Neural Networks Lecture 1: Introduction, Logistic Regression, Backpropagation, MLP

Podcast

Play an AI-generated podcast conversation about this lesson

Questions and Answers

What is the loss function used in linear regression?

Mean Squared Error (MSE) / Quadratic Loss

What is the function that 'squeezes in' the weighted input into a probability space in logistic regression?

Logistic Sigmoid Function

What is the measure of the uncertainty associated with a random variable in logistic regression?

Entropy (H)

What is the model that specifies the probability of binary output given an input in logistic regression?

Logistic regression model Signup and view all the answers

What method of estimating the parameters of a statistical model maximizes the likelihood of making the observations given the parameters?

Maximum Likelihood Estimation (MLE) Signup and view all the answers

What property does Maximum Likelihood Estimation (MLE) have for i.i.d. data?

MLE minimizes Cross-Entropy Signup and view all the answers

What distribution is used to denote the probability of a binary output in logistic regression?

Bernoulli distribution Signup and view all the answers

What is the dataset used to estimate the parameters of a statistical model in logistic regression?

Empirical Data Distribution Signup and view all the answers

What is the function used to minimize the negative log likelihood in logistic regression?

Cross-Entropy Signup and view all the answers

What is the purpose of the Logistic Sigmoid Function in logistic regression?

To 'squeeze in' the weighted input into a probability space Signup and view all the answers

What is the model that specifies the probability of binary output given an input in logistic regression?

Logistic regression model Signup and view all the answers

What is the measure of the uncertainty associated with a random variable in logistic regression?

Entropy (H) Signup and view all the answers

What is the measure of difference between two probability distributions in logistic regression?

Kullback-Leibler Divergence Signup and view all the answers

What method is used to solve the loss function in logistic regression when it no longer has a closed-form solution?

Gradient Descent Signup and view all the answers

What does the gradient vector point in the direction of in gradient descent?

Steepest ascent Signup and view all the answers

In logistic regression, what is the generalization of a neural network from binary classification to multiclass?

Softmax Signup and view all the answers

What is the derivative of the logit function in logistic regression?

σ'(x) = σ(x)(1−σ(x)) Signup and view all the answers

What does the loss function in logistic regression equal to?

J(θ) = −log P(y∣X, θ) = −∑ yi log πi +(1−yi) log(1−πi) Signup and view all the answers

What is the objective of a Multi-Layer Perceptron (MLP)?

Minimize cross-entropy error Signup and view all the answers

What does each neuron in a Multi-Layer Perceptron (MLP) compute?

A separation plane on the space of its inputs Signup and view all the answers

What is the influence of the Activation Functions in the Neural Net Playground?

Observe influence of: Activation Functions Signup and view all the answers

In logistic regression, what is the method used to solve the loss function with a closed-form solution?

Gradient Descent Signup and view all the answers

What is the derivative of the loss function in logistic regression with respect to θ?

∂J(θ)/∂θ = ∑xi(πi−yi) Signup and view all the answers

What is the measure of difference between two probability distributions in logistic regression that needs to be minimized?

Cross-Entropy Signup and view all the answers

What is the size of the output volume after applying a convolution layer with a kernel (filter) of size 5 × 5 × 3 to an input volume of dimension 32 × 32 × 3?

28 × 28 × 1 (D) Signup and view all the answers

What is the purpose of the hyperparameter 'stride' in convolutional neural networks?

To control the step taken when sliding the filter over the input volume (D) Signup and view all the answers

How does the 'zero-padding' hyperparameter affect the output volume in convolutional neural networks?

It preserves the spatial size of the input volume (A) Signup and view all the answers

What is the constraint on strides in convolutional neural networks?

The result of the division has to be an integer (D) Signup and view all the answers

What is the purpose of parameter sharing in convolutional neural networks?

To reduce the computational complexity of convolutional layers (D) Signup and view all the answers

What is the purpose of the torch.nn.Conv1d function in PyTorch?

To apply a one-dimensional convolution over an input signal (C) Signup and view all the answers

What is the main disadvantage of using a fully connected layer in convolutional neural networks?

It loses information about spatial arrangement of pixels (D) Signup and view all the answers

What is the purpose of a convolution layer in CNNs?

To connect each neuron to only a local region of the input volume (B) Signup and view all the answers

In the given case study, what is the dimension of the output map after applying a 5x5x3 filter to a 32x32x3 input volume?

28x28x1 (B) Signup and view all the answers

What does the term 'receptive field' refer to in the context of convolutional layers in CNNs?

The spatial extent of the local connectivity of each neuron (A) Signup and view all the answers

Which of the following is true about the output map dimension in a convolutional layer when applying a filter to an input volume?

The width and height of the output map are determined by the size of the filter and padding (B) Signup and view all the answers

What is the primary reason for using pooling layers in CNNs?

To reduce overfitting and improve computational efficiency (B) Signup and view all the answers

What is the function of a convolution kernel in the spatial domain?

It applies a filter to the image to extract features. (C) Signup and view all the answers

What does the convolution operation in the spatial domain imply?

It uses a convolution kernel to process the image pixels. (C) Signup and view all the answers

How is an RGB image represented as a function in the spatial domain?

As a function mapping coordinates to color values. (A) Signup and view all the answers

What does the convolution product between two functions represent in the continuous case?

The weighted combination of function values. (A) Signup and view all the answers

What is the purpose of applying operators to an image in the spatial domain?

To enhance specific image features. (C) Signup and view all the answers

In the discrete case, how is the convolution operation between two functions represented?

As a summation of transformed function values. (A) Signup and view all the answers

What is the measure of difference between two probability distributions in logistic regression that needs to be minimized?

Kullback-Leibler Divergence (B) Signup and view all the answers

What method is used to solve the loss function in logistic regression when it no longer has a closed-form solution?

Stochastic Gradient Descent (SGD) (D) Signup and view all the answers

What does the gradient vector point in the direction of in gradient descent?

Steepest Ascent (A) Signup and view all the answers

What is the derivative of the logit function in logistic regression?

$σ(x)(1−σ(x))$ (A) Signup and view all the answers

What is the model that specifies the probability of binary output given an input in logistic regression?

Logistic Regression Model (C) Signup and view all the answers

What is the measure of uncertainty associated with a random variable in logistic regression?

Entropy (C) Signup and view all the answers

What function is used to minimize the negative log likelihood in logistic regression?

$−log(πi)$ (D) Signup and view all the answers

What distribution is used to denote the probability of a binary output in logistic regression?

Bernoulli Distribution (D) Signup and view all the answers

What property does Maximum Likelihood Estimation (MLE) have for i.i.d. data?

$E[MLE]=θ$ (A) Signup and view all the answers

What does each neuron in a Multi-Layer Perceptron (MLP) compute?

Separation Plane on the space of its inputs (A) Signup and view all the answers

What does the loss function in logistic regression equal to?

Cross-Entropy Loss (D) Signup and view all the answers

What is the main purpose of using dilated convolutions in semantic segmentation?

To enable larger effective receptive field sizes and capture spatial context information (D) Signup and view all the answers

What is the limitation of the output stride (reduction factor for image resolution) in semantic segmentation using dilated convolutions?

Limited to 16, as larger values are harmful for semantic segmentation (C) Signup and view all the answers

What is the main idea behind Atrous Spatial Pyramid Pooling in semantic segmentation?

Resampling features at different scales for more accurate region classification (B) Signup and view all the answers

What are the two main streams of methods in Instance Segmentation?

Proposal based and Segmentation based (B) Signup and view all the answers

What is the architecture that builds upon Faster-RCNN in the case of MaskRCNN for Instance Segmentation?

Region of Interest (RoI) proposal based architecture (C) Signup and view all the answers

What is the primary function of the RoI proposal based approach in Instance Segmentation?

Propose possible regions of interest where an object might be found and extract features from that region (D) Signup and view all the answers

What is the main difference between proposal based and segmentation based methods in Instance Segmentation?

Proposal based methods propose possible regions of interest where an object might be found, while segmentation based methods start from segmentation as the first objective. (C) Signup and view all the answers

MaskRCNN is an extension of which object detection architecture?

Faster-RCNN (B) Signup and view all the answers

Semantic Segmentation aims to classify each pixel in an image into a specific class. What technique does DeepLab-v3 use to capture spatial context information effectively?

Dilated convolutions and Atrous Spatial Pyramid Pooling (B) Signup and view all the answers

How does Atrous Spatial Pyramid Pooling in DeepLab-v3 implement the idea of resampling features at different scales?

Using a combination of atrous convolutions and global average pooling (A) Signup and view all the answers

What is the formula for calculating the height of the output volume (Hout) in a convolutional layer?

Hout = Hin + 2 * padding - dilation * (kernel size - 1) - 1 + 1 (C) Signup and view all the answers

How does the 'zero-padding' hyperparameter affect the output volume in convolutional neural networks?

It increases the spatial size of the output volume. (A) Signup and view all the answers

What is the purpose of the hyperparameter 'stride' in convolutional neural networks?

To control the step taken when sliding the filter. (A) Signup and view all the answers

How does Atrous Spatial Pyramid Pooling in DeepLab-v3 implement the idea of resampling features at different scales?

By dilating the filters to capture spatial context at different scales. (D) Signup and view all the answers

What is the constraint on strides in convolutional neural networks?

The result of the division has to be an integer. (B) Signup and view all the answers

What is the purpose of parameter sharing in convolutional neural networks?

To control the number of learnable parameters. (C) Signup and view all the answers

What is the primary focus of YOLOv3 compared to its predecessors?

Improvement focus on accuracy and speed (A) Signup and view all the answers

Which factor contributes to the struggles of original YOLO in detecting objects of small sizes that appear in groups?

Limited number of grid cells predicting only two boxes (B) Signup and view all the answers

What is the purpose of using anchor boxes in YOLOv2?

To scale the predicted box w.r.t. defined anchor boxes (A) Signup and view all the answers

What is the key difference in the activation function used in YOLO v1 as compared to YOLO v2?

Use of Leaky ReLU function in YOLO v2 instead of ReLU (C) Signup and view all the answers

What does the YOLO algorithm use to optimize directly for detection of objects?

Bounding box regression (B) Signup and view all the answers

Which feature is emphasized in YOLOv2 to tackle the vanishing gradient problem?

Batch Normalization on all layers (B) Signup and view all the answers

What is the metric used to force predicted output boxes to coincide with ground truth in YOLO v1?

Intersection Over Union (IOU) (D) Signup and view all the answers

How does YOLOv1 process frames compared to its competitors at the time?

Processes frames in real time and uses a single pipeline for detection and localization (C) Signup and view all the answers

What is the limitation related to small bounding boxes versus large bounding boxes in the original YOLO architecture?

'Struggles with objects of small sizes that appear in groups' (A) Signup and view all the answers

What inspired the architecture of YOLO v1?

Inspired by GoogLeNet (B) Signup and view all the answers

What is the philosophy behind the inception module in GoogleNet?

Designing good local network topology and stacking the modules on top of another (A) Signup and view all the answers

What is the main purpose of using a global average pooling layer in GoogleNet?

To spatially average across each feature map before the final FC layer (D) Signup and view all the answers

What is the primary function of a skip connection in Residual Network (ResNet)?

To fit residual F(x) = H(x) − x, rather than H(x) directly (A) Signup and view all the answers

What is the key aspect focused on in ResNeXt for network performance?

Implementing multi-branch aggregated transformations (B) Signup and view all the answers

What is the main takeaway regarding feature reuse in Wide Residual Networks?

Diminishing feature reuse can be addressed by using residual blocks (D) Signup and view all the answers

What is the purpose of using grouped convolutions in ResNeXt?

Implementing multi-branch aggregated transformations for improved network performance (C) Signup and view all the answers

What is the purpose of the RoI Pooling in the Faster R-CNN architecture?

