Deep learning models have shown great promises in the past few years in many areas, including machine learning, natural language processing and computer vision. There has been much progress in this area in recent years, in terms of methods, applications as well as theoretical understanding. Many challenges remain: scalable models, faster learning, learning from smaller sample sets, and adoption to novel tasks. The workshop aims to bring together researchers to discuss recent advances and ongoing developments in deep learning as well as exchange their experience of current methods.

Key topics will include:

• General discussion on current deep architectures as well as recent developments in deep learning. We would like participants to get up to speed with modern methods and understand the future research challenges.
• Practical sessions: One aspect of the workshop is to help researchers become more familiar with various existing tools in deep learning (Tensorflow, Torch, Theano, caffe, etc). Support will be provided for researchers to use existing models on some fundamental problems in various areas such as in computer vision or natural language processing.
• Theoretical understanding: Training deep learning models give rise to a non-convex optimization problem. While optimizing convex functions is fairly well understood, non-convex functions are much more challenging. We will review existing analyses and discuss how they give rise to practical insights to optimize such models.

The workshop will include several research talks, a poster/demo sessions as well as practical sessions and reading groups.

We are excited to have the following invited speakers:

• Andrea Vedaldi, University of Oxford
• Koray Kavukcuoglu, Google DeepMind
• Matthias Bethge, Centre for Integrative Neuroscience,Tübingen

All attendees are encouraged to present a poster or a demonstration of their current research.

July 11th (Monday)

July 12th (Tuesday)

July 13th (Wednesday)