Understanding Deep Neural Networks Training Course

Part 1 – Deep Learning and DNN Concepts

Introduction to AI, Machine Learning & Deep Learning

• An overview of the history, fundamental concepts, and common applications of artificial intelligence, distinguishing reality from fiction
• Collective Intelligence: aggregating knowledge shared across multiple virtual agents
• Genetic algorithms: evolving populations of virtual agents through selection processes
• Machine Learning basics: definitions and core principles
• Task types: supervised learning, unsupervised learning, and reinforcement learning
• Action types: classification, regression, clustering, density estimation, and dimensionality reduction
• Examples of machine learning algorithms, including linear regression, Naive Bayes, and Random Trees
• Machine Learning vs. Deep Learning: identifying problems where traditional ML (e.g., Random Forests & XGBoost) remains the state of the art

Neural Network Fundamentals (Application: Multi-layer Perceptron)

• Review of essential mathematical foundations
• Defining neural networks: classical architecture, activation functions
• Weighting of previous activations and network depth
• Learning in neural networks: cost functions, back-propagation, stochastic gradient descent, and maximum likelihood
• Modelling neural networks: adapting input and output data to problem types (regression, classification, etc.) and addressing the curse of dimensionality
• Distinguishing between multi-feature data and signals; selecting appropriate cost functions based on data characteristics
• Function approximation by neural networks: theory and practical examples
• Distribution approximation by neural networks: theory and practical examples
• Data Augmentation: techniques for balancing datasets
• Generalising neural network results
• Initialisation and regularisation of neural networks, including L1/L2 regularisation and Batch Normalisation
• Optimisation and convergence algorithms

Standard ML/DL Tools

A brief overview covering the advantages, disadvantages, ecosystem positioning, and usage of key tools.

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

Convolutional Neural Networks (CNNs)

• Overview of CNNs: core principles and applications
• Basic CNN operations: convolutional layers and kernel usage
• Padding, stride, feature map generation, pooling layers, and 1D/2D/3D extensions
• Presentation of prominent CNN architectures that have defined state-of-the-art classification performance
• Key image architectures: LeNet, VGG Networks, Network in Network, Inception, and ResNet, including their innovations and broader applications (e.g., 1x1 convolution or residual connections)
• Implementing attention models
• Applying CNNs to standard classification tasks (text or image)
• CNNs for generation: super-resolution and pixel-to-pixel segmentation
• Strategies for enhancing feature maps in image generation

Recurrent Neural Networks (RNNs)

• Overview of RNNs: core principles and applications
• Basic RNN operations: hidden activations, back-propagation through time, and unfolded versions
• Evolution towards Gated Recurrent Units (GRUs) and Long Short-Term Memory (LSTM) networks
• Presentation of different states and improvements brought by these architectures
• Addressing convergence and vanishing gradient problems
• Classic architectures: time series prediction, classification, and more
• RNN Encoder-Decoder architectures and the use of attention models
• NLP applications: word/character encoding and machine translation
• Video applications: predicting the next frame in a video sequence

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

• Introduction to generative models and their relationship with CNNs
• Autoencoders: dimensionality reduction and limited generation capabilities
• Variational Autoencoders: generative modelling, distribution approximation, latent space definition, reparameterisation trick, applications, and limitations
• Generative Adversarial Networks: foundational concepts
• Dual network architecture (Generator and Discriminator), alternating learning strategies, and available cost functions
• GAN convergence and associated challenges
• Improved convergence techniques: Wasserstein GAN, Began, and Earth Moving Distance
• Applications in image/photograph generation, text generation, and super-resolution

Deep Reinforcement Learning

• Introduction to reinforcement learning: controlling agents within defined environments
• Understanding states and possible actions
• Using neural networks to approximate state functions
• Deep Q Learning: experience replay and applications in video game control
• Policy optimisation: on-policy vs. off-policy approaches, Actor-Critic architecture, and A3C
• Applications: controlling single video games or digital systems

Part 2 – Theano for Deep Learning

Theano Fundamentals

• Introduction
• Installation and Configuration

Theano Functions

• Inputs, outputs, updates, and givens

Training and Optimising Neural Networks with Theano

• Neural Network Modelling
• Logistic Regression
• Hidden Layers
• Training the Network
• Computing and Classification
• Optimisation
• Log Loss

Model Testing

Part 3 – DNNs using TensorFlow

TensorFlow Fundamentals

• Creating, initialising, saving, and restoring TensorFlow variables
• Feeding, reading, and preloading TensorFlow data
• Leveraging TensorFlow infrastructure to train models at scale
• Visualising and evaluating models using TensorBoard

TensorFlow Mechanics

• Data Preparation
• Downloading
• Inputs and Placeholders
• Building Graphs
  • Inference
  • Loss
  • Training
• Training the Model
  • The Graph
  • The Session
  • Training Loop
• Evaluating the Model
  • Building the Evaluation Graph
  • Evaluation Output

The Perceptron

• Activation functions
• The perceptron learning algorithm
• Binary classification using the perceptron
• Document classification using the perceptron
• Limitations of the perceptron

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 artificial neural networks
• Multilayer perceptrons
• Minimising the cost function
• Forward propagation
• Back propagation
• Techniques for improving neural network learning

Convolutional Neural Networks

• Objectives
• Model Architecture
• Principles
• Code Organisation
• Launching and Training the Model
• Evaluating a Model

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

TensorFlow – Advanced Usage

• Threading and Queues
• Distributed TensorFlow
• Writing Documentation and Sharing Your Model
• Customising Data Readers
• Manipulating TensorFlow Model Files

TensorFlow Serving

• Introduction
• Basic Serving Tutorial
• Advanced Serving Tutorial
• Serving Inception Model Tutorial