To extract features from the convolutional feature map for region proposals (A) Signup and view all the answers

What fundamental concepts are associated with Faster R-CNN?

Region Proposal Network, Anchor Boxes, RoI Pooling (D) Signup and view all the answers

What changes were made in Mask-RCNN in comparison to Faster R-CNN?

All of the above (D) Signup and view all the answers

What is the function of the Anchor Boxes in Faster R-CNN?

To propose regions where objects might be found (A) Signup and view all the answers

What are the downsampling ratios of CNN feature maps used in Anchor Boxes for object detection?

3 different scales, 3 different aspect ratios (A) Signup and view all the answers

What is the main drawback that deformable convolutions aim to address?

Limited sampling points only from the discrete grid (A) Signup and view all the answers

What is the key improvement of RoI Align Layer over RoI Pooling?

Avoids quantization of RoI coordinates or spatial bins to feature map grid (B) Signup and view all the answers

What is the trade-off made by setting a constant spatial-offset (k, x, y) for each channel C in deformable convolutions?

Between modelling capacity and compute (A) Signup and view all the answers

What is the role of the backbone network (VGG-16) in Faster RCNN?

To feed the Region Proposal Network and the Class and BBox regressor network (A) Signup and view all the answers

What changes were made in Mask-RCNN in comparison to Faster R-CNN?

All of the above (D) Signup and view all the answers

Which task is an example of using Recurrent Neural Networks (RNNs) for sequential processing of non-sequence data?

Classifying images by taking a series of 'glimpses' (C) Signup and view all the answers

What is the purpose of the Elman RNN model?

Modeling sequential data (A) Signup and view all the answers

Which type of task involves translating a sequence of words into another sequence of words using RNNs?

Translation (D) Signup and view all the answers

What concept addresses the issue of vanishing and exploding gradients in RNN training?

LSTM Intuitions (C) Signup and view all the answers

In the context of RNNs, what is the primary focus of LSTM?

Addressing vanishing and exploding gradients (D) Signup and view all the answers

Which task involves classifying images by taking a series of 'glimpses'?

Sequential Processing of Non-Sequence Data (B) Signup and view all the answers

What is the primary reason for using LSTM in RNNs?

Addressing vanishing and exploding gradients (A) Signup and view all the answers

What is a key feature of using RNNs for image captioning?

Generating images one piece at a time (A) Signup and view all the answers

'Drawing a Recurrent Neural Network For Image Generation' is associated with which task?

'Generating images one piece at a time' (B) Signup and view all the answers

How does 'Video classification on frame level' relate to the application of Recurrent Neural Networks (RNNs)?

It involves processing non-sequence data sequentially (D) Signup and view all the answers

What is the purpose of truncated backpropagation through time (TBPTT)?

To carry hidden states forward in time forever and backpropagate for some smaller number of steps (A) Signup and view all the answers

What is the main difference between Long Short Term Memory (LSTM) and vanilla RNN in terms of preserving information over many timesteps?

LSTM provides an easier way for the model to learn long-distance dependencies, while vanilla RNN does not (A) Signup and view all the answers

What does the LSTM architecture make easier for the RNN in terms of gradient flow?

Balancing gradient values through forget, input, and output gates (C) Signup and view all the answers

What is the primary focus of Long Short Term Memory (LSTM) compared to vanilla RNN?

Preserving information over many timesteps (D) Signup and view all the answers

What is the main advantage of using Long Short Term Memory (LSTM) over vanilla RNN?

Learning long-distance dependencies (B) Signup and view all the answers

What is the role of the input gate (i) in the LSTM cell?

Scales input to cell (write) (B) Signup and view all the answers

What does TBPTT(k1, k2), where k1 < 1, lead to?

Exploding gradients (C) Signup and view all the answers

What is the significance of the forget gate (f) in the LSTM cell?

'Scales old cell values (reset memory)' (B) Signup and view all the answers

What does Truncated BPTT (TBPTT) with n=1 imply?

The network likely does not have enough temporal context to learn, relying heavily on internal state and inputs (A) Signup and view all the answers

Which paper discusses 'Batch-instance normalization for adaptively style-invariant neural networks'?

Nam & Kim, 2018 (B) Signup and view all the answers

In which conference was 'Group normalization' presented?

European conference on computer vision (ECCV) (A) Signup and view all the answers

Which paper introduces 'Semantic image synthesis with spatially-adaptive normalization'?

Park et al., 2019 (B) Signup and view all the answers

Who presented the concept of 'Micro-batch training with batch-channel normalization and weight standardization'?

Qiao et al., 2019 (C) Signup and view all the answers

Which paper discusses 'Batch-instance normalization for adaptively style-invariant neural networks'?

[Nam & Kim, 2019] (A) Signup and view all the answers

What is the primary focus of Dense captioning Events in Videos?

Describing events in a video with dense natural language descriptions (B) Signup and view all the answers

What does the term 'Vanilla RNN Model' refer to?

A recurrent neural network model with a single hidden layer (B) Signup and view all the answers

What task is an example of using Recurrent Neural Networks (RNNs) for sequential processing of non-sequence data?

Image classification (C) Signup and view all the answers

What is the key aspect focused on in Sequence to Sequence Learning with Neural Networks?

Producing output sequence from a single input vector (A) Signup and view all the answers

What is the purpose of applying operators to an image in the spatial domain?

Representing an image as a function (C) Signup and view all the answers

What does the term 'receptive field' refer to in the context of convolutional layers in CNNs?

The region of the input space that a particular feature is looking at (A) Signup and view all the answers

What is the function of the RoI proposal based approach in Instance Segmentation?

Proposing regions likely to contain objects of interest (B) Signup and view all the answers

Semantic Segmentation aims to classify each pixel in an image into a specific class. What technique does DeepLab-v3 use to capture spatial context information effectively?

Atrous (with holes) convolution (C) Signup and view all the answers

What does the loss function in logistic regression equal to?

Cross-entropy loss (C) Signup and view all the answers

What is a disadvantage of using BatchNorm in tasks such as video prediction, segmentation, and medical image processing?

Small batch size leads to inaccurate estimates (D) Signup and view all the answers

In what scenario is Layer Normalization suitable?

When the statistics need to be independent of the batch (B) Signup and view all the answers

What is the primary advantage of Instance Normalization?

It computes statistics across channels and spatial dimensions (C) Signup and view all the answers

What problem can arise in classification tasks when using Batch Instance Normalization?

Style information as the discriminating factor (D) Signup and view all the answers

When can Group Normalization be used?

For object detection tasks with large batch sizes (B) Signup and view all the answers

In what scenario is Adaptive Instance Normalization used for channel-wise alignment?

For style transfer in image classification tasks (B) Signup and view all the answers

What does Batch Instance Normalization learn to control?

How much style information to propagate through each channel (D) Signup and view all the answers

What is the primary advantage of Group Normalization over Layer Normalization?

Stable across a greater range of batch sizes (C) Signup and view all the answers

In what scenarios is Layer Normalization primarily used?

'Frame wise' normalization at each time step across batch (C) Signup and view all the answers

What is the main difference between Layer Normalization and Instance Normalization?

'Frame wise' normalization vs. 'channel wise' normalization (D) Signup and view all the answers

What problem does the Reformer architecture address?

All of the above (D) Signup and view all the answers

What is the key idea behind Linformer for reducing memory complexity?

All of the above (D) Signup and view all the answers

How is attention interpreted in the context of kernel interpretation?

As the Gram Matrix of an exponential kernel (C) Signup and view all the answers

What is the primary function of the FAVOR+ mechanism in Performer?

Utilizing positive orthogonal random features for fast attention computation (A) Signup and view all the answers

What does the FAVOR+ mechanism approximate using positive orthogonal random features?

$\phi(Q) \times \phi(K)^T$ (D) Signup and view all the answers

When is adding recurrence useful for long sequences?

When dealing with very long sequences or when previous approaches do not fit in available hardware (A) Signup and view all the answers

What problem does Transformer-XL address?

Storing forward-pass activations for each transformer layer (B) Signup and view all the answers

What does Truncated BPTT (TBPTT) with $n=1$ imply?

$n$-step unrolling of the forward pass during backpropagation through time (B) Signup and view all the answers

What is the primary purpose of the Logistic Sigmoid Function in logistic regression?

Mapping input values to probabilities between 0 and 1 (C) Signup and view all the answers

What does the term 'receptive field' refer to in the context of convolutional layers in CNNs?

The region in the input volume that affects a unit in the convolutional layer (B) Signup and view all the answers

In the context of efficient attention, which technique involves dividing the sequence into local blocks and restricting attention within them?

Blockwise Patterns (D) Signup and view all the answers

What attention pattern reduces time complexity to be linear in sequence length and window size?

Diagonal (sliding window) Patterns (C) Signup and view all the answers

Which example of efficient attention pattern showcases the use of sliding, strided, and global attention patterns?

Longformer (A) Signup and view all the answers

In the context of efficient attention, which pattern is applied to a few special tokens that are often prepended to the sequence and is usually combined with other attention patterns?

Global Attention Patterns (B) Signup and view all the answers

Which technique showcases the use of dilation configurations, multi-headed attention, and position embeddings?

Longformer (B) Signup and view all the answers

What efficient attention pattern involves reaching a receptive field that can be 10^4 tokens wide for small values of d?

Dilated Sliding Window (D) Signup and view all the answers

Which technique showcases the use of global, sliding, and random patterns of token blocks?

BigBird (A) Signup and view all the answers

Which efficient attention pattern showcases the use of sliding window and global attention patterns in addressing the problem of handling large documents?

Longformer (C) Signup and view all the answers

Which type of GNN layer is useful for homophilous graphs and is highly scalable?

Attentional GNN (B) Signup and view all the answers

In which GNN layer are the features of neighbors aggregated with implicit weights (attention)?

Attentional GNN (B) Signup and view all the answers

Which GNN layer computes arbitrary vectors (messages) to be sent across edges?

Message-passing GNN (C) Signup and view all the answers

Which function defines a neighborhood aggregation function according to the given model design overview?

Loss function on the embeddings (A) Signup and view all the answers

What is the primary model mentioned for building and training GNNs in the given text?

GCN (C) Signup and view all the answers

Which type of GNN layer is ideal for computational chemistry, reasoning, and simulation tasks?

Message-passing GNN (C) Signup and view all the answers

In which GNN layer do edges give a 'recipe' for passing data and may have scalability or learnability issues?

Message-passing GNN (C) Signup and view all the answers

What is the common feature of GraphNets, Interaction Nets, and MPNN?

"Middle ground" w.r.t. capacity, scale, interpretability (C) Signup and view all the answers

What is the correct definition of permutation invariance for 𝑓(𝐗)?

𝑓(𝐗) is permutation invariant if, for all permutation matrices 𝐏: 𝑓[𝐏𝐗] = 𝑓[𝐗] (D) Signup and view all the answers

Which type of model is suitable for set-level outputs?

Deep Sets model (A) Signup and view all the answers

What is the purpose of extracting neighbourhood features in graph neural networks?

To enforce locality in equivariant set functions (A) Signup and view all the answers

For graph neural networks, which operation ensures permutation equivariance?

Applying a local function to every node in isolation (D) Signup and view all the answers

What is the main difference between permutation invariance and equivariance on graphs?

Permutations now also accordingly act on the edges in equivariance (B) Signup and view all the answers

What does it mean to ensure equivariance for graph neural networks?

Ensuring the local function does not depend on the order of nodes (B) Signup and view all the answers

What is a common lingo used for the shared application of a local permutation-invariant function in graph neural networks?

"GNN layer" (C) Signup and view all the answers

What is the primary focus of Graph Neural Networks (GNNs)?

