Artificial Intelligence Training

5 Day Training Course: Neural Networks and Deep Learning Enroll in a public course Enroll in a public course Request a private course

5 Day Training Course: Neural Networks and Deep Learning

This five-day hands-on course is designed for data scientists seeking a better understanding and knowledge of main technology trends driving Deep Learning.

Attendees will get a clear understanding of Deep Learning technology, practical scenarios to build, train and apply algorithms of fully connected Deep Neural Networks, strategies to configure the key parameters in a neural network's architecture.

The participants will gain experience in building and applying Deep Neural Networks, with most popular frameworks, such as Keras, TensorFlow, Theano, scikit-learn.

After the course, you will become fully skilled of implementing deep learning for your own applications.

Day 1

Topic	Description
Introduction to Deep Learning	Agenda for the training. General introduction for neural networks and deep learning.
Neural Networks	Components of an artificial neural network. Connections and weights. Propagation function. Choosing a cost function. Learning paradigms. Types of neural networks.
Activation Functions	Identity function, unit step (binary step) function, sigmoid function, hyperbolic function, inverse trigonometric function, softmax function, rectified linear unit (ReLU), exponential linear unit (ELU), maxout.
Learning of Neural Networks	Training, test, and validation sets. Selection of a validation dataset: holdout method and cross-validation. Instance space decomposition.
Supervised Learning	Labeled training data. Determination of the type of training examples. Gathering a training set. Determination of the input feature representation of the learned function. Running the learning algorithm on the gathered training set. Evaluation of the accuracy of the learned function.
Unsupervised Learning	Approaches to unsupervised learning. Clustering, anomaly detection, Autoencoders, Generative Adversarial Networks, self-organizing map.
Reinforcement Learning	Markov decision process. Algorithms for control learning. Optimality criteria. Brute force approach. Value function approaches. Monte Carlo methods. Temporal difference methods. End-to-end reinforcement learning.

Day 2

Topic	Description
Perceptron	Initialization of the weights and the threshold. Training of the algorithm. Calculation of the actual output. Multiclass perceptron.
Multilayer Perceptron	Activation function. Layers of nonlinearly-activating nodes. Learning of the perceptron.
Recurrent Neural Networks	Directed graph along a sequence. Infinite and finite impulses. Long short-term memory networks. Fully Recurrent Neural Network. Independently Recurrent Neural Network. Recursive Neural Network. Hopfield network. Boltzmann machine. Restricted Boltzmann machine. Recurrent Multilayer Perceptrons network. Multiple Timescales Recurrent Neural Network.
Autoencoders	Dimensionality reduction. Denoising Autoencoder. Sparse Autoencoder. Variational Autoencoder models. Contractive Autoencoder. Training algorithm for an Autoencoder.
Generative Adversarial Networks	Generative and discriminative networks. Error rate of the discriminative network. Fully convolutional feedforward Generative Adversarial Network.
Convolutional Neural Networks	Convolutional layers, pooling layers, fully connected layers and normalization layers. Number of filters. Filter shape. Hierarchical coordinate frames.

Day 3

Topic	Description
Deep Learning in Computer Vision	Machines vision. Applications of computer vision techniques. Image acquisition. Pre-processing images. Feature extraction. Detection and segmentation. High-level processing. Decision making.
Image Classification	Image processing. Supervised and unsupervised classification.
Object Detection	Sliding windows. Viola-Jones detector. Region-based Convolutional Neural Networks. Single shot detector models.
Object Tracking and Action Recognition	Computer vision architectures for video analysis. Optical flow estimation. Visual object tracking. Action recognition.
Image Segmentation and Synthesis	Semantic image segmentation. Image synthesis problems.

Day 4

Topic	Description
Deep Learning in Natural Language Processing	Challenges in natural-language processing. Statistical natural-language processing.
Text Classification	Linear classifiers and deep learning techniques. News flows classification. Sentiment analysis. Spam filtering.
Language Modeling and Sequence Tagging	Prediction of next words techniques. Methods to predict a sequence of tags for a sequence of words.
Machine Translation	Word sense disambiguation. Rule-based machine translation (MT). Transfer-based MT. Interlingual MT. Dictionary-based MT. Statistical MT. Example-based MT. Hybrid MT. Neural MT.
Dialog Systems	Architectures for dialog systems. Input recognizer. Natural language understanding. Dialog manager. Task manager. Output generator. Output renderer. Types of systems and current frameworks.

Day 5

Topic	Description
Meta-optimization	Meta-learning approach to neural network optimization. Optimization strategy. Optimization method benchmarks. Combination of optimization methods. Evaluation of methods.
Neural Network Ensemble	Basic principles of Neural Network Ensemble. Individual neural network generation methods. Conclusion generation methods.

The Neural Networks and Deep Learning training was developed for data scientists seeking a better understanding and knowledge of:

Main technology trends driving Deep Learning
How to build, train and apply fully connected deep Neural Networks
Key parameters in a neural network architecture optimization

Provide the information and, through labs, the experience necessary for the students to:

Gain a deep understanding of Deep Learning concept
Use main different types of Neural Networks
Build Deep Neural Networks applications

Altoros recommends that all students have:

Programming: Basic R and Python programming skills, with the capability to work effectively with data structures
Experience with the RStudio and the Jupyter Notebook applications
Basic experience with git
A basic understanding matrix vector operations and notation
A basic knowledge of Statistics
Basic command line operations

A workstation with the following capabilities:

A web browser (Chrome/Firefox)
Internet connection
A firewall allowing outgoing connections on TCP ports 80 and 443

The following developer utilities should be installed:

Anaconda
RStudio
Jupyter Notebook

Attendees will get a deep understanding of the essential concepts of Deep Learning and acquire a clear knowledge of how to apply Deep Learning algorithms to computer vision, text understanding, medical informatics and other areas.

PRIVATE TRAINING FOR YOUR TEAM

Request a quote