Connectionists: PhD Thesis proposal

[Emmanuel Mazer]

*PhD Thesis proposal*

*Rebooting computing : Design of stochastic machines *

*Advisor:*Emmanuel Mazer (LIG)

Contact: <mailto:Emmanuel.Mazer at inria.fr>Emmanuel.Mazer at inria.fr

*Co-advisors:*Laurent Fesquet (TIMA), Didier Piau (Institut Fourier)

Contacts: <mailto:Laurent.Fesquet at imag.fr>Laurent.Fesquet at imag.fr, 
<mailto:Didier.Piau at ujf-grenoble.fr>Didier.Piau at ujf-grenoble.fr

The Persyval Labex is offering a PhD grant for a 3-year period.

*Overview:*

As the physical limits of Moore's law are being reached, a research 
effort is launched to achieve further performance improvements by 
exploring computation paradigms departing from standard approaches.

The project aims at developing hardware dedicated to probabilistic 
computation, which extends logic computation realized by Boolean gates 
in current computer chips. Such probabilistic computing devices would 
allow to solve faster and at a lower energy cost a wide range of 
Artificial Intelligence applications, especially when decisions need to 
be taken from incomplete data in an uncertain environment.

**

*More detailed presentation: *

The overall goal of the proposed research is to design, build and 
program new types of information processing machines and to demonstrate 
their effectiveness in low-level sensor signal
processing such as sound source localization and separation.

The goal of the research is to replace the Von Neuman architecture and 
the floating point arithmetic units found in CPU and GPU by innovative 
designs of processing random bit stream in order to reduce the power 
consumption of computing devices while increasing their processing speed 
and robustness to noise.

To address this multi-disciplinary subject, the candidate will work in 
close cooperation with four laboratories (Institut Fourier, Gipsa, LIG, 
TIMA), part of the Persyval Labex. The candidate will firstly improve 
the design of an existing stochastic machine based on the Gibbs sampling 
technique. For instance, by using generating sets, one of the goals is 
to make generic the random walk used by the Gibbs sampler. Another is to 
elaborate the stochastic machine architecture and to implement it in 
hardware. Therefore the candidate will have to become familiar with 
computer architecture, stochastic computing, Bayesian programming, 
generating sets for random exploration as well as circuit design.

After an initial period for acquiring the interdisciplinary technical 
background, the PhD candidate will develop a novel architecture 
dedicated to stochastic computing and will validate it on an FPGA 
platform. As a proof a concept, the stochastic machine will be 
integrated in a signal processing chain in order to evaluate it on a 
sound source localization application or a source separation problem. 
The hardware experimentation will be done in collaboration with post-doc 
hired on the project.

*Scientific and technical skills:*

Digital circuit design, computer architecture, programming skills, 
basics in probabilities, Bayesian techniques, mathematical skills, 
digital signal processing

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20160616/a3606dd5/attachment.html>