Document Actions

You are here: Home / Bernstein Conference / Past Conferences / 2017 / Satellite Workshops / Deep Neural Networks Tutorial for Computational Neuroscientists (Bethge, Rauber)

Deep Neural Networks Tutorial for Computational Neuroscientists (Bethge, Rauber)


This is not a workshop but a hands-on tutorial with the goal to provide a gentle introduction on how to study deep neural networks (DNNs). ln recent years DNNs have become increasingly proficient in mimicking the perceptual inference abilities of humans and animals. Thus high-performance  deep neural networks (DNNs) open a new door to study how neural networks can solve ecologically  relevant tasks. Additionally,  deep neural networks can serve as a powerful modeling and analysis tool for experimental data. While those advantages have been noted by many, there is still an entrance barrier that prevents many computational neuroscientists  from taking advantage of Deep Learning.

We will start the day with a gentle overview over the broad Deep Learning Iandscape and the merits DNNs promise for the field of Computational  Neuroscience.  The second half of the morning session will introduce the basics of Tensorflow, currently the most widely used open-source  machine learning package. The afternoon will be split into three hands-on sessions dealing with the modeling of biological computations,  human decision making and population representations.

For the programming  lectures a Iaptop as weel as some experience with Python are required. Each participant will get access to a cloud-based  GPU instance and a fully set-up Python environment  (using Jupyter Notebook).

The goal of this tutorial is to provide the necessary background  and the tools needed to run first practical experiments  with deep neural networks.