Last Friday, we enjoyed a visit by Dr. Martin Rehn.
Title: Cell assemblies and computation in cortical networks
Abstract: Recurrent neural networks are powerful computational structures. Intractable in the general case, their power is yet to be harnessed, both for practical applications and as a model for the brain. One class of recurrent networks that is theoretically well understood is attractor memory networks. Starting from this idea, we explore extensions that have non-trivial temporal dynamics, and how they apply to sensory coding. It will also be shown how an attractor memory can operate on top of a fairly realistic cortical circuitry, with some conclusions for cortical modelling.
Bio: Martin Rehn is a postdoctoral fellow at the Redwood Center for Theoretical Neuroscience, UC Berkeley, and a Research Scientist at Google. He received a PhD in Computer Science from the Royal Institute of Technology in Stockholm in 2006 and an MSc in Engineering Physics from the same institution in 1999. He is interested in representation and computation in early sensory cortices, associative memory models, and cortical simulations.
On Dec 4, 2008, we had a spirited and wonderful talk by Dr. Dileep George who is Co-Founder and CTO of Numenta. You can find his thesis here.
Title: Towards a Mathematical Model of Cortical Circuits Based on Hierarchical Temporal Learning in the Brain
Abstract: It is well known that the neocortex is organized as a hierarchy. Hierarchical Temporal Memory is a theory of the neocortex that models the necortex using a spatio-temporal hierarchy. The HTM hierarchy is organized in such a way that the higher levels of the hierarchy incorporate larger amounts of space and longer durations of time. The states at the higher levels of the hierarchy vary at a slower rate compared to the lower levels. It is speculated that this kind of organization leads to efficient learning and generalization because it mirrors the organization of the world.
I will start this talk by demonstrating the recent advances at Numenta in using HTM for object recognition. We are able to recognize objects in clutter with a high degree of accuracy. Top-down attention based feedback is used to recognize multiple objects in a scene. Feedback is used to segment out objects from clutter.
I will then describe how the assumptions of hierarchical temporal learning can lead to a mathematical model for cortical circuits. An HTM node is abstracted using a coincidence detector and a mixture of variable memory Markov chains. Bayesian belief propagation equations on this HTM node gives a set of operation related constraints for the cortical circuits. Anatomical and physiological data provide a second set of constraints related to organization of the circuits. The combination of these two constraints can be used to derive a set of cortical circuits that explain many anatomical and physiological features and predict several other. I will then demonstrate the application of these circuits in the modeling of the subjective contour effect.
Bio: Dileep George is the Chief Technology Officer of Numenta — a company he co-founded with Jeff Hawkins and Donna Dubinsky. His primary research interests are in understanding the organizational properties of the world and in linking that to the cortical architecture and micro-circuitry.
Dileep joined the Redwood Neuroscience Institute as a Graduate Research Fellow and began working closely with Jeff Hawkins in extending and expressing Jeff’s neuroscience theories in mathematical terms. He created the first proof-of-concept program to illustrate these concepts, which triggered the launch of Numenta in 2005. Within five months of Numenta¹s founding, Dileep and his team created the first prototype of HTM technology. Prior to his graduate studies, Dileep worked on developing algorithms for 3G wireless modems.
Dileep holds a Bachelor’s degree in Electrical Engineering from the Indian Institute of Technology in Bombay and Master’s and Ph.D degrees in Electrical Engineering from Stanford University. Dileep’s PhD thesis provides a detailed study of the hierarchical temporal learning in the neocortex.