Machine Learning Fundamentals Quiz

<p>Regression models, decision trees, Random Forests, and neural networks. (D)</p> Signup and view all the answers

<p>Potential absence of causal insights, which are important for policy implications. (D)</p> Signup and view all the answers

<p>The process through which a model learns a function that maps input to output. (B)</p> Signup and view all the answers

<p>It requires input data (X) and its corresponding labels (Y). (D)</p> Signup and view all the answers

<p>Reinforcement learning. (C)</p> Signup and view all the answers

<p>To minimize the risk of overfitting to a specific dataset configuration. (C)</p> Signup and view all the answers

<p>It combines predictions from several models to create a single, more accurate prediction. (B)</p> Signup and view all the answers

<p>Bootstrap sampling, generating subsets by randomly sampling with replacement. (C)</p> Signup and view all the answers

<p>Number of trees (<code>n_estimators</code>). (D)</p> Signup and view all the answers

<p>Random Forests use multiple decision trees in parallel, while Boosting trains models sequentially. (A)</p> Signup and view all the answers

<p>A model that performs very well on training data but poorly on new unseen data. (D)</p> Signup and view all the answers

<p>Random forests are generally more difficult to interpret. (C)</p> Signup and view all the answers

<p>Errors introduced by the assumptions made by the learning algorithm. (B)</p> Signup and view all the answers

<p>To select a random subset of features used for each split in the tree. (A)</p> Signup and view all the answers

<p>A technique that stops training when the model’s performance on a validation set starts to degrade, thus preventing overfitting. (C)</p> Signup and view all the answers

<p>Implementing the model in a production environment. (C)</p> Signup and view all the answers

<p>Statistical approaches are built on more assumptions and give more interpretable parameters, while machine learning methods tend to have fewer assumptions. (A)</p> Signup and view all the answers

<p>Equal-Area Projection (B)</p> Signup and view all the answers

<p>They are easy to understand, use, and interpret, making them useful for feature selection. (D)</p> Signup and view all the answers

<p>They are prone to overfitting and can yield unstable results with slight variations in the data. (D)</p> Signup and view all the answers

<p>Mean Absolute Error (MAE) (C)</p> Signup and view all the answers

<p>To distribute feature contributions based on their marginal impact in various combinations. (B)</p> Signup and view all the answers

<p>Complexity: Higher feature impact leads to lower scores. (D)</p> Signup and view all the answers

<p>It provides global consistency in its explanations. (D)</p> Signup and view all the answers

<p>Waterfall plot. (C)</p> Signup and view all the answers

<p>How much the predictions change when a feature's values are shuffled randomly. (A)</p> Signup and view all the answers

<p>By measuring how variations in input affect the output. (C)</p> Signup and view all the answers

<p>How temperature and seasons affect the number of bicycle rentals. (D)</p> Signup and view all the answers

<p>To increase understandability, identify biases, and provide consistent explanations. (D)</p> Signup and view all the answers

<p>By sequentially training trees on the residuals of previous trees. (C)</p> Signup and view all the answers

<p>max_features (C)</p> Signup and view all the answers

<p>Embeddings transform data into a continuous vector space, making it suitable for algorithms. (C)</p> Signup and view all the answers

<p>Using the bottleneck of an autoencoder. (D)</p> Signup and view all the answers

<p>Ensembles have increased training complexity and are more difficult to interpret. (B)</p> Signup and view all the answers

<p>It refers to maintaining relationships between the original data in the embedded space. (A)</p> Signup and view all the answers

<p>Association, temporary order, and no false connections. (C)</p> Signup and view all the answers

<p>Boosting is more focused on correcting errors and is effective with complex data. (D)</p> Signup and view all the answers

<p>The risk of privacy violation and reinforcement of stereotypes. (C)</p> Signup and view all the answers

<p>The risk of unintentional leaks of private information. (B)</p> Signup and view all the answers

<p>Through AI-driven supply and demand balancing. (B)</p> Signup and view all the answers

<p>Simulating the effects of emission reduction strategies. (C)</p> Signup and view all the answers

<p>Contextual and domain-specific requirements. (D)</p> Signup and view all the answers

<p>The tension between maximizing model performance and explanation clarity. (C)</p> Signup and view all the answers

<p>To enhance the understanding of how black-box models work. (C)</p> Signup and view all the answers

<p>The better integration of XAI methodologies into socio- technical systems. (C)</p> Signup and view all the answers

The field of study that allows computers to learn from data without explicit instructions.

Machine learning models aim to make accurate predictions by learning the relationship between input and output data.

Statistical models aim to understand the underlying relationships and reasons behind observed data.

Machine learning models often have a large number of parameters, making them less interpretable compared to statistical models.

