Overview of Artificial Intelligence & Machine Learning

Overview of Artificial Intelligence

• The presentation introduces Artificial Intelligence (AI)
• The speaker is TS. Lương Văn Thiện, a faculty member at the Faculty of DS&AI, NEU.
• The website provided is www.tvluong.wordpress.com

Basic Concepts of Machine Learning and Deep Learning

• Core concepts include machine learning and deep learning.
• Neural network (DNN) and deep learning are discussed.
• Applications of machine learning (ML) and deep learning (DL) are presented.
• Essential tools and skills for ML and DL are highlighted as topics.

Image Classification as a Core Task in Computer Vision

• Image classification is a fundamental computer vision task.
• An example is given as classifying an image of a cat.

Beyond Image Classification

• The presentation dives deeper than basic image classification.
• Topics covered include semantic segmentation, object detection, and instance segmentation.

Three Key Factors Driving AI Advancements

• The presentation outlines three factors that are driving the advancement of AI:
  • Computation
  • Algorithms
  • Data

Traditional vs. AI-Driven Algorithms

• A comparison of traditional and AI-driven approaches to algorithms is provided.
• Traditional software development relies on pre-defined inputs and algorithms.
• AI-driven approaches, specifically machine learning, work with input/output data pairs to deduce the algorithm.

Example of Applying AI: Converting Celsius to Fahrenheit

• An illustrative example of a simple conversion task (Celsius to Fahrenheit) showcases the contrast between traditional and AI-based solutions.
• Traditional methods involve explicit calculations, while machine learning methods derive the conversion relationship from provided data.

Machine Learning/Deep Learning Algorithm

• The algorithm uses, input/output data.
• This is a core concept in deriving the formula for the conversion

Understanding Data Types

• The presentation highlights different data types:
  • Structured data (databases, data warehouses, ERP/CRM)
  • Semi-structured data (CSV, XML, JSON)
  • Unstructured data (Video, audio, image, document, sensor data)

Image Labeling Task

• The purpose of labeling is to annotate images.

Five Steps for AI Problem Solving

• The presentation outlines a five-step process for approaching AI problems:
  • Problem Setup
  • Data Collection
  • Model Building and Training
  • Model Selection
  • Model Deployment

Machine Learning Process

• Preprocessing, Learning, Evaluation, and Prediction are discussed as essential stages of the machine learning process.

Fundamental Concepts in Deep Learning

• Features, examples, labels, and models are defined as fundamental concepts.

Differentiating AI, Machine Learning, and Deep Learning

• The presentation clarifies the relationship between AI, machine learning, and deep learning.

Categorization of Machine Learning Algorithms

• The types of algorithms in machine learning are classified: supervised learning, unsupervised learning, reinforcement learning.
• Example tasks include: classification, regression, and clustering.

Classification vs. Regression

• The presentation differentiates between classification and regression:
• Classification: predicting a discrete value (e.g., category labels).
• Regression: predicting a continuous value (e.g., a numerical output).

Deep Learning Concepts: Neural Networks and Units

• Artificial Neural Networks (ANNs) and neurons/units, are explained.
• ANNs are interconnected groups of nodes inspired by the human brain.
• Neurons are basic building blocks of neural networks, performing calculations based on inputs.

Deep Learning Concept: Tensors

• Tensors are multi-dimensional arrays representing data.
• The presentation relates tensors to multi-dimensional arrays and cube-like structures of data.

Deep Learning Concepts: Epochs, Iterations, and Batches

• Epochs, iterations, and batches are described in relation to deep learning training.
• Epoch: complete cycle of training data.
• Iteration: number of batches required for one epoch.
• Batch: the size of a sample for the training data.

Deep Learning Concept: Cost Function/Loss Function

• The cost function/loss function evaluates the model's performance regarding the difference between predicted and actual values.
• Lower cost function indicates better model quality.

Deep Learning Concepts: Activation Functions

• Activation functions control whether a neuron/unit is activated or not.
• The presentation includes examples such as sigmoid, tanh, ReLU, and Leaky ReLU.

Summary of Key Loss Functions

• Mean Squared Error (MSE)
• Root Mean Squared Error (RMSE)
• Cross Entropy (CE)

Deep Learning Concepts: Forward and Backward Propagation

• Forward propagation passes input data through neural network layers to generate output.
• Backpropagation calculates gradients, updating weights and reducing the loss function.

Deep Learning Concepts: Hyperparameter Tuning and Cross-Validation

• Hyperparameters are defined as parameters set before training.
• The presentation explains the process of selecting optimal hyperparameters.
• Cross-validation is a technique used in assessing how well a model generalizes.

Deep Learning Concepts: Overfitting and Underfitting

• Overfitting occurs when a model excels on training data but performs poorly on unseen data (too complex).
• Underfitting occurs when a model is too simple and does not represent underlying data patterns well.

Additional Concepts in Deep Learning

• CNN (Convolutional Neural Networks)
• RNN (Recurrent Neural Networks)
• Batch normalization
• Autoencoder
• Generative Adversarial Network (GAN)
• Transfer learning
• Online learning
• Weight decay
• Dropout
• Data augmentation
• Generative AI.

AI Applications in Different Domains

• AI use-cases are explored, covering various areas:
  • Face and text recognition
  • Image and video
  • Natural language processing (NLP)
  • Product recommendations (recommendation systems)

Tools and Libraries for Building AI Applications

• The presentation introduces essential Python libraries and tools, including:
  • Jupyter Notebook, Spyder and Anaconda
• Libraries for data visualization (e.g., matplotlib, seaborn).

Deep Learning Frameworks

• Comparison of various Deep Learning Frameworks
• TensorFlow, PyTorch, Keras, Caffe, mxnet, theano, Chainer, and DL4J are discussed and compared in terms of capabilities and use-cases.

Top AI Companies

• Businesses, including Google
• Microsoft, Facebook, Amazon, IBM and others listed are highlighted for their significance and contributions in the field.

China vs. U.S. in AI

• The presentation explores global comparisons in AI research output - 2022.

Discussion Questions about AI

• The presentation concludes with discussion questions focusing on practical issues related to AI:
  • Concerns about job displacement
  • Necessary skills to excel in AI
  • Challenges and risks related to AI
  • Question on whether playing the lottery can be categorized as a data science activity.

This quiz covers essential concepts of Artificial Intelligence, including machine learning and deep learning. It discusses core topics such as neural networks and image classification, along with advanced areas like semantic segmentation and object detection. Join to test your understanding of these pivotal AI topics.