Obtaining embeddings of information contained in graphs for use in deep learning pipelines (B) Signup and view all the answers

What are some examples of structured data that are ever present and can be represented as graphs?

User preference/consumption and social network graphs (C) Signup and view all the answers

What is the recent and hot topic in machine learning research as mentioned in the text?

Graph Neural Networks (GNNs) (B) Signup and view all the answers

What is the challenge addressed by Graph Neural Networks (GNNs) as stated in the text?

Applying deep learning techniques to graph-based information representations (C) Signup and view all the answers

In what real-world applications have Graph Neural Networks (GNNs) made an impact, as mentioned in the text?

Drug discovery and fake news detection on social media (C) Signup and view all the answers

What is the primary function of Graph Convolutional Networks as part of GNN models?

Applying convolution operations to graph-structured data (A) Signup and view all the answers

What is the main focus of Graph Attentional Networks, a foundational GNN model?

Learning attention weights over graph nodes for message passing (A) Signup and view all the answers

What is the general framework for building and training GNNs, as mentioned in the text?

'Message passing' and aggregation over neighborhood nodes (A) Signup and view all the answers

In what scenarios have GNNs broken into the real world, as mentioned in the text?

'Drug discovery' including for COVID-19 treatment (D) Signup and view all the answers

Structured data is ever present. How can we apply deep learning techniques to graph-based information representations?

By leveraging MPNNs to perform message passing over the graph structure. (D) Signup and view all the answers

What is the main challenge in deep learning for graph data when it comes to mapping nodes to d-dimensional embeddings?

Learning the mapping function f (B) Signup and view all the answers

What is the desirable property for a graph convolutional layer in terms of parameters?

Fixed number of parameters (independent of input size) (D) Signup and view all the answers

What is the goal of the encoder in the context of deep learning methods based on graph neural networks (GNNs)?

Mapping each node to a low-dimensional vector (D) Signup and view all the answers

What are the tasks that can be solved with GNNs according to the text?

Node classification, link prediction, and community detection (B) Signup and view all the answers

What is the primary challenge associated with networks in comparison to simple sequences and grids?

Arbitrary size and complex topological structure (B) Signup and view all the answers

What is the purpose of symmetry group 𝔊 and its group element 𝔤 in the context of learning on sets?

Permutations and permutation matrices for changing node order (A) Signup and view all the answers

What does permutation invariance aim to achieve in functions 𝑓(𝐗) over sets?

Result should not depend on the order (A) Signup and view all the answers

What does learning on sets initially assume about the graph being analyzed?

No edges in the graph (i.e. Ω = 𝒱, set of nodes) (D) Signup and view all the answers

What does the symmetry group 𝔊 consist of in the context of learning on sets?

|𝒱|-element Permutation group Σ𝒱 (D) Signup and view all the answers

What is the useful notion that arises from permutation invariance according to the text?

Permutation invariance for functions over sets (A) Signup and view all the answers

What is the main purpose of Transformer-XL's relative position encoding scheme?

To facilitate the recurrence strategy (D) Signup and view all the answers

In the context of efficient attention, what does Transformer XL's query content to key content Uj replaced with its relative position counterpart signify?

Attending to some positions more than others (D) Signup and view all the answers

What is the distinctive feature of Longformer, as compared to other efficient transformers?

Facilitation of long document understanding (A) Signup and view all the answers

In the arena of efficient transformers, what does Long-Range Arena Challenge benchmark primarily aim to assess?

Performance in handling long-range dependencies (D) Signup and view all the answers

According to the provided text, what is the main focus of the Big Bird transformer?

Efficiently processing longer sequences (B) Signup and view all the answers

What does the 'Reformer' model primarily aim to achieve?

Efficiently processing and understanding long documents (A) Signup and view all the answers

What is the key aspect focused on in Linformer for network performance enhancement?

Efficient self-attention with linear complexity (D) Signup and view all the answers

What is the primary focus of Rethinking Attention with Performers in terms of attention mechanisms?

'Self-attention with linear complexity' incorporation (A) Signup and view all the answers

According to the provided text, what is the main focus of Efficient transformers: A survey by Tay et al?

'Long-document understanding' benchmarking (C) Signup and view all the answers

What is the role of Efficient transformers: A survey by Tay et al in the context of transformer models?

'Efficient transformers' benchmarking and assessment (D) Signup and view all the answers

What is the formula for the loss function in linear regression?

$J(\theta) = \sum_{i=1}^{n} (y_i - \theta_1 - x_i\theta_2)$ (C) Signup and view all the answers

What does the Logistic Sigmoid Function do?

Squeezes the input into a probability space (D) Signup and view all the answers

What is the purpose of Maximum Likelihood Estimation (MLE) in logistic regression?

Maximizing likelihood of making the observations given the parameters (D) Signup and view all the answers

What is the distribution used to denote the probability of a binary output in logistic regression?

Bernoulli distribution (B) Signup and view all the answers

What property does Maximum Likelihood Estimation (MLE) have for i.i.d. data?

Maximizes likelihood of making the observations given the parameters (C) Signup and view all the answers

What does Cross-Entropy measure in logistic regression?

Uncertainty associated with a random variable (C) Signup and view all the answers

What does the Logistic regression model specify for binary output given an input?

Probabilities of positive and negative classes (C) Signup and view all the answers

What is the purpose of the Hessian matrix in optimization?

To find the minimum or maximum of a function (C) Signup and view all the answers

What does the gradient vector represent in the context of optimization?

The rate of change of the function with respect to the independent variables (B) Signup and view all the answers

In optimization, what role does the gradient descent algorithm play?

To find the minimum of a function by iteratively moving in the direction opposite to the gradient (C) Signup and view all the answers

What is the primary purpose of Stochastic Gradient Descent (SGD) in optimization?

To handle large datasets efficiently (C) Signup and view all the answers

In the context of optimization, what does the Hessian matrix's diagonal represent?

The curvature of the function along each variable axis (B) Signup and view all the answers

What is the significance of second-order derivatives in optimization?

They determine whether a critical point is a minimum, maximum, or saddle point (A) Signup and view all the answers

What is the key concept behind second-order optimization methods?

Utilizing both first and second derivatives for optimization (D) Signup and view all the answers

What does the term 'stochastic' refer to in Stochastic Gradient Descent (SGD)?

It refers to using random subsets of data for each iteration (C) Signup and view all the answers

What distinguishes second-order optimization methods from gradient descent?

'Utilizing both first and second derivatives for optimization' (D) Signup and view all the answers

What does the Hessian matrix help determine in optimization?

The curvature and nature of critical points (B) Signup and view all the answers

What distinguishes Stochastic Gradient Descent (SGD) from traditional gradient descent?

Random selection of subsets from large datasets for each iteration (B) Signup and view all the answers

What does the gradient vector help determine in optimization?

The rate of change with respect to independent variables (A) Signup and view all the answers

What is the primary function of a Convolution Layer in a CNN?

To apply a filter to the input volume and produce an output map (A) Signup and view all the answers

What is the disadvantage of using a Fully Connected Layer in a CNN?

Loss of information about spatial arrangement of pixels (B) Signup and view all the answers

In the context of CNNs, what does the term 'receptive field' refer to?

The spatial extent of local connectivity of each neuron (C) Signup and view all the answers

What is the primary purpose of applying a filter in a Convolution Layer?

To capture spatial features from the input volume (B) Signup and view all the answers

For an input volume of 32 × 32 × 3 and applying a filter of size 5 × 5 × 3, what is the dimension of the output map?

28 × 28 × 1 (C) Signup and view all the answers

What is the primary objective of connecting each neuron to only a local region of the input volume in a Convolution Layer?

To capture detailed spatial information from the input volume (B) Signup and view all the answers

What is the primary advantage of using depthwise separable convolution?

Reduced computational load and fewer kernel parameters (A) Signup and view all the answers

What is the main purpose of using a pooling layer in a convolutional neural network?

To modify the input volume into a smaller and more manageable representation (A) Signup and view all the answers

What is the purpose of batch normalization in convolutional neural networks?

To address covariate-shift problem and enable higher learning rates (C) Signup and view all the answers

What is a distinctive feature of VGG-16 architecture compared to other classic networks?

Uses only 3x3 convolutions with stride 1 and 2x2 max pool with stride 2 (D) Signup and view all the answers

What does the transpose convolution operation aim to achieve?

Upsampling and capturing spatial information (C) Signup and view all the answers

What problem does ReLU activation function primarily address in CNNs?

Vanishing gradient problem for deeper networks (A) Signup and view all the answers

What is the primary reason for using Mosaic Data Augmentation in YOLOv4?

To improve performance on small object detection (B) Signup and view all the answers

Why does YOLOv4 choose CSPDarknet53 as the backbone network?

To compromise in receptive field size and FPS (C) Signup and view all the answers

What is the main limitation of Temporal Convolutional Network (TCN) for sequence modeling?

Limitations in test/evaluation mode (C) Signup and view all the answers

In what way does InceptionTime reduce variance in classification performance?

By using ensembling with multiple network architectures (A) Signup and view all the answers

What is the purpose of Adaptive Feature Pooling in YOLOv4?

To aggregate features from all levels of the bottom up path (C) Signup and view all the answers

How does Path Aggregation Net contribute to YOLOv4?

By aggregating features from several stages of the feature extractor network (B) Signup and view all the answers

What is a key task that can be solved using Recurrent Neural Networks (RNNs) according to the provided text?

Image Captioning (C) Signup and view all the answers

In what scenario is the generation of images one piece at a time discussed in the provided text?

Sequential Processing of Non-Sequence Data (D) Signup and view all the answers

What type of data processing is discussed in the context of classifying images by taking a series of 'glimpses'?

Sequential Processing of Non-Sequence Data (A) Signup and view all the answers

What is the primary focus of Long Short Term Memory (LSTM) compared to vanilla RNN according to the provided text?

Mitigating vanishing and exploding gradients (D) Signup and view all the answers

In logistic regression, what does the loss function represent?

The measure of uncertainty associated with a random variable. (B) Signup and view all the answers

What is one task that can be solved with Recurrent Neural Networks (RNNs) according to the provided text?

Language Translation (A) Signup and view all the answers

What is the primary application of sequence-to-sequence models?

Machine translation (B) Signup and view all the answers

What is the purpose of the encoder in a sequence-to-sequence model?

To generate the context vector (C) Signup and view all the answers

What is the significance of using teacher forcing in sequence-to-sequence models?

It allows the model to learn from its own predictions (B) Signup and view all the answers

In a sequence-to-sequence model, when is the loop broken during decoding?

When the decoder predicts the (END) token (A) Signup and view all the answers

What is a key advantage of sequence-to-sequence models?

They can work with variable-length input and output sequence pairs (D) Signup and view all the answers

What type of models are seq2seq models commonly referred to as?

Encoder-decoder models (C) Signup and view all the answers

What does the decoder receive during the forward pass in a seq2seq model?

Input and context vector (D) Signup and view all the answers

What does the context vector represent in a seq2seq model?

The context of the whole input sequence (B) Signup and view all the answers

What is the primary function of the decoder in a seq2seq model?

To predict the next input token (B) Signup and view all the answers

Which task can be performed using seq2seq models?

Text summarization (A) Signup and view all the answers

What is achieved by using RNNs again in the decoder of a seq2seq model?

Decoding of the 'context' vector (D) Signup and view all the answers

What is an advantage of using seq2seq models in auto-encoding setup?

It enables self-learning capabilities in the model (B) Signup and view all the answers

What is the primary purpose of the relative position encoding scheme in Transformer-XL?

To facilitate the recurrence strategy (A) Signup and view all the answers

What is the key aspect focused on in Reformer for network performance enhancement?

Linear complexity self-attention (C) Signup and view all the answers

In Efficient Attention, what is the purpose of adding a component that feeds the hidden states of previous segments as inputs to current segment layers in Transformer-XL?

Facilitating the recurrence strategy (B) Signup and view all the answers

What is the main idea behind Linformer for network performance enhancement?

Self-attention with linear complexity (A) Signup and view all the answers

Which paper introduces 'Semantic image synthesis with spatially-adaptive normalization'?

Choromanski et al., 2020, Rethinking Attention with Performers (B) Signup and view all the answers

What is the function of the Anchor Boxes in Faster R-CNN?

To encode position-sensitive information about objects (C) Signup and view all the answers

What is the primary focus of Rethinking Attention with Performers in terms of attention mechanisms?

Efficient attention patterns (B) Signup and view all the answers

What is the recent and hot topic in machine learning research as mentioned in the text?

'A Survey of Methods for Incorporating Long-Term Context' (C) Signup and view all the answers

What is the primary reason for using LSTM in RNNs?

To facilitate the recurrence strategy (A) Signup and view all the answers

What is a key feature of using RNNs for image captioning?

Capturing sequential dependencies (C) Signup and view all the answers

What is the primary focus of the Long-Range Arena Challenge benchmark?

Evaluating the efficiency of attention mechanisms in transformers (B) Signup and view all the answers

In Transformer-XL, what is the purpose of the relative position encoding scheme?

To facilitate the recurrence strategy for long sequences (A) Signup and view all the answers

What distinguishes Longformer from other efficient transformers?

It is designed for processing longer sequences (C) Signup and view all the answers

Which paper introduces the concept of Big Bird: Transformers for longer sequences?

Zaheer et al., 2020 (B) Signup and view all the answers

What do performers in Rethinking Attention with Performers focus on?

Efficient computation of attention without full attention matrices (B) Signup and view all the answers

According to the given text, what does Efficient transformers: A survey primarily focus on?

Providing an overview and comparison of efficient transformers (D) Signup and view all the answers

Which paper introduces Linformer: Self-attention with linear complexity?

Wang et al., 2020 (C) Signup and view all the answers

Which method for missing value imputation can be computationally intensive when the dataset is very large?

k-NN imputation (D) Signup and view all the answers

What is the main purpose of Seasonal and Trend Decomposition using Loess (STL)?

To estimate models for seasonal, trend, and residual components independently (C) Signup and view all the answers

What transformation function can be applied to obtain variance stabilization in data?

Box-Cox transform (D) Signup and view all the answers

When is Mean Normalization useful or required in time series data?

When the numeric values in the time series are large (D) Signup and view all the answers

What does the AR component of ARIMA attempt to predict?

Future values based on past values (A) Signup and view all the answers

What does the I (Integrated) model component in ARIMA expect of the time series?

To be stationary (A) Signup and view all the answers

What is the primary purpose of using (partial-) Auto Correlation Function plots in ARIMA?

To determine the appropriate hyperparameters AR(p), MA(q), and I(d) (B) Signup and view all the answers

According to the provided text, what does RNN stand for in the context of time series forecasting?

Recurrent Neural Networks (D) Signup and view all the answers

What benchmarking paper is referenced for Recurrent Neural Networks (RNNs) in time series forecasting?

Hewamalage et al. (2021). Recurrent neural networks for time series forecasting: Current status and future directions. (A) Signup and view all the answers

What post-processing step is required for final error metric computation when using RNN models for time series forecasting?

Reverse trend normalization by subtracting the trend value of the last input point. (A) Signup and view all the answers

Which type of graphs are considered a generalization of images according to the text?

Grid graphs (D) Signup and view all the answers

What is a desirable property for a graph convolutional layer according to the text?

Specifying different importances to different neighbors (C) Signup and view all the answers

What property does a function 𝑓(𝐗) have if, for all permutation matrices 𝐏, 𝑓 𝐏𝐗 = 𝑓 𝐗?

Permutation invariance (B) Signup and view all the answers

What is the goal of the similarity function mentioned in the text?

To specify relationships in the original network (A) Signup and view all the answers

In the context of deep sets, what is the critical operation for the sum aggregation?

Sum aggregation (D) Signup and view all the answers

In the context of graph neural networks, what does the term 'neighbourhood' refer to?

A node's 1-hop connectivity (C) Signup and view all the answers

What is an example task that can be solved with Graph Neural Networks (GNNs) according to the text?

Node classification (A) Signup and view all the answers

What are networks far more complex than, according to the text?

Dynamic networks (B) Signup and view all the answers

What is the main difference between permutation invariance and permutation equivariance on graphs?

Permutations acting on edges (A) Signup and view all the answers

What operation is necessary to construct permutation equivariant functions on graphs?

Local permutation-invariant function (D) Signup and view all the answers

What is the focus of learning on sets, as mentioned in the text?

Learning on unordered sets (A) Signup and view all the answers

What is the primary focus of Graph Neural Networks (GNNs)?

Learning on graph-structured data (C) Signup and view all the answers

What does the symmetry group 𝔊 aim to achieve in the context of learning on sets?

To change node order (D) Signup and view all the answers

What does permutation invariance aim to achieve according to the text?

Consistency irrespective of feature ordering (C) Signup and view all the answers

What is a useful notion achieved by permutation invariance as stated in the text?

Permutation equivariance (C) Signup and view all the answers

What is emphasized as a distinctive feature of networks compared to simple sequences & grids, according to the text?

Dynamic topological structure (C) Signup and view all the answers

What is the primary purpose of Graph Neural Networks (GNNs) as stated in the text?

To apply deep learning techniques to graph-based information representations (D) Signup and view all the answers

What is the main challenge addressed by Graph Neural Networks (GNNs) according to the text?

Obtaining embeddings of information contained in graphs (C) Signup and view all the answers

What is one of the recent and hot topics in machine learning research, as mentioned in the text?

Drug discovery, including for COVID-19 (D) Signup and view all the answers

Which type of data is mentioned as an example of structured data that is ever present?

Knowledge graphs (C) Signup and view all the answers

Where has Graph Neural Networks (GNNs) broken into the real world, according to the text?

Drug discovery, including for COVID-19 (B) Signup and view all the answers

What are some examples of applications of GNNs mentioned in the text?

Drug discovery and fake news detection on social media (B) Signup and view all the answers

What is described as one of the fastest growing areas at ICLR (International Conference on Learning Representations) in recent years?

Graph Neural Networks (GNNs) (A) Signup and view all the answers

What does the text describe as a challenge related to structured data?

Applying deep learning techniques to graph-based information representations (A) Signup and view all the answers

Where can Graph Neural Networks be applied according to the text?

Detecting fake news on social media platforms (A) Signup and view all the answers

What is mentioned as a potential application area of Graph Neural Networks?

Detecting anomalies in network traffic data (B) Signup and view all the answers

Which type of GNN features neighbors aggregated with fixed weights?

Convolutional GNN (A) Signup and view all the answers

Which GNN type computes arbitrary vectors (messages) to be sent across edges?

Message-passing GNN (C) Signup and view all the answers

Which GNN type features neighbors aggregated with implicit weights (attention)?

Attentional GNN (B) Signup and view all the answers

Which function is used to compute the attention weights in the Attentional GNN?

$\alpha# = a(\phi(x#), \psi(x))$ (A) Signup and view all the answers

What is the key feature of the Message-passing GNN?

Arbitrary message computation (C) Signup and view all the answers

What is the primary application of the Convolutional GNN?

Homophilous graphs (A) Signup and view all the answers

Which model is useful for computational chemistry, reasoning, and simulation tasks?

Message-passing GNN (C) Signup and view all the answers

What is shared for all nodes in Graph Neural Networks?

$W$ parameters (D) Signup and view all the answers

In Graph Neural Networks, what does the aggregation function $z=$ represent?

Neighborhood aggregation function (C) Signup and view all the answers

In Graph Convolutional Networks (GCN), what does each node compute as a message?

An arbitrary vector to be sent across edges (C) Signup and view all the answers

What does a convolution product between two functions f and g represent in the continuous case?

The integral of the multiplication of the functions over all time (A) Signup and view all the answers

In the discrete case, what does (f ∗ g )(n) represent?

The sum of the multiplication of the functions at different positions (A) Signup and view all the answers

How is an RGB image represented as a function in the spatial domain?

As a 3D function mapping to a single value (A) Signup and view all the answers

What is involved in the convolution operation on images?

Multiplication of pixel values with a convolution kernel (B) Signup and view all the answers

What does the convolution operation in spatial domain imply?

Feature extraction and pattern recognition in images (B) Signup and view all the answers

What does O(i, j) = I ∗ K represent in the context of convolution operation on images?

The element-wise multiplication of input image and convolution kernel (C) Signup and view all the answers

What are the dimensions of the output map after applying a filter of size 5 × 5 × 3 to an input volume of 32 × 32 × 3?

28 × 28 × 1 (B) Signup and view all the answers

Which hyperparameter controls the step taken when sliding the filter?

Stride (A) Signup and view all the answers

What is the primary purpose of parameter sharing in CNNs?

To reduce the computational cost (D) Signup and view all the answers

In the context of PyTorch's Conv2D class, what does the formula Hout = Hin +2∗padding −dilation∗(kernel size−1)−1 +1 represent?

Calculation for output height (C) Signup and view all the answers

What is the constraint on strides as mentioned in the text?

The result of division by stride has to be an integer (B) Signup and view all the answers

What is the main function of the 'groups' parameter in PyTorch's Conv3D class?

To specify the number of groups in a filter (B) Signup and view all the answers

What is the dimension of the output map if a filter of size 5 × 5 × 3 is applied to an input volume of dimension 32 × 32 × 3?

28 × 28 × 1 (C) Signup and view all the answers

What is the primary disadvantage of using a Fully Connected Layer in a CNN?

Loss of spatial information (B) Signup and view all the answers

What does a Convolution Layer with a kernel (filter) of size 5 × 5 × 3 aim to achieve for an input volume of dimension 32 × 32 × 3?

Feature extraction through local connectivity (C) Signup and view all the answers

What is the spatial extent of the local connectivity of each neuron in a Convolution Layer?

Receptive Field (C) Signup and view all the answers

What is the primary function of a Pooling Layer in a CNN?

Downsampling and feature reduction (C) Signup and view all the answers

What is the purpose of linearizing an image in the context of CNN architectures?

To convert it into a vector for feeding into Fully Connected Layers (C) Signup and view all the answers

What is the advantage of using spatially separable convolutions?

Reduced computational load due to fewer kernel parameters (B) Signup and view all the answers

What is the primary purpose of using the pooling layer in a convolutional neural network?

To modify the input volume into a smaller, more manageable representation (B) Signup and view all the answers

What role does batch normalization play in deep learning networks?

It addresses the covariate-shift problem and allows for higher learning rates (D) Signup and view all the answers

What is the main distinguishing feature of VGG-16 architecture in terms of convolutional operations?

It uses 3x3 CONV with a stride of 1, padding of 1, and 2x2 MAX POOL with a stride of 2 (A) Signup and view all the answers

What is the primary function of fully convolutional networks in deep learning applications?

To capture context information and perform spatial analysis simultaneously (A) Signup and view all the answers

What is the computational advantage of using depthwise separable convolutions over typical 2D convolutions?

Reduced number of multiplications and overall computational load (D) Signup and view all the answers

What is the primary purpose of using 1x1 convolutions in GoogleNet's inception module?

To reduce computational complexity by decreasing the number of channels over which 3x3 and 5x5 convolutions operate (A) Signup and view all the answers

What is the main advantage of using residual blocks in ResNet architectures?

To mitigate the optimization problems encountered in deep networks (A) Signup and view all the answers

In Wide Residual Networks, what does 'widening' consistently improve?

