TITLE:

Enabling Computation for Machine Learning Algorithms Inspired by Neurobiology

DATE:

Friday, January 27th, 2017

TIME:

3:30 PM

LOCATION:

GMCS 314

SPEAKER:

Dr. Doug Bergman, Algorithmic Data Scientist at KnuEdge

ABSTRACT:

Natural neural systems tightly couple a wide diversity of neuron types and connections to perform extremely efficient pattern recognition processing with very little training. On the other hand, most neural networks in use today must be trained with hundreds or thousands of supervised images. Current artificial neural network (NN) research has centered around homogeneous systems of identical neuron types in mostly multilayered perceptron models. This owes partly to the availability of powerful GPU machines, which can process data very quickly but are limited to a single-instruction, multiple-data (SIMD) paradigm.

KnuEdge’s machine architecture supports a MPMD (Multiple-Program, Multiple Data) model, allowing researchers to look to nature, to develop more powerful models of machine learning inspired by the awesome learning and recognition abilities of animal and even human brains. Sparse and heterogeneous neural network models pipeline data through specialized processing algorithms to perform specific interpretive tasks. Running on MPMD architecture, the next generation of heterogeneous NN models promise to save a lot of training time and computational resources compared to homogenous networks. The algorithms that will enable sparse NN processing can also be leveraged for other problems, such as graph search.

HOST:

Dr. Jose Castillo

DOWNLOAD: