[Comp-neuro] Joint A*STAR (Singapore) and King’s College London PhD Studentship

Yansong Chua james4424 at gmail.com
Wed Dec 19 18:14:48 CET 2018


We would like to encourage potential PhD candidates interested to work
on understanding the role of stochastic neuron models in neuromorphic
computing to apply for the above PhD studentship.

We are including below a non-technical introduction to the project
involved. The candidate will be jointly supervised by Prof. Osvaldo
Simeone, Professor of Information Engineering, King's College London
and Dr. Yansong Chua, Institute for Infocomm Research, A*STAR.

For further queries, we maybe reached at:
Prof Simeone:
osvaldo.simeone at kcl.ac.uk
Dr Chua:
chuays at i2r.a-star.edu.sg

More application details are also available at

https://www.kcl.ac.uk/research/funding-opportunities/doctoral-research-opportunities/homeeu-funding.aspx

https://www.kcl.ac.uk/research/funding-opportunities/doctoral-research-opportunities/current-phd-opportunities/astar-phd-studentships.aspx

For all its recent breakthroughs, modern machine learning based on
deep neural networks is becoming increasingly unaffordable in terms of
computing and energy resources needed to run training algorithms that
achieve state-of-the-art performance. This poses possibly
insurmountable challenges for the implementation of efficient learning
methods on resource-limited devices such as smart sensors or
wearables. A possible solution to this problem is the adoption of the
new paradigm of neuromorphic computing, which relies on
energy-efficient sparse spike-domain processing and communication that
are inspired by the operation of the brain. Whether Spiking Neural
Networks (SNNs) can overcome the limitations of conventional deep
networks for the implementation of low-power machine learning is a
fundamental question that is currently being investigated by major
technology companies and universities. In this project, this issue
will be tackled both theoretically and through hands-on experiments by
leveraging the complementary expertise of the respective research
teams at KCL and A*STAR. Specifically, this research will seek to
understand whether conventional deterministic models for SNNs can be
improved by probabilistic models, which are typically used in
neuroscience to model the brain operation, in terms of accuracy,
speed, and robustness. In the first two years, at KCL, the project
will concentrate on deriving models and learning rules for
probabilistic SNNs. In the last two years, at A*STAR, the research
will shift to aspects related to implementation, with a focus on the
comparison between deterministic and probabilistic SNNs and on the use
of nano-scale devices for the implementation of probabilistic SNNs.


More information about the Comp-neuro mailing list