-
-
Notifications
You must be signed in to change notification settings - Fork 1k
GSoC_2015_project_deep_learning
Sergey Lisitsyn edited this page Feb 18, 2015
·
18 revisions
Taking last year's project to a new level.
- Theofanis Karaletsos
- Gunnar Rätsch
- Sergey (github: lisitsyn, IRC: lisitsyn)
Pretty difficult. Requires:
- C++ programming
- Understanding what neural nets are and how do we learn them
- OpenCL programming is a plus
- Understanding Of inference in graphical models, variational inference is a plus
- Experience with Stochastic Optimization is a plus
Last year we had an introductory project completed by Khaled Nasr on deep learning which introduced most essential components to use neural networks with Shogun. This project considers extending and improving things we already have with some cutting edge algorithms.
- First of all, you'd have to dive into the existing goods. We believe the easiest way to do that is to improve the code through small iterative changes that improve code quality
- Training neural nets with CPU only is really slow and we're actually pretty near to have GPU training in Shogun so the next step is to finish up GPU training and finally train some deep net with a GPU
- We expect this project to consider active research problems so probably you can even come up with something really-really new!
Deep learning is everywhere - it is on the bleeding edge of research. Most importantly, in recent attempts deep learning is getting the benefit of a bayesian treatment and the goal of this project is to marry deep nets and variational inference in the spirit of Reverend Bayes.
- Auto-Encoding Variational Bayes by Kingma and Welling
- Stochastic Backpropagation and Approximate Inference in Deep Generative Models by Rezende et al
- Neural Variational Inference and Learning in Belief Networks by Mnih and Gregor
- Deep Exponential Families by Ranganath, Blei et al
- "Neural Networks for Machine Learning" by Geoffrey Hinton (https://www.coursera.org/course/neuralnets)
- "Deep learning" by Yoshua Bengio, Ian Goodfellow and Aaron Courville (http://www.iro.umontreal.ca/~bengioy/dlbook/)