2014/Workgroups/Discussion

Benchmarking Neuromorphic Systems

Members: Cheston Tan, Chetan Singh Thakur, Daniel Neil, Guillaume Garreau, Garrick Orchard, Jonathan Tapson, Marcello Mulas, Manu Rastogi, Michael Pfeiffer, Sadique Sheik, Sergio Davies, Shih-Chii Liu, Timmer Horiuchi, Thomas Trappenberg, Vikram Ramanarayanan, Andre van Schaik, Xavier Lagorce, Yulia Sandamirskaya

Organizers:: Jonathan Tapson (Univ. of Western Sydney)

Goal

"In God we trust; all others bring data." - W. Edwards Deming.

The neuromorphic engineering research community does not have a history of benchmarking its systems against standard problems (in the manner in which e.g. the machine learning community uses the MNIST, TIMIT and other data sets to rank the performance of machine learning algorithms). It has been argued recently that this is a barrier to improvement, in that we don't actually know how our systems are performing vis-a-vis other systems which address the same problem. This discussion group will focus on this issue, and specifically identify whether we want to benchmark our systems; if so, what kind of benchmarks would be meaningful in a neuromorphic context (e.g. factors such as power consumption, speed and latency could be as important as accuracy); and how to introduce a benchmarking paradigm without becoming unconstructively competitive. A good candidate area for this problem is in spiking networks for spatio-temporal pattern recognition, and we may choose to use this particular area to develop a set of benchmarks and explore whether they are likely to drive progress in useful ways.

Network Encoding in the Presence of Noise Correlations

Members: Chetan Singh Thakur, Shih-Chii Liu, victor minces

Organizers:: victor minces (Univ. of California, San Diego)

Goal

Discuss the effect of correlated noise in network encoding.

Noise degrades the amount of information that can be encoded in a network. If the noise is independent across the network’s elements (neurons, registers, etc.) it can be reduced by averaging. However, this is not possible when the noise is correlated. We will discuss the influence of the signal and noise correlation structures (matrices) in network encoding. I developed this theory based on experimental results regarding the influence of learning, attention, and neuromodulation on the correlation structure of neurons, but I think it could be applied understand encoding efficiency in noisy electronic circuits. I would greatly appreciate discussing these issues with the neuromorphs.

Conscious Systems

Members: Chetan Singh Thakur, Daniel Neil, Guillaume Garreau, Jasmine Berry, John Chiasson, Marcello Mulas, Paul Verschure, Sadique Sheik, Stephen Deiss, Sergio Davies, Tim C. Pearce, Yulia Sandamirskaya

Organizers:: Stephen Deiss (Univ. California, San Diego) Paul Verschure (Universitat Pompeu Fabra)

Goal

Clarify what consciousness is, what it does, and the full range of possible instantiations.

We all know consciousness is a tricky subject hampered by lack of consensus on definition, lack of functional characterization, and disagreement about what kinds of systems can have it. We are most familiar with our own awareness and that which we infer in our social contacts. Historically there has been reluctance to ascribe it to many other life forms. Nevertheless, we speak freely even of our artifacts as sensing, perceiving and behaving active agents in the world (or approaching same). Since we are building robotic systems we should be cognisant of these ambiguities especially when talking to the public. There is optimism about building 'brains.' But opinions vary on whether the end products will experience the world at all, let alone much as we do. Come listen to Steve (an engineer/panpsychist) and Paul (a scientist with a comprehensive view) dialogue about their views and others, and bring your own if you have thought it through and you have tenure and/or thick skin.

References: ALSO SEE ATTACHMENTS AT BOTTOM OF PAGE '''

Steve's (aka David's) view in a humorous nutshell -  http://www.appliedneuro.com/UNCC_Zombie_Blues_Cure.pdf

and a more systematic exposition -  http://www.appliedneuro.com/UNCC_Submission_Draft.pdf

Stephen Deiss, “The Universal Correlates of Consciousness” Chapter 7 in Mind that Abides: Panpsychism in the new millennium, edited by David Skrbina, John Benjamins, Advances in Consciousness Research Series, 2009.

David Skrbina, Panpsychism in the West, MIT Press, 2007.

Paul's (aka Goliath's) view presented in a Ted talk -  https://www.youtube.com/watch?v=xr1WkV6tSjo

and an interview -  https://player.fm/series/convergent-science-network-podcast/interview-with-paul-verschure

Attachments