Session: Deep Learning With Keras


This tutorial is designed to introduce software engineers, developers and data scientists with the tools and knowledge needed in order to use deep learning with their everyday work. The main concepts and ideas of deep learning will be presented. Deep learning is one of the leading tools in data analysis this days and one of the two common frameworks for deep learning are Theano and Keras. The Tutorial will provide an introduction to these libraries with practical code examples.

Note: An Expert level of python will be assumed.

Note: Every "Hands-On" Includes Explanations, Code Samples and exercises and all "Hands-On"s jupyter-notebooks will be shared before the tutorial. 1. Introduction About Deep Learning, time-line, theory, uses, examples Brief Into to Neural Networks: - From Threshold logic and Turing machines to ANNs - Warren Sturgis McCulloch's famous article - Perceptron and Gradient decent. - Multilayer Perceptron network and Backpropagation. – Definitions: Dataset, Learning set, Training set, Training vectors, Epoch, Bias, Learning rate, Momentum.

  1. Theano Tutorial Motivation for theano, theano syntax, Function and variables, using theano for gradient calculations, high performance computing with theano on the gpu: live demo and code tutorial (won’t be a hands on because not all the audice laptops may be supporting.) Hands-on: Implementing Regression with Theano. Hands-on: Implementing Logistic Regression with Theano. Hands-on: implementing one layer neural net architecture with theano

  2. Complex ANN structures:

  3. Deep learning Feed-Forward Multilayer-Perceptron Neural Networks
  4. Implementing fully connected network for MNIST Hands-on: Implementing Multilayer Perceptron with Theano.

  5. Convolutional Neural Networks: LeCun, Yann; Léon Bottou; Yoshua Bengio; Patrick Haffner (1998). "Gradient-based learning applied to document recognition" Motivation, the power of convolution, uses today, optimization.

  6. Keras Tutorial: Motivation for Keras, Keras syntax. Hand-on: implement a deep architecture with keras on MNIST.

  7. Practical ConvNets: Image Recognition, Fine tuning pretraied networks, Style transfer Hands-On: Convolutional neural networks Keras. Hands-On: Fine tuning convolutional neural networks Keras.

  8. Recurrent Neural Networks: - Architecture - Training - “Simply, how does it work?” Hands-On: Implementing RNN Network with Keras Dataset: bAbI tasks

Hands-On: Implementing LSTM Network with Keras – Dataset: Nietzsche's writings Hands-On: Implementing Image caption generator CNN and LSTM with Keras

  1. Extending Keras and some more Fun with Deep Learning :) Deep Q Learning Demo We want you! To contribute to Keras!
    Keras internals in a nutshell and how to find your way in the code. Invitation: sprint on keras :)