Understanding Deep Neural Networks Training Course

Part 1 – Concepts of Deep Learning and DNNs

Overview of AI, Machine Learning, and Deep Learning

• Historical context, fundamental concepts, and standard applications of artificial intelligence, moving beyond mere theoretical fantasies
• Collective Intelligence: the aggregation of knowledge shared among numerous virtual agents
• Genetic algorithms: evolving a population of virtual agents through selection processes
• Definition of a standard learning machine
• Task categories: supervised learning, unsupervised learning, and reinforcement learning
• Action types: classification, regression, clustering, density estimation, and dimensionality reduction
• Examples of machine learning algorithms: Linear regression, Naive Bayes, and Random Tree
• Machine Learning versus Deep Learning: identifying problems where machine learning remains the state of the art (e.g., Random Forests & XGBoosts)

Foundational Neural Network Concepts (Application: Multi-layer Perceptron)

• Review of mathematical foundations
• Definition of a neural network: classical architecture, activation functions, and the weighting of previous activations
• Network depth and its implications
• Defining network learning: cost functions, back-propagation, stochastic gradient descent, and maximum likelihood
• Neural network modeling: adapting input and output data modeling to problem types (regression, classification, etc.), including the curse of dimensionality
• Distinguishing between multi-feature data and signals; selecting appropriate cost functions based on data characteristics
• Function approximation via neural networks: theory and examples
• Distribution approximation via neural networks: theory and examples
• Data Augmentation: techniques for balancing datasets
• Generalizing neural network results
• Initialization and regularization of neural networks: L1/L2 regularization and Batch Normalization
• Optimization and convergence algorithms

Standard ML/DL Toolkits

A brief overview is provided, highlighting the advantages, disadvantages, ecosystem positioning, and usage of various tools.

• Data management: Apache Spark and Apache Hadoop tools
• Machine Learning libraries: Numpy, Scipy, and Scikit
• High-level DL frameworks: PyTorch, Keras, and Lasagne
• Low-level DL frameworks: Theano, Torch, Caffe, and TensorFlow

Convolutional Neural Networks (CNNs)

• Introduction to CNNs: core principles and applications
• Fundamental CNN operations: convolutional layers, kernel usage,
• Padding and stride, feature map generation, and pooling layers, including 1D, 2D, and 3D extensions
• Review of CNN architectures that have established the state of the art in classification
• Image processing examples: LeNet, VGG Networks, Network in Network, Inception, and ResNet; discussing innovations like 1x1 convolutions and residual connections
• Integration of attention models
• Application to common classification tasks (text or image)
• CNNs for generation: super-resolution and pixel-to-pixel segmentation
• Key strategies for enhancing feature maps in image generation

Recurrent Neural Networks (RNNs)

• Introduction to RNNs: core principles and applications
• Fundamental RNN operations: hidden activation, back-propagation through time, and unfolded representations
• Evolution towards Gated Recurrent Units (GRUs) and Long Short-Term Memory (LSTM) networks
• Analysis of architectural states and the advancements brought by these models
• Addressing convergence and vanishing gradient issues
• Classical architectures: time series prediction and classification
• RNN Encoder-Decoder architectures and the use of attention models
• NLP applications: word/character encoding and translation
• Video applications: predicting the next frame in a video sequence

Generative Models: Variational Autoencoders (VAE) and Generative Adversarial Networks (GAN)

• Overview of generative models and their connection to CNNs
• Autoencoders: dimensionality reduction and basic generation capabilities
• Variational Autoencoders: generative modeling and distribution approximation; definition and use of latent space, reparameterization trick, and associated limitations
• Generative Adversarial Networks: fundamental concepts
• Dual network architecture (Generator and Discriminator) with alternating training and cost functions
• GAN convergence challenges and common difficulties
• Improved convergence methods: Wasserstein GAN, Began, and Earth Mover's Distance
• Applications in image/photograph generation, text generation, and super-resolution

Deep Reinforcement Learning

• Introduction to reinforcement learning: agent control within a defined environment
• State representation and possible actions
• Utilizing neural networks to approximate state functions
• Deep Q-Learning: experience replay and application to video game control
• Policy optimization: on-policy vs off-policy approaches, Actor-Critic architecture, and A3C
• Applications in controlling video games or digital systems

Part 2 – Deep Learning with Theano

Foundations of Theano

• Introduction
• Installation and configuration

Theano Functions

• Managing inputs, outputs, updates, and givens

Training and Optimizing Neural Networks with Theano

• Modeling neural networks
• Implementing logistic regression
• Incorporating hidden layers
• Network training procedures
• Computation and classification tasks
• Optimization techniques
• Log Loss calculation

Model Testing

Part 3 – DNNs using TensorFlow

TensorFundamentals

• Creating, initializing, saving, and restoring TensorFlow variables
• Feeding, reading, and preloading data in TensorFlow
• Leveraging TensorFlow infrastructure for large-scale model training
• Visualizing and evaluating models using TensorBoard

TensorFlow Mechanics

• Data preparation
• Downloading data
• Managing inputs and placeholders
• Constructing Graphs
  • Inference
  • Loss calculation
  • Training operations
• Model Training
  • Graph management
  • Session handling
  • Training loops
• Model Evaluation
  • Building evaluation graphs
  • Evaluating outputs

The Perceptron

• Activation functions
• Perceptron learning algorithm
• Binary classification using perceptrons
• Document classification with perceptrons
• Limitations of the perceptron model

From Perceptrons to Support Vector Machines

• Kernels and the kernel trick
• Maximum margin classification and support vectors

Artificial Neural Networks

• Nonlinear decision boundaries
• Feedforward and feedback neural networks
• Multilayer perceptrons
• Minimizing cost functions
• Forward propagation
• Back propagation
• Enhancing neural network learning strategies

Convolutional Neural Networks

• Objectives
• Model architecture
• Core principles
• Code organization
• Launching and training models
• Model evaluation

Brief introductions to the following modules will be provided, subject to time availability:

Advanced TensorFlow Usage

• Threading and queues
• Distributed TensorFlow
• Documentation and model sharing
• Customizing data readers
• Manipulating TensorFlow model files

TensorFlow Serving

• Introduction
• Basic serving tutorial
• Advanced serving tutorial
• Serving Inception models tutorial