Understanding Deep Neural Networks Training Course

Part 1 – Deep Learning and DNN Concepts

Introduction to AI, Machine Learning & Deep Learning

• Historical context, basic concepts, and common applications of artificial intelligence, distinct from the often romanticized fantasies of the field
• Collective Intelligence: aggregating knowledge shared among numerous virtual agents
• Genetic algorithms: evolving populations of virtual agents through selection processes
• Standard Learning Machines: definition and overview
• Task types: 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 Trees
• Machine Learning vs. Deep Learning: identifying problems where Machine Learning (e.g., Random Forests & XGBoost) remains the state of the art

Basic Concepts of Neural Networks (Application: multi-layer perceptron)

• Review of mathematical foundations.
• Definition of a neural network: classical architecture, activation functions, and
• Weighting of previous activations and network depth
• Defining the learning process for neural networks: cost functions, back-propagation, stochastic gradient descent, and maximum likelihood.
• Modeling neural networks: structuring input and output data based on the problem type (regression, classification, etc.). The curse of dimensionality.
• Distinguishing between multi-feature data and signals. Selecting appropriate cost functions based on data characteristics.
• Function approximation via neural networks: theoretical overview and practical examples
• Distribution approximation via neural networks: theoretical overview and practical examples
• Data Augmentation: techniques for balancing datasets
• Generalizing results from neural networks.
• Initializing and regularizing neural networks: L1 / L2 regularization, and Batch Normalization
• Optimization and convergence algorithms

Standard ML / DL Tools

A concise overview will be provided, highlighting the advantages, disadvantages, ecosystem positioning, and usage of each tool.

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

Convolutional Neural Networks (CNN).

• Overview of CNNs: fundamental principles and applications
• Core operations of a CNN: convolutional layers, kernel usage,
• Padding & stride, feature map generation, and pooling layers. Extensions to 1D, 2D, and 3D.
• Overview of various CNN architectures that have set the state of the art in classification
• Image processing: LeNet, VGG Networks, Network in Network, Inception, Resnet. Discussing innovations introduced by each architecture and their broader applications (e.g., 1x1 convolution or residual connections)
• Implementing attention models.
• Application to standard classification tasks (text or image)
• CNNs for generation: super-resolution and pixel-to-pixel segmentation. Overview of
• Key strategies for enhancing feature maps in image generation.

Recurrent Neural Networks (RNN).

• Overview of RNNs: fundamental principles and applications.
• Core RNN operations: hidden activations, back propagation through time, and unfolded versions.
• Evolutions towards Gated Recurrent Units (GRUs) and LSTM (Long Short-Term Memory).
• Overview of the different states and advancements brought by these architectures
• Convergence issues and the vanishing gradient problem
• Classical architectures: Time series prediction, classification, etc.
• RNN Encoder-Decoder type architecture. Use of attention models.
• NLP applications: word / character encoding and translation.
• Video Applications: predicting the next frame in a video sequence.

Generative Models: Variational AutoEncoder (VAE) and Generative Adversarial Networks (GAN).

• Overview of generative models and their connection to CNNs
• Auto-encoders: dimensionality reduction and limited generation capabilities
• Variational Auto-encoders: generative models approximating data distributions. Definition and use of latent space, reparameterization trick, observed applications, and limitations
• Generative Adversarial Networks: Fundamentals.
• Dual Network Architecture (Generator and Discriminator) with alternate learning and available cost functions.
• GAN convergence and common difficulties.
• Improved convergence methods: Wasserstein GAN, Began. Earth Mover’s Distance.
• Applications: image/photograph generation, text generation, and super-resolution.

Deep Reinforcement Learning.

• Overview of reinforcement learning: controlling an agent within a defined environment
• Characterized by states and possible actions
• Utilizing neural networks to approximate state-value functions
• Deep Q Learning: experience replay, and application to video game control.
• Policy learning optimization. On-policy & off-policy methods. Actor-Critic architecture. A3C.
• Applications: controlling video games or digital systems.

Part 2 – Theano for Deep Learning

Theano Basics

• Introduction
• Installation and Configuration

Theano Functions

• inputs, outputs, updates, givens

Training and Optimization of a Neural Network using Theano

• Neural Network Modeling
• Logistic Regression
• Hidden Layers
• Training a network
• Computing and Classification
• Optimization
• Log Loss

Testing the Model

Part 3 – DNN using Tensorflow

TensorFlow Basics

• Creating, initializing, saving, and restoring TensorFlow variables
• Feeding, reading, and preloading TensorFlow Data
• Leveraging TensorFlow infrastructure to train models at scale
• Visualizing and evaluating models with TensorBoard

TensorFlow Mechanics

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

The Perceptron

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

From the Perceptron 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
• Minimizing the cost function
• Forward propagation
• Back propagation
• Improving neural network learning methods

Convolutional Neural Networks

• Goals
• Model Architecture
• Principles
• Code Organization
• Launching and Training the Model
• Evaluating a Model

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

TensorFlow - Advanced Usage

• Threading and Queues
• Distributed TensorFlow
• Writing Documentation and Sharing your Model
• Customizing Data Readers
• Manipulating TensorFlow Model Files

TensorFlow Serving

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