Recently deep learning is everywhere and, along with it, TensorFlow, Google’s deep learning library. Learning TensorFlow can be quite challenging. Hence, in this tutorial we will give a deep insight into TensorFlow by introducing its most important features from scratch. Starting with a couple of simple lines of code, we will arrive at the implementation of a full neural network. Every line of code will be discussed and live executed. The audience will have full access to the demos, in order to be able to gain direct experience with TensorFlow.
Maximum number of participants: 25
Participants should bring along their own laptop: preconfigured with Python and TensorFlow.
Target Audience: Data Scientists, Big Data Developers, Developers
Prerequisites: Basic knowledge in Python, basic knowledge in Deep Learning
TensorFlow has become the most popular deep learning library, thus in this tutorial participants will get a comprehensive introduction to TensorFlow, and at the end of it they will be able to use this tool independently. This tutorial is meant to be highly interactive in order to clarify any question which might arise.
TensorFlow has been developed by the researchers and engineers of the Google Brain Team and it is now supported by the Google community. Moreover, this library offers APIs in several programming languages, e.g. Python and Java, it operates easily with multiple GPUs and it offers the possibility to visualize the network topology and its performances using TensorBoard. All these elements combined have made TensorFlow the most popular deep learning library.
We will start from a couple of simple lines of code, and will arrive at the implementation of a full neural network (NN). Every line of code will be discussed and executed live. The audience will have full access to the demos, in order to be able to gain direct experience with the tool.
The tutorial is going to be divided into two parts. The first part of the tutorial will be devoted to understanding the basic mechanics of TensorFlow. A standard execution starts with the definition of the computational graph. This graph defines the placeholders of the input parameters, which have to be fed with data on execution, and the operations that one wants to execute. Once the definition is complete, a session, i.e. a runtime context for executing the graph, is launched. I will first introduce these concepts and then we will move to executing the first demo. From this example, participants will understand the basics of graphs, sessions, and more generally the basic functioning of TensorFlow itself.
The second part of the tutorial will be dedicated to implementing a NN. I will introduce the dataset that we will use and then I will briefly review the architecture of a NN, focusing on input and output of every layer, units, activation functions and cost function. During the second demo we will implement a NN which will be composed of a few fully-connected layers in order to solve a very simple classification problem using the introduced dataset. Particular attention will be given to the graph definition in the case of a NN. In addition, we will visualize the topology and performances of the NN using TensorBoard.