The performance across residual networks of different depth (B) Signup and view all the answers

What is the key concept behind ResNeXt's approach to multi-branch aggregated transformations?

Utilizing grouped convolutions to implement multi-branch aggregated transformations (D) Signup and view all the answers

What is the primary focus of the Wide Residual Networks (WRN) paper by Zagoruyko and Komodakis?

Exploring the problem of diminishing feature reuse in original ResNet designs (D) Signup and view all the answers

What distinguishes ResNeXt's approach from VGG, ResNet, and Inception architectures?

Implementing multi-branch aggregated transformations using grouped convolutions (C) Signup and view all the answers

What is the primary focus of DenseNet architecture?

Obtaining representational power through feature reuse (A) Signup and view all the answers

In DenseNet, what is concatenated to subsequent volumes with the same feature-map size?

Information from previous volumes (C) Signup and view all the answers

In transfer learning with CNNs, what is the norm according to the text?

Using CNNs trained on ImageNet in object detection (D) Signup and view all the answers

What is the recommended approach if a dataset has less than 1 million images for training a ConvNet?

Find a large dataset with similar data and train a big ConvNet there (A) Signup and view all the answers

What task can be solved using CNN + RNN according to the provided text?

Image captioning (D) Signup and view all the answers

Which paper is a source for understanding and visualizing DenseNets?

Densely Connected Convolutional Networks, Huang et al., 2018 (B) Signup and view all the answers

In which type of problem is transfer learning with CNNs commonly used?

Image classification (B) Signup and view all the answers

What is the main goal of simplifying the connectivity pattern between layers in DenseNet?

To improve representational power through feature reuse (C) Signup and view all the answers

What does DenseNet focus on in terms of network architectures?

"Skip connections" between layers with different feature-map sizes (C) Signup and view all the answers

What is the primary function of a convolution operation on images?

Applying filters to an image (A) Signup and view all the answers

What is the dimension of the convolution kernel (filter) used in the convolution operation?

3x3 (B) Signup and view all the answers

In the context of convolutions, what does the term 'receptive field' refer to?

The area of an image covered by a filter (A) Signup and view all the answers

What do RGB images represent as a function in the context of convolutional operations?

Color values of pixels (D) Signup and view all the answers

What is the primary purpose of normalization in Convolutional Neural Networks (CNNs)?

To scale input data to a similar range (A) Signup and view all the answers

What does the convolution product between two functions represent in the continuous case?

A spatial correlation (D) Signup and view all the answers

What is the primary disadvantage of using a fully connected layer in Convolutional Neural Networks (CNNs)?

Loss of information about the spatial arrangement of pixels (B) Signup and view all the answers

What is the dimension of the output map when applying a 5x5x3 filter to a 32x32x3 input volume in a Convolution Layer?

28 × 28 × 1 (C) Signup and view all the answers

What is the purpose of using a Convolution Layer in Convolutional Neural Networks (CNNs)?

To capture local connectivity of each neuron (A) Signup and view all the answers

What is achieved by using a filter of size 5x5x3 on a 32x32x3 input volume in a Convolution Layer?

Extraction of features from local regions of the input (B) Signup and view all the answers

What does the size of the receptive field represent in a Convolution Layer?

The spatial extent of the local connectivity of each neuron (D) Signup and view all the answers

What is the dimension of an output map when applying a filter to an input volume in a Convolution Layer?

Depends on the size and number of filters used (D) Signup and view all the answers

What is the formula to compute the height of the output map in a convolution layer?

Height = Height + 2padding - dilation(kernel size-1) - stride + 1 (C) Signup and view all the answers

Which hyperparameter controls the size of the output volume by determining the step taken when sliding the filter?

Stride (C) Signup and view all the answers

What does parameter sharing in CNNs aim to control?

The number of filter parameters (C) Signup and view all the answers

What is the main purpose of using a backbone network like VGG-16 in Faster RCNN?

To extract high-level features from input images (C) Signup and view all the answers

What is the significance of ensuring equivariance for graph neural networks?

It ensures that the model preserves symmetries present in the input data (B) Signup and view all the answers

What does the formula Hout = Hin +2padding - dilation(kernel size-1)-1 +1 represent in PyTorch's Conv2D class?

It represents the computation for height in the output volume (B) Signup and view all the answers

What does the Hessian matrix of a scalar-valued function represent?

Second-order partial derivatives (B) Signup and view all the answers

In ofﬂine learning, what type of data is typically used to optimize functions?

Batch data (C) Signup and view all the answers

For linear regression with Mean Squared Error (MSE) loss function, what does the gradient represent?

First-order partial derivatives (C) Signup and view all the answers

What is the primary purpose of the Gradient Descent algorithm?

Minimize the function value (C) Signup and view all the answers

With second-order optimization using Newton’s algorithm, what kind of updates are performed?

Second-order updates (B) Signup and view all the answers

What is a challenge associated with Second Order Optimization?

Need to invert the Hessian matrix (C) Signup and view all the answers

What distinguishes Stochastic Gradient Descent (SGD) from traditional Gradient Descent?

Computation of noisy gradients (D) Signup and view all the answers

What is the primary concern when using Stochastic Gradient Descent (SGD)?

Data normalization and scaling (A) Signup and view all the answers

What is the main advantage of using Momentum in the SGD algorithm?

Dampens oscillations in optimization (C) Signup and view all the answers

What is the key feature of Adagrad in optimization?

Adaptive learning rate for individual parameters (B) Signup and view all the answers

What is a concern addressed by Nesterov Accelerated Gradient in optimization?

Momentum overshooting in SGD algorithm (C) Signup and view all the answers

What does Adagrad aim to achieve by adapting learning rates for individual parameters?

Reduced variance in gradient updates for each parameter (C) Signup and view all the answers

What is the advantage of spatially separable convolutions?

Reduced computational load and fewer kernel parameters (A) Signup and view all the answers

What is the primary purpose of the pooling layer in a convolutional neural network?

To modify the input volume into a smaller and more manageable representation (C) Signup and view all the answers

What is the computational advantage of depthwise separable convolutions?

Fewer multiplications and reduced computational load (C) Signup and view all the answers

What is the purpose of batch normalization in convolutional neural networks?

To address covariate-shift problem and allow higher learning rates (B) Signup and view all the answers

What is the primary focus of Fully Convolutional Networks (FCNs)?

Semantic segmentation and generative networks (B) Signup and view all the answers

What are the downsampling ratios commonly used for CNN feature maps in Anchor Boxes for object detection?

(2, 2) and (3, 3) (B) Signup and view all the answers

What is the primary purpose of using bottleneck layers in the GoogleNet architecture?

To reduce the computational complexity (C) Signup and view all the answers

What is the main benefit of using residual blocks in the Residual Network (ResNet) architecture?

To make the network deeper without increasing the number of parameters (D) Signup and view all the answers

What is the primary focus of Wide Residual Networks (Wide ResNet)?

Increasing both depth and width of residual networks (A) Signup and view all the answers

What is the main objective of using grouped convolutions in ResNeXt?

Implementing multi-branch aggregated transformations (C) Signup and view all the answers

What is the significance of using a global average pooling layer in GoogleNet?

To spatially average across each feature map before final FC layer (A) Signup and view all the answers

What is the main benefit of using skip connections in the Residual Network (ResNet) architecture?

To address optimization problems in deep networks (B) Signup and view all the answers

What is the role of the Region Proposal Network (RPN) in Faster R-CNN?

To propose regions where objects might be found (B) Signup and view all the answers

What is the purpose of the RoI Align Layer in Mask-RCNN?

To avoid quantization of RoI coordinates or spatial bins to feature map grid (D) Signup and view all the answers

What are the downsampling ratios of CNN feature maps used in Faster R-CNN?

3 scales, 3 aspect ratios (D) Signup and view all the answers

What is a limitation of using regular convolutions for learning spatially-local biases?

Limited in processing patterns using rectangular patterns (B) Signup and view all the answers

What is the main difference between single stage predictors and multi-stage predictors in object detection approaches?

Single stage predictors perform both object classification and object bounding box regression in a single forward pass (B) Signup and view all the answers

What does the deformation mechanism aim to achieve in deformable convolutions?

Learn non-rectangular patterns by computing pixel offsets (B) Signup and view all the answers

What are the different backbones used in Mask-RCNN?

ResNext-101, FPN (Feature Pyramid Network) (D) Signup and view all the answers

What changes were made to Mask-RCNN compared to Faster R-CNN?

Replaces RoI Pooling with a RoI Align Layer (C) Signup and view all the answers

What is the primary difference between proposal-based and segmentation-based methods in instance segmentation?

Proposal-based methods propose regions of interest and extract features, while segmentation-based methods start from segmentation as the first objective and learn specially designed transformations or instance boundaries. (B) Signup and view all the answers

What is the purpose of Atrous Spatial Pyramid Pooling in DeepLab-v3 for semantic segmentation?

To resample features at different scales for more accurate region classification. (D) Signup and view all the answers

What is the main idea behind using dilated convolutions in DeepLab-v3 for semantic segmentation?

To enable larger effective receptive field sizes and capture spatial context information. (D) Signup and view all the answers

What is the architecture that builds upon Faster-RCNN in the case of MaskRCNN for instance segmentation?

Region of Interest (RoI) proposal based architecture. (B) Signup and view all the answers

What are the two main streams of methods in instance segmentation?

Proposal-based and Segmentation-based. (D) Signup and view all the answers

What does Semantic Segmentation in DeepLab-v3 emphasize through the use of dilated convolutions and Atrous Spatial Pyramid Pooling?

Capturing spatial context information and resampling features at different scales for accurate region classification. (C) Signup and view all the answers

What is the purpose of resampling features at different scales in Atrous Spatial Pyramid Pooling?

To enable more accurate region classification by capturing spatial context information. (C) Signup and view all the answers

What does the reduction factor for image resolution need to be limited to in semantic segmentation according to DeepLab-v3?

16 (A) Signup and view all the answers

What does the Atrous Spatial Pyramid Pooling use to extract content information from several scale levels at the same time?

(A combination of) atrous convolutions and global average pooling. (B) Signup and view all the answers

(Atrous Convolution Layer) vs (Dilated Convolution Layer), which one is used in DeepLab-v3 to extract larger information context?

(Dilated Convolution Layer) (C) Signup and view all the answers

What does the Hessian matrix represent in optimization?

It represents the square matrix of second-order partial derivatives. (D) Signup and view all the answers

What does the gradient vector point in the direction of in gradient descent?

It points in the direction of maximum decrease of the function. (B) Signup and view all the answers

What is the main purpose of using a global average pooling layer in GoogleNet?

To reduce the computational complexity of the network. (C) Signup and view all the answers

What does the Logistic Sigmoid Function do?

It introduces non-linearity to the network. (C) Signup and view all the answers

What is the key concept behind second-order optimization methods?

They use second-order derivatives to directly optimize the loss function. (C) Signup and view all the answers

What is the primary focus of Rethinking Attention with Performers in terms of attention mechanisms?

To improve interpretability of attention mechanisms. (D) Signup and view all the answers

What is one of the recent and hot topics in machine learning research, as mentioned in the text?

"Group normalization" for improving network performance. (B) Signup and view all the answers

What is the function that 'squeezes in' the weighted input into a probability space in logistic regression?

"Logistic Sigmoid Function" (D) Signup and view all the answers

What problem does ReLU activation function primarily address in CNNs?

Vanishing gradient problem. (B) Signup and view all the answers

Which architecture was the YOLOv4 backbone selected based on?

CSPDarknet53 (A) Signup and view all the answers

What is one of the limitations of using BatchNorm in tasks such as video prediction, segmentation, and medical image processing?

Difficulty with training small mini-batches or non-i.i.d. data distribution. (C) Signup and view all the answers

What is the primary purpose of using Bag of Freebies and Bag of Specials in YOLOv4?

Improved performance on small object detection (A) Signup and view all the answers

What is the main limitation of Temporal Convolutional Network (TCN) in test/evaluation mode?

Requires a large memory space (B) Signup and view all the answers

What is the focus of InceptionTime, introduced in the article 'InceptionTime: Finding AlexNet for Time Series Classification'?

Time series classification (B) Signup and view all the answers

How does the InceptionTime Network reduce variance in classification accuracy?

By using ensemble modeling (A) Signup and view all the answers

What is the primary advantage of using causal convolutions in Temporal Convolutional Network (TCN)?

No leakage from the past (C) Signup and view all the answers

What is the main modification in YOLOv5 compared to YOLOv4?

Implementation primarily in PyTorch (D) Signup and view all the answers

What is the primary focus of YOLOv2 in comparison to YOLOv1?

Improving detection accuracy for small objects (D) Signup and view all the answers

Which loss function addresses the problem of nonoverlapping bounding boxes in YOLOv4?

CioU (complete IoU loss) (C) Signup and view all the answers

What is the purpose of using anchor boxes in YOLOv2?

To scale the predicted box w.r.t. defined anchor boxes (C) Signup and view all the answers

What is the primary function of a Convolution Layer in a CNN?

Dimensionality reduction and feature extraction (D) Signup and view all the answers

What was the significant change in object class classification in YOLOv3 compared to YOLOv1 and YOLOv2?

Using individual binary cross-entropy for each class label (A) Signup and view all the answers

What does the gradient vector point in the direction of in gradient descent?

The steepest ascent direction (C) Signup and view all the answers

What is the limitation related to small bounding boxes versus large bounding boxes in the original YOLO architecture?

Loss function treats errors the same in small and large bounding boxes (C) Signup and view all the answers

What was the primary improvement focus of YOLOv3 compared to its predecessors?

Improvement in detection accuracy, especially for small objects (C) Signup and view all the answers

What was the key purpose of using Darknet-53 in YOLOv3?

Tackling the vanishing gradient problem using residual layers (C) Signup and view all the answers

What is the concept of anchor boxes in YOLOv2?

Scaling the predicted box w.r.t. defined anchor boxes (B) Signup and view all the answers

What was the primary limitation of the original YOLO architecture related to small objects appearing in groups?

Struggles with objects of small sizes that appear in groups (B) Signup and view all the answers

What was a significant change in class conditional probability prediction in YOLOv1?

Each grid cell predicts B bounding boxes + class conditional probability (B) Signup and view all the answers

What was the emphasis of YOLOv2 to tackle the vanishing gradient problem?

Using 23 residual layers at regular intervals (C) Signup and view all the answers

Which type of recurrent neural network (RNN) cells are commonly used due to their additive interactions improving gradient flow?

LSTM (A) Signup and view all the answers

What technique can be used to control exploding gradients in RNNs?

Gradient Clipping (A) Signup and view all the answers

What is the primary reason for using Layer Normalization in linear mappings of the RNN?

Control exploding gradients (B) Signup and view all the answers

What is the default initialization for the initial state (h(0)) in RNNs?

0 (B) Signup and view all the answers

What is the main purpose of using noisy initial state in RNNs?

Prevent overfitting (C) Signup and view all the answers

In the context of RNNs, what is the primary objective of using stacked recurrent nets?

Handle a high number of weights effectively (A) Signup and view all the answers

What is the primary purpose of summing the outputs of all layers in stacked recurrent nets?

Handle sequence length variations effectively (A) Signup and view all the answers

What technique is commonly used to address the slow remembering issue in RNNs?

"Forget gate bias" initialization (D) Signup and view all the answers

When does vanishing gradient in RNNs get controlled with additive interactions?

"LSTM" usage (C) Signup and view all the answers

What is a common method for preventing overfitting in RNNs?

Regularization with dropout (B) Signup and view all the answers

What is an example task that can be solved with Recurrent Neural Networks (RNNs) according to the text?

Translate a sequence of words into another sequence of words (A) Signup and view all the answers

What is the primary focus of LSTM in the context of Recurrent Neural Networks (RNNs)?

Remembering long-term dependencies in sequential data (A) Signup and view all the answers

What task involves classifying images by taking a series of 'glimpses'?

Image Captioning (B) Signup and view all the answers

What does the Elman RNN model primarily aim to achieve?

Remembering long-term dependencies in sequential data (D) Signup and view all the answers

What does the term 'vanishing and exploding gradients' refer to in the context of RNN training?

Decrease or increase in the magnitude of gradients during backpropagation (D) Signup and view all the answers

What task involves generating images one piece at a time?

Translation (D) Signup and view all the answers

What does LSTM primarily focus on in the context of RNNs?

Remembering long-term dependencies in sequential data (A) Signup and view all the answers

What task is an example of sequential processing of non-sequence data?

Classify images by taking a series of 'glimpses' (C) Signup and view all the answers

What is achieved by using RNNs again in the decoder of a seq2seq model?

Incorporating long-term dependencies in sequential data (C) Signup and view all the answers

What distinguishes Longformer from other efficient transformers?

Incorporating global and sliding patterns for token blocks processing (A) Signup and view all the answers

What type of neural network has an 'internal state' that is updated as a sequence is processed?

Recurrent Neural Network (RNN) (D) Signup and view all the answers

In the context of RNNs, what does the 'unrolled RNN' diagram visually represent?

The unfolding of RNN through time (C) Signup and view all the answers

What function is used to update the hidden state in a vanilla RNN at each time step?

$h = f(W_hh h_{t-1} + W_{ih}x_t)$ (D) Signup and view all the answers

What does the 'Sequence to Sequence' model aim to achieve in the context of RNNs?

Producing output sequence from single input vector (B) Signup and view all the answers

In the provided text, what example task demonstrates the need for RNNs to handle variable sequence length inputs and outputs?

Video captioning (B) Signup and view all the answers

What is the primary focus of the 'Character-level Language Model' example discussed in the text?

Text generation (C) Signup and view all the answers

What is the purpose of 'Sampling Softmax' in the 'Character-level Language Model' example?

Sampling characters one at a time at test-time (B) Signup and view all the answers

What does the 'Many-to-one' computational graph represent in the context of RNNs?

Encoding input sequence in a single vector (D) Signup and view all the answers

What is the purpose of truncated backpropagation through time (TBPTT) in recurrent neural networks?

To carry hidden states forward in time forever, but only backpropagate for some smaller number of steps (B) Signup and view all the answers

What does the Long Short Term Memory (LSTM) architecture provide an easier way for the model to learn?

Temporal context (B) Signup and view all the answers

What makes it easier for the RNN to preserve information over many timesteps in the LSTM architecture?

Forget gate (f) (B) Signup and view all the answers

What control does the LSTM architecture provide over gradient values through suitable parameter updates?

Elementwise multiplication by f, i, g, and o gates (A) Signup and view all the answers

What scenario does Truncated BPTT (TBPTT) with k1=1 imply?

Not enough temporal context to learn (A) Signup and view all the answers

What change in RNN architecture addressed the vanishing/exploding gradient problem?

Long Short Term Memory (LSTM) (A) Signup and view all the answers

What is the main advantage of using Long Short Term Memory (LSTM) over vanilla RNN?

Easier for the model to learn long-distance dependencies (B) Signup and view all the answers

What operation ensures that information of a cell is preserved indefinitely in the LSTM architecture?

$f = 1$ and $i = 0$ (D) Signup and view all the answers

Which scenario leads to exploding gradients in TBPTT(k1, k2)?

>1 (B) Signup and view all the answers

What does TBPTT(1, n) in recurrent neural networks imply?

Timesteps are processed one at a time followed by an update that covers all timesteps seen so far (C), Timesteps are processed one at a time followed by an update that covers all timesteps seen so far (D) Signup and view all the answers

What is a disadvantage of using RNNs for long input sequences?

Context vector is forced to 'compress' a lot of information before decoding any of it (A) Signup and view all the answers

What is the purpose of using Bidirectional LSTM?

To address the issue of long range dependencies (A) Signup and view all the answers

What is the customization point in Bidirectional LSTM?

Sum (D) Signup and view all the answers

What type of time series analysis is ConvLSTM applied to?

Visual information (B) Signup and view all the answers

What does ConvLSTM replace internal matrix multiplications with?

Convolution operations (B) Signup and view all the answers

What are the advantages of ConvLSTM over fully connected LSTM?

Better at handling spatio-temporal relations (A) Signup and view all the answers

What is the primary difference between univariate and multivariate time series?

The number of features in each time step (B) Signup and view all the answers

What does single-step learning setup in time series forecasting focus on predicting?

$x_t$ given $[x_{t-k}, x_{t-k+1}, …, x_{t-1}]$ (B) Signup and view all the answers

What considerations need to be addressed when applying RNNs to timeseries?

Seasonal and Trend Decomposition - STL, Variance stabilization, Multiple Output Strategy selection (B) Signup and view all the answers

What is discussed in the context of regularization and normalization in RNNs?

All options are correct (A) Signup and view all the answers

What is the primary focus of DenseNet architecture?

Introducing direct connections between any two layers with the same feature-map size (D) Signup and view all the answers

What is the norm for transfer learning with Convolutional Neural Networks (CNNs)?

Using CNNs trained on ImageNet in object detection and image captioning (A) Signup and view all the answers

What does DenseNet use to control the amount of concatenation between feature maps?

Growth factor (C) Signup and view all the answers

What is the primary focus of the Long-Range Arena Challenge benchmark?

Natural language processing tasks (B) Signup and view all the answers

In DenseNet, what does the growth factor control?

Amount of feature-map concatenation (A) Signup and view all the answers

What is the primary function of the RoI proposal based approach in Instance Segmentation?

Generating region proposals for object detection (D) Signup and view all the answers

What is the main benefit of using skip connections in the Residual Network (ResNet) architecture?

Improving gradient flow and easing optimization (B) Signup and view all the answers

'Transfer learn to your dataset' is a key takeaway when dealing with a dataset that has:

Less than 1 million images (D) Signup and view all the answers

What is the primary weakness of Adagrad according to the text?

Accumulation of squared gradients leading to a shrinking learning rate (C) Signup and view all the answers

What is Adadelta's solution to Adagrad's weakness?

Restricting past gradients to a window and computing a decaying running average (A) Signup and view all the answers

What does the RMSProp optimization algorithm aim to address?

Biased estimation of the first and second moment averages (D) Signup and view all the answers

What is the primary similarity between Adadelta and RMSProp optimization algorithms?

Both algorithms store exponentially decaying averages of squared gradients (A) Signup and view all the answers

What is the main distinguishing feature of Adam optimization algorithm?

It stores exponentially decaying averages of past gradients with momentum (A) Signup and view all the answers

What is the purpose of early stopping in optimization?

To monitor error on validation set and stop if validation error does not improve enough (D) Signup and view all the answers

What transformation function can be applied for variance stabilization in data?

Batch Normalization (C) Signup and view all the answers

What is used to make training more robust to poor initialization or when having deep and complex networks?

Gradient Noise (C) Signup and view all the answers

What factor contributes to the struggles of original YOLO in detecting objects of small sizes that appear in groups?

Receptive field size (B) Signup and view all the answers

What does each neuron in a Multi-Layer Perceptron (MLP) compute?

A weighted sum of its inputs followed by a nonlinear activation function (D) Signup and view all the answers

In which GNN layer are the features of neighbors aggregated with implicit weights (attention)?

GAT Layer (D) Signup and view all the answers

What is the primary purpose of sequence-to-sequence models in neural networks?

To work with variable-length input and output sequence pairs (C) Signup and view all the answers

What is the purpose of the encoder-decoder model in sequence-to-sequence models?

To encode the source (input) into a single vector and decode it into a sequence of output tokens (C) Signup and view all the answers

What is the advantage of using sequence-to-sequence models?

They can be used in auto-encoding setup (D) Signup and view all the answers

What does the decoder model in sequence-to-sequence models do during the forward pass?

Receive prediction, next state, and next cell state (D) Signup and view all the answers

When is the loop broken during decoding in a sequence-to-sequence model?

When the decoder predicts the (END) token (A) Signup and view all the answers

What type of tasks can sequence-to-sequence models handle effectively?

Time series prediction tasks (C) Signup and view all the answers

What is the primary function of an RNN in the context of sequence-to-sequence models?

To decode the 'context' vector into a sequence of output tokens (A) Signup and view all the answers

What distinguishes seq2seq models from other neural network architectures?

They are designed for auto-encoding setup (A) Signup and view all the answers

What capability allows seq2seq models to work with variable-length input and output sequences?

'Context' vector generation capability (A) Signup and view all the answers

What type of analysis tasks can benefit from using the 'context vector' (z) generated by seq2seq models?

Time series analysis tasks (C) Signup and view all the answers

What is the purpose of approximate attention computation using more efficient operations?

To make self-attention more efficient (A) Signup and view all the answers

What role do Key and Query embeddings play in defining the attention pattern?

They are used for hashing and clustering (A) Signup and view all the answers

What does the Blockwise Attention pattern do?

Divides the sequence into local blocks and restricts attention within them (D) Signup and view all the answers

What is the purpose of Strided Patterns in the context of efficient attention?

To decrease time complexity to one linear in sequence length / stride (D) Signup and view all the answers

How does the Diagonal (sliding window) Patterns reduce time complexity?

It reduces time complexity to one linear in sequence length (A) Signup and view all the answers

What is the primary purpose of Global Attention Patterns?

To apply attention to special tokens like [CLS] (C) Signup and view all the answers

What is the distinctive feature of Longformer, as compared to other efficient transformers?

It showcases use of sliding, strided, and global attention patterns (C) Signup and view all the answers

What does BigBird's attention pattern compose of?

Global, sliding and random patterns of token blocks (D) Signup and view all the answers

What does Dilated sliding Window achieve in Longformer?

It achieves a l x d x w receptive field with different dilation configs per head (A) Signup and view all the answers

What are the two sets of projections learned in Longformer?

Qs, Ks, Vs for sliding window; Qg, Kg, Vg for global attention (D) Signup and view all the answers

What is the main purpose of the Multi-Head Attention in the Transformer Architecture?

To summarize values based on similarity between value keys and the query (C) Signup and view all the answers

What is the primary addition in Transformer-XL to facilitate the recurrence strategy?

Incorporates relative position encoding scheme (D) Signup and view all the answers

What does the Scaled Dot-Product Attention compute in the Transformer Architecture?

Similarity based on dot-product and a function ϕ (B) Signup and view all the answers

What is the primary focus of the Long-Range Arena Challenge in relation to efficient transformers?

Benchmarking the performance of efficient attention mechanisms (B) Signup and view all the answers

What is the primary challenge when dealing with large sequences in the Transformer Architecture?

Quadratic bottleneck in sequence length due to multi-headed attention (C) Signup and view all the answers

Which paper presents a method for long document understanding using blockwise self-attention?

Qiu et al., 2014 (C) Signup and view all the answers

What is the purpose of the Efficient Transformer Techniques discussed in the text?

To address challenges in handling large sequence lengths (D) Signup and view all the answers

Which paper introduces 'Big bird: Transformers for longer sequences'?

Zaheer et al., 2018 (C) Signup and view all the answers

What is the primary feature of Linformer, as discussed in the text?

Self-attention with linear complexity (D) Signup and view all the answers

What represents the sequence length (l) and feature dimensionality (d) in Scaled Dot-Product Attention?

Sequence length and feature dim (C) Signup and view all the answers

What is the main idea behind 'Reformer: The efficient transformer'?

Efficiently handling long sequences using reversible layers (B) Signup and view all the answers

What does the Attention Operation in the Transformer Architecture summarize based on?

Similarity between value keys and the query (A) Signup and view all the answers

What does the Rethinking Attention with Performers paper primarily focus on?

Introducing a novel attention mechanism (C) Signup and view all the answers

What is a serious challenge when large sequences are required in the Transformer Architecture?

Quadratic bottleneck in sequence length due to multi-headed attention (B) Signup and view all the answers

Which paper discusses 'Longformer: The long-document transformer'?

Beltagy et al, 2020 (A) Signup and view all the answers

What does the Dot-Product Similarity compute in Scaled Dot-Product Attention?

$ ext{exp}( ext{softmax}(x_i) = Q_i K_j^T rac{1}{ oot{2} t{d}})$ (B), $ ext{exp}( ext{softmax}(x_i) = Q_i K_j^T rac{1}{ oot{2} t{d}})$ (D) Signup and view all the answers

'Data-Independent Attention Patterns' and 'Data-Dependent Attention Patterns' fall under which category of Efficient Transformer Techniques?

'Efficient computation of attention patterns' (A) Signup and view all the answers

What is the main focus of the Efficient transformers: A survey paper?

'Surveying and discussing efficient transformers' (C) Signup and view all the answers

'What is a key takeaway from the Transformer Survey Blog?'

Methods for incorporating long-term context in transformers (A) Signup and view all the answers

'Recurrence in Transformer Architectures' presents challenges related to which aspect of computation?

Dealing with long-range dependencies. (A) Signup and view all the answers

What is the purpose of using global and random attention patterns?

To reduce the overall memory consumption (C) Signup and view all the answers

What problem does the Reformer architecture address?

Memory storage for backpropagation in multi-layer models (D) Signup and view all the answers

What is the key idea behind Linformer's approach to reduce memory complexity?

Down-projecting the sequence length (A) Signup and view all the answers

How is Attention interpreted in the context of Performer's approach?

As an infinite feature space kernel function (B) Signup and view all the answers

What does Angular Locality Sensitive Hashing strive to achieve?

Dividing tokens into buckets based on their angular proximity (D) Signup and view all the answers

What is the primary focus of Efficient Transformers with respect to attention mechanisms?

Optimizing the computation and memory usage of attention mechanisms (B) Signup and view all the answers

What problem does Reversible Residual Layer aim to address?

The need to store forward-pass activations for backpropagation (C) Signup and view all the answers

What does the Kernel Interpretation approach enable in terms of attention?

Approximation of softmax-based dot-product attention using polynomial kernels (C) Signup and view all the answers

What is an advantage of Atrous Spatial Pyramid Pooling when dealing with long sequences?

It extracts features from different scales at once (B) Signup and view all the answers

What does Linformer aim to achieve by using low-rank matrix approximation?

Efficient approximation of softmax-based dot-product attention (B) Signup and view all the answers

In the context of time series analysis, what is a typical task related to industrial settings?

Detecting data traffic outliers in telecommunication (D) Signup and view all the answers

Which domain is mentioned as an example in the context of time series analysis?

Healthcare monitoring (D) Signup and view all the answers

What is a specific example of time series data mentioned from the domain of economics and finance?

Stock prices (B) Signup and view all the answers

In the context of time series analysis, what type of prediction task is mentioned in relation to industrial settings?

Forecasting water demand in London (D) Signup and view all the answers

What is an example of a typical analysis task mentioned in the context of time series analysis?

Forecasting and classification (A) Signup and view all the answers

Which task is mentioned as a typical domain for time series analysis?

Weather forecasting (A) Signup and view all the answers

What is an example of a domain mentioned in the context of time series analysis?

Financial trading strategy development (A) Signup and view all the answers

In the context of time series analysis, what is a specific example from the domain of healthcare?

Long term patient monitoring to predict disease outcomes (C) Signup and view all the answers

What is a specific type of data mentioned as an example in the context of time series analysis?

Weather patterns from meteorological stations (A) Signup and view all the answers

In the context of industrial settings, what is an example task related to transportation mentioned for time series analysis?

Traffic prediction and route optimization (D) Signup and view all the answers

What is the purpose of Mean Absolute Error (MAE) in time series forecasting?

Captures both large and small errors (B) Signup and view all the answers

What is the primary challenge in classification tasks for time-ordered sequences?

Variable length sequences (A) Signup and view all the answers

What is the main objective of anomaly detection in time series analysis?

Identify points or periods where data distribution shifts significantly (A) Signup and view all the answers

Which benchmark dataset is typically used for short-term forecasting analysis?

M4 benchmark (D) Signup and view all the answers

What is the purpose of Exponential Smoothing in time series analysis?

Capture short-term trends (B) Signup and view all the answers

What are the typical challenges encountered in time series classification tasks?

Variable length sequences and Transferability across domains (A) Signup and view all the answers

Which metric is useful for comparing forecast accuracies across different time series with varying scales?

Mean Absolute Percentage Error (MAPE) (D) Signup and view all the answers

What is the main challenge associated with detecting anomalies in time series data?

Seasonality, Trends, non-stationarity and Noise and Outliers (D) Signup and view all the answers

Which method is typically used to remove noise and transient outliers in time series data?

Savitzky-Golay Filter (D) Signup and view all the answers

What is the advantage of using Root Mean Squared Error (RMSE) as a forecasting metric?

Interpretable and still penalizes large errors (B) Signup and view all the answers

What is the main purpose of STL decomposition in time series analysis?

To break down the time series into seasonal, trend, and residual components (A) Signup and view all the answers

What technique can be used to replace missing values with the mean, median, or mode of available values in time series data?

Mean/Median/Mode imputation (B) Signup and view all the answers

What is the primary function of Trend Normalization in time series analysis?

To subtract the trend value of the last item in the input sequence from both input and output (C) Signup and view all the answers

What does an ARMA model expect from the time series data?

The time series to be stationary (D) Signup and view all the answers

How are the hyperparameters for ARIMA model chosen?

By minimizing information criteria such as AIC and BIC (C) Signup and view all the answers

What is the main focus of RNN models for time series forecasting?

To predict future values based on past values and forecasting errors (C) Signup and view all the answers

What is the purpose of reverse deseasonalization in post-processing for RNN models?

To add back the seasonality components after forecasting (A) Signup and view all the answers

What efficient attention pattern involves reaching a receptive field that can be 10^4 tokens wide for small values of d?

(Performers) in terms of attention mechanisms (D) Signup and view all the answers

Which transformation function can be applied for variance stabilization in data?

(Log) transform (A) Signup and view all the answers

What is the primary benefit of using skip connections in RNN models?

To allow gradients to flow directly through to earlier layers during training and alleviate vanishing gradient problem (C) Signup and view all the answers

What is the primary focus of the paper 'Recurrent neural networks for time series forecasting: Current status and future directions'?

Forecasting using RNN models (D) Signup and view all the answers

Which paper introduces a model designed for long-term predictions and large input windows, involving a built-in Series Decomposition Block and replacing standard self-attention with auto-correlation?

Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting (D) Signup and view all the answers

Which technique involves converting the 1D time series to a 2D space to simultaneously model intra- and inter-period variations?

Temporal 2D-Variation Modeling (C) Signup and view all the answers

What type of analysis tasks can benefit from using the 'context vector' (z) generated by seq2seq models?

Classification and Anomaly Detection (D) Signup and view all the answers

What does the model 'TS2VEC' primarily aim to achieve?

Universal representation of time series (A) Signup and view all the answers

What is the main focus of the paper 'Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting'?

Forecasting using RNN models (A) Signup and view all the answers

Which model is associated with 'Temporal 2D-Variation Modeling' for general time series analysis?

'TimesNet' (B) Signup and view all the answers

What does 'Hierarchical Contrasting' aim to achieve according to the text?

'Representations at various scales learn Contextual Representations' (A) Signup and view all the answers

'Drawing a Recurrent Neural Network For Image Generation' is associated with which task?

'Image Generation' (B) Signup and view all the answers

In what scenarios have GNNs broken into the real world, as mentioned in the text?

Solving real-world problems such as recommendation systems and fraud detection (D) Signup and view all the answers

What is the primary focus of Graph Neural Networks (GNNs) according to the text?

Obtaining embeddings of information contained in graphs (A) Signup and view all the answers

Where has Graph Neural Networks (GNNs) broken into the real world, according to the text?

Drug discovery (C) Signup and view all the answers

What is emphasized as a distinctive feature of networks compared to simple sequences & grids, according to the text?

Handling structured data (B) Signup and view all the answers

What is one of the recent and hot topics in machine learning research, as mentioned in the text?

Graph Neural Networks (GNNs) (C) Signup and view all the answers

What does the Scaled Dot-Product Attention compute in the Transformer Architecture, according to the text?

Semantic similarity between words (B) Signup and view all the answers

What transformation function can be applied for variance stabilization in data, according to the text?

Log transformation (B) Signup and view all the answers

When does vanishing gradient in RNNs get controlled with additive interactions, according to the text?

At the hidden layers (D) Signup and view all the answers

What property does Maximum Likelihood Estimation (MLE) have for i.i.d. data, according to the text?

Convergence guarantee (C) Signup and view all the answers

What is a desirable property for a graph convolutional layer according to the text?

'Local and translation invariant' behavior (B) Signup and view all the answers

What is the purpose of using global and random attention patterns, according to the text?

Solving combinatorial optimization problems (C) Signup and view all the answers

What type of GNN layer features fixed weights for neighbor aggregation?

Convolutional GNN (A) Signup and view all the answers

Which GNN layer uses attention to compute implicit weights for neighbor aggregation?

Attentional GNN (B) Signup and view all the answers

Which GNN layer is most suitable for computing arbitrary vectors (messages) to be sent across edges?

Message-passing GNN (C) Signup and view all the answers

What is the primary principle for building and training GNNs outlined in the text?

Train on a set of nodes (B) Signup and view all the answers

Which foundational GNN models are specifically mentioned in the text?

GCN, MoNet, GraphNets (A) Signup and view all the answers

Which type of function is permutation invariant?

A function that remains the same when the order of elements is changed (D) Signup and view all the answers

What characterizes a Deep Sets model according to the text?

Learnable functions like MLPs and sum aggregation are critical (A) Signup and view all the answers

What is the main difference in applying permutation invariance and equivariance to graphs?

Permutations act accordingly on both rows and columns of the adjacency matrix for invariance and equivariance (C) Signup and view all the answers

What does enforcing locality in equivariant set functions involve?

Applying a shared function to every node in isolation (D) Signup and view all the answers

How can permutation equivariant functions on graphs be constructed?

By applying a permutation-invariant local function over all neighbourhoods (B) Signup and view all the answers

What is the purpose of a GNN layer according to the text?

To aggregate features from neighbours with implicit weights (C) Signup and view all the answers

What is a common lingo used for 𝐅 in the context of graph neural networks?

"GNN layer" (A) Signup and view all the answers

What does extracting neighbourhood features involve in graph neural networks?

Aggregating features from all nodes in the graph (C) Signup and view all the answers

What is an important constraint for ensuring equivariance in graph neural networks?

"Permutation-invariant" local functions over all neighbourhoods (B) Signup and view all the answers

What does applying a permutation matrix to 𝜙 involve in graph neural networks?

"Preserving the order" of nodes in 𝐗 𝒩. while applying 𝜙 (B) Signup and view all the answers

What is the main challenge in learning the mapping function for graph data?

Learning the mapping function f (D) Signup and view all the answers

How are graphs similar to images?

Graphs can be seen as a generalization of images as each node corresponds to a pixel (A) Signup and view all the answers

What is a desirable property for a graph convolutional layer?

Applicability to inductive problems (B) Signup and view all the answers

What does the encoder do in the context of deep learning methods based on graph neural networks?

Maps each node to a low-dimensional vector (A) Signup and view all the answers

What tasks can be solved with Graph Neural Networks (GNNs)?

Identification of densely linked clusters of nodes (B) Signup and view all the answers

What is a key property of node embedding in the context of building and training GNNs?

Permutation Invariance (A) Signup and view all the answers

What is the focus of learning on sets within the context of graph analysis?

Analyzing point clouds and LiDAR data (D) Signup and view all the answers

What does the symmetry group 𝔊 aim to achieve in the context of graph analysis?

Change the node order using permutation matrices (B) Signup and view all the answers

What is a key aspect focused on in Linformer for network performance enhancement?

Improvement in network performance by reducing memory complexity (D) Signup and view all the answers

What is the primary focus of Long-Range Arena Challenge in relation to efficient transformers?

The evaluation of efficient transformers for longer sequences (A) Signup and view all the answers

What are the desirable properties for a graph convolutional layer?

Computational and storage efficiency, fixed number of parameters, localisation, specifying different importances to different neighbors, applicability to inductive problems Signup and view all the answers

What are the tasks that can be solved with GNNs according to the text?

Node classification, link prediction, community detection, network similarity Signup and view all the answers

What is the symmetry group 𝔊 defined in the context of learning on sets?

𝔊 is the n-element permutation group Σn, where n is the number of nodes. Signup and view all the answers

What is the purpose of permutation invariance in the context of learning on sets?

Permutation invariance ensures that functions 𝑓(𝐗) over sets do not depend on the order of the nodes. Signup and view all the answers

What are the node embedding properties mentioned in the text for building and training GNNs?

Permutation invariance, permutation equivariance Signup and view all the answers

What is the general focus of GNNs according to the text?

To generalize the operation of CNNs to work on arbitrary graphs. Signup and view all the answers

What is the general framework for building and training GNNs?

Node, Subgraph, Graph Encoders Signup and view all the answers

What is the encoder's role in deep learning methods based on graph neural networks?

The encoder maps each node to a low-dimensional vector in the input graph. Signup and view all the answers

What are the challenges associated with graph convolutions according to the text?

What does the similarity function specify in the context of deep learning methods based on graph neural networks?

It specifies how the relationships in vector space map to the relationships in the original network. Signup and view all the answers

What are the three 'flavours' of GNN layers?

Convolutional GNN, Attentional GNN, Message-passing GNN Signup and view all the answers

What are the features of neighbors aggregated with fixed weights in GNN?

Convolutional GNN Signup and view all the answers

Which GNN layer is useful for homophilous graphs and highly scalable applications?

Convolutional GNN Signup and view all the answers

What are the attention weights computed as in Attentional GNN?

Attention weights are computed as 𝛼#) = 𝑎(𝐱 # , 𝐱) Signup and view all the answers

Which GNN layer is ideal for computational chemistry, reasoning, and simulation tasks?

Message-passing GNN Signup and view all the answers

What are the four steps outlined in the model design overview for building and training GNNs?

(1) Define a neighborhood aggregation function, (2) Define a loss function on the embeddings, (3) Train on a set of nodes, (4) Generate embeddings for nodes as needed Signup and view all the answers

What is the primary advantage of shared aggregation parameters for all nodes in GNN?

Sublinear number of model parameters and generalization to unseen nodes Signup and view all the answers

What is the purpose of generating embeddings 'on the fly' in GNNs?

To generate new embeddings for previously unseen nodes Signup and view all the answers

What does each node compute in the message-passing step of GNN?

Each node computes a message Signup and view all the answers

What are the three components in the message-passing process of GNN?

Message, Aggregation, Nonlinearity Signup and view all the answers

What are some examples of structured data mentioned in the text?

Knowledge graph, Protein graphs, Recommender Systems, User preference/consumption, Social Network Graphs Signup and view all the answers

What are some recent real-world applications of Graph Neural Networks (GNNs) mentioned in the text?

Drug discovery (including for COVID-19), Trip/wait time prediction on Google Maps, Fake news detection on Social Media (e.g. Tweets) Signup and view all the answers

What is the primary challenge mentioned in the text regarding structured data and deep learning techniques?

Applying deep learning techniques to graph-based information representations and obtaining embeddings of information contained in graphs for further use in deep learning pipelines. Signup and view all the answers

What are some foundational models of Graph Neural Networks (GNNs) mentioned in the text?

Graph Convolutional Networks, Graph Attentional Networks, Message Passing Neural Network (MPNN) Signup and view all the answers

What are some examples of tasks that can be handled effectively by sequence-to-sequence models according to the text?

Semantic Segmentation, Instance Segmentation Signup and view all the answers

What is the main objective of using grouped convolutions in ResNeXt according to the text?

To control the amount of concatenation between feature maps Signup and view all the answers

According to the text, what is the primary focus of Graph Neural Networks (GNNs)?

Capturing and processing information from graph-structured data Signup and view all the answers

What is the key purpose of using Darknet-53 in YOLOv3 as mentioned in the text?

To serve as the backbone for feature extraction Signup and view all the answers

According to the text, what is the primary function of a convolution operation on images?

To extract features from the input image Signup and view all the answers

What is the primary purpose of using normalization in Convolutional Neural Networks (CNNs) according to the text?

To improve the training speed and stability of the network Signup and view all the answers

What is the definition of permutation invariance for a function 𝑓(𝐗)?

𝑓(𝐗) is permutation invariant if, for all permutation matrices 𝐏, 𝑓 𝐏𝐗 = 𝑓 𝐗 Signup and view all the answers

How is the concept of locality enforced in equivariant set functions?

Locality in equivariant set functions is enforced through a shared function 𝜓 applied to every node in isolation. Signup and view all the answers

What is the formula for extracting neighbourhood features from a graph?

𝐗 𝒩. = {{ 𝐱) ∶ 𝑗 ∈ 𝒩# }} Signup and view all the answers

What is the key requirement for ensuring equivariance in the local function 𝜙 used in graph neural networks?

The local function 𝜙 should not depend on the order of the nodes in 𝐗 𝒩. Signup and view all the answers

What is the main difference in applying permutation invariance and equivariance on graphs compared to sets?

The main difference is that permutations now also accordingly act on the edges in addition to the nodes. Signup and view all the answers

How are permutation equivariant functions 𝐅(𝐗, 𝐀) constructed on graphs?

Permutation equivariant functions 𝐅(𝐗, 𝐀) are constructed by applying a local permutation-invariant 𝜙 over all neighbourhoods. Signup and view all the answers

What is the common lingo used to refer to the shared application of a local permutation-invariant function in graph neural networks?

The common lingo used is 'diffusion', 'propagation', or 'message passing'. Signup and view all the answers

What is the definition of a GNN layer in the context of graph neural networks?

A GNN layer is a permutation equivariant function 𝐅(𝐗, 𝐀) on graphs by shared application of a local permutation-invariant 𝜙. Signup and view all the answers

What is the broader context considered in graphs that gives rise to a node's neighbourhood?

The broader context in graphs that gives rise to a node's neighbourhood is the edges between nodes. Signup and view all the answers

What is the exercise posed in the text regarding ensuring equivariance in the local function 𝜙?

The exercise is to prove that it is sufficient if 𝜙 does not depend on the order of the nodes in 𝐗 𝒩. Signup and view all the answers

Flashcards

Mean Squared Error (MSE)

The loss function used in linear regression. It measures the average squared difference between predicted and actual values.

Logistic Sigmoid Function

A function that transforms the weighted input in logistic regression into a probability between 0 and 1.

Entropy (H)

A measure of the uncertainty associated with a random variable in logistic regression. It quantifies the randomness of the predicted probabilities.

Logistic Regression Model

A statistical model that predicts the probability of a binary output (e.g., success or failure) given an input.