Supervised learning involves training a model with labelled data, where both input (X) and output (Y) are provided.

Unsupervised learning involves analyzing unlabeled data (X) to find patterns and structures within the data.

Reinforcement learning trains a model through interactions and rewards or punishments based on decisions made.

Machine learning models can be used for various tasks, including email spam filtering, chatbots, fraud detection, recommendation systems, and advertisement placement.

A model's tendency to fit the training data too closely, leading to poor performance on new, unseen data. It captures noise and random fluctuations present in the training set, hindering generalization to new examples.

A model that is too simple and fails to capture important patterns in the data, resulting in poor predictive accuracy on both training and new data.

The balance between bias (errors due to assumptions made by the model) and variance (sensitivity to data variations). A model with high bias makes strong assumptions and might miss important details, while a model with high variance is prone to overfitting.

A technique used to evaluate the performance of a machine learning model by splitting the dataset into two parts: training data used to train the model and test data used to assess how well it generalizes to unseen data.

A process for developing a machine learning model that involves 5 steps: 1. Studying the phenomenon and cleaning the data, 2. Discovering relevant data features, 3. Exploring relationships through visualizations and correlations, 4. Training a basic model, and 5. Evaluating the model's performance using appropriate metrics.

Models designed to predict continuous values, such as income, age, temperature, or house prices. Examples include linear regression and multiple regression.

A type of regression model that uses a single explanatory variable to predict a continuous target variable.

A type of regression model that uses multiple explanatory variables to predict a continuous target variable.

A boosting technique where decision trees are trained sequentially on residuals, aiming to correct errors from previous trees.

A family of algorithms that combine multiple weak learners (often decision trees) to create a stronger predictive model.

The amount of influence each new tree has on the overall model prediction, controlling the learning speed and preventing overfitting.

Representation of discrete data in a continuous, lower-dimensional vector space, effectively capturing relationships between data points.

A relationship where a change in one variable (cause) directly leads to a change in another variable (effect), keeping all other factors constant.

The process of training a model with labeled data where both inputs (X) and outputs (Y) are provided.

The process of analyzing unlabeled data to find patterns and structures without explicit output guidance.

Training a model through interactions with an environment, where the model receives rewards or punishments based on its decisions.

A method used to explain how each feature contributes to a machine learning model's prediction. Imagine it as dividing the model's output among the input features to quantify their contribution.

AI systems developed with ethical considerations, focusing on fairness, transparency, accountability, and responsible use.

A property of SHAP Values that ensures the sum of all feature contributions equals the model's final output.

The potential for AI to perpetuate existing societal biases and discrimination in its data and algorithms.

A property of SHAP Values that states features with equal contributions receive the same score, regardless of their order.

Collecting and analyzing data while respecting individual privacy and maintaining data security.

Techniques that help explain the reasoning behind AI models' decisions, making them more transparent and understandable.

A method for explaining the importance of features in a model by randomly shuffling each feature's values and measuring the impact on predictions.

A method for visually explaining how features affect model decisions.

Using AI to address climate change challenges, such as optimizing energy use and predicting climate impacts.

Challenges involved in ensuring the ethical and responsible implementation of AI systems.

A method for visually explaining the impact of a feature on model predictions.

A technique used to explain the predictions of Convolutional Neural Networks (CNNs) by highlighting the parts of the input image that were most influential in the prediction.

Techniques that improve the accuracy and fairness of AI models by incorporating human feedback and insights.

A type of explainable AI method that focuses on providing explanations after a model has been trained. It explores the reasons behind the model's decision by analyzing the features and their contributions.

Using AI for policy analysis and decision-making, such as simulating climate scenarios and informing policy interventions.

A technique for minimizing the risk of overfitting a model to a specific dataset. It involves training and evaluating the model on multiple combinations of data folds, averaging the performance to optimize hyperparameters.

A model that combines multiple ‘weak’ models to create a stronger, more robust model. This technique draws inspiration from the "wisdom of the crowd" principle, where diversity reduces bias and improves generalization.

A type of ensemble model that uses multiple decision trees to make predictions. It addresses the problem of overfitting inherent in decision trees by introducing diversity and stability. These trees are trained on different bootstrap samples of the data, using a random subset of features for splitting at each node.

A technique used to build random forests, where multiple bootstrap datasets are created by randomly sampling data with replacement. This allows the model to be trained on different subsets of data, reducing overfitting.

A technique used in random forests, where a randomly selected subset of features is considered at each split in a decision tree. This helps to improve diversity and prevent the model from relying too heavily on any specific features.

A hyperparameter in random forests that controls the number of decision trees in the ensemble. Increasing this number generally improves performance up to a certain point.

A hyperparameter in random forests that controls the maximum number of features considered during each split in a decision tree.