How is spacetime represented in the brain? : The math behind a predictive fuzzy memory network
Speaker 
Karthik Shankar
Boston University, USA


When 
Jan 19, 2018
from 03:30 PM to 04:30 PM 
Where  LH 006 
Add event to calendar 
vCal iCal 
Abstract: Neural Networks perform parallel computations in a fuzzy (meaning imprecise) fashion to represent space, time and memory, which in turn can be flexibly used for decisionmaking with regard to potential future events. A basic question to be addressed is— How will the mathematical representation of the past (memory) determine the ability to predict the future? I will present a neuralnet model where the connection weights are derived as an inverseLaplace transform operation, and is uniquely suited for flexibly timetranslating memory states by modulating those connections. I will try to convince you that the computational simplicity of this neuralnetwork architecