Understanding Deep Neural Networks Training Course

Part 1 – Deep Learning and DNN Concepts

Introduction to AI, Machine Learning & Deep Learning

• History, basic concepts, and common applications of artificial intelligence, separating reality from the fantasies often associated with this domain.
• Collective Intelligence: aggregating knowledge shared by multiple virtual agents.
• Genetic algorithms: evolving a population of virtual agents through selection.
• Standard Learning Machines: definition.
• Types of tasks: supervised learning, unsupervised learning, and reinforcement learning.
• Types of actions: 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: problems where Machine Learning remains the state-of-the-art (e.g., Random Forests & XGBoosts).

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

• Review of mathematical foundations.
• Definition of a network of neurons: classical architecture, activation functions, and...
• Weighting of previous activations and network depth.
• Definition of neural network learning: cost functions, back-propagation, Stochastic Gradient Descent, and maximum likelihood.
• Modeling a neural network: modeling input and output data based on the type of problem (regression, classification, etc.) and addressing the curse of dimensionality.
• Distinguishing between multi-feature data and signals. Choosing a cost function based on the data.
• Function approximation by a neural network: presentation and examples.
• Distribution approximation by a neural network: presentation and examples.
• Data Augmentation: techniques for balancing a dataset.
• Generalization of results from a neural network.
• Initialization and regularization of a neural network: L1 / L2 regularization, Batch Normalization.
• Optimization and convergence algorithms.

Standard ML / DL Tools

A simple presentation outlining advantages, disadvantages, ecosystem positioning, and use cases will be provided.

• 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).

• Presentation of CNNs: fundamental principles and applications.
• Basic operation of a CNN: convolutional layer, use of a kernel, ...
• Padding & stride, feature map generation, pooling layers, and 1D, 2D, and 3D extensions.
• Presentation of different CNN architectures that achieved state-of-the-art results in classification.
• Images: LeNet, VGG Networks, Network in Network, Inception, Resnet. Presentation of innovations brought by each architecture and their broader applications (e.g., Convolution 1x1 or residual connections).
• Use of attention models.
• Application to common classification cases (text or image).
• CNNs for generation: super-resolution, pixel-to-pixel segmentation. Presentation of...
• Main strategies for increasing feature maps for image generation.

Recurrent Neural Networks (RNN).

• Presentation of RNNs: fundamental principles and applications.
• Basic operation of the RNN: hidden activation, backpropagation through time, and the unfolded version.
• Evolutions towards Gated Recurrent Units (GRUs) and LSTM (Long Short-Term Memory).
• Presentation of different states and the evolutions brought by these architectures.
• Convergence and vanishing gradient problems.
• Classical architectures: temporal series prediction, classification, etc.
• RNN Encoder-Decoder type architecture. Use of an attention model.
• NLP applications: word / character encoding, translation.
• Video Applications: prediction of the next generated image in a video sequence.

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

• Presentation of generative models and their link with CNNs.
• Auto-encoder: dimensionality reduction and limited generation.
• Variational Auto-encoder: generative model and approximation of a given distribution. Definition and use of latent space. Reparameterization trick. Applications and observed limitations.
• Generative Adversarial Networks: Fundamentals.
• Dual Network Architecture (Generator and Discriminator) with alternating learning and available cost functions.
• Convergence of a GAN and difficulties encountered.
• Improved convergence: Wasserstein GAN, BegAN, Earth Mover's Distance.
• Applications for the generation of images or photographs, text generation, and super-resolution.

Deep Reinforcement Learning.

• Presentation of reinforcement learning: controlling an agent in a defined environment.
• Through a state and possible actions.
• Use of a neural network to approximate the state function.
• Deep Q Learning: experience replay and application to video game control.
• Optimization of learning policy. On-policy && off-policy. Actor-critic architecture. A3C.
• Applications: control of a single video game or a digital system.

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

• Creation, Initializing, Saving, and Restoring TensorFlow variables
• Feeding, Reading and Preloading TensorFlow Data
• How to use TensorFlow infrastructure to train models at scale
• Visualizing and Evaluating models with TensorBoard

TensorFlow Mechanics

• Prepare the Data
• Download
• Inputs and Placeholders
• Build the Graphs
  • Inference
  • Loss
  • Training
• Train the Model
  • The Graph
  • The Session
  • Train Loop
• Evaluate 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 the way neural networks learn

Convolutional Neural Networks

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

Basic Introductions to be given to the below modules (Brief Introduction to be provided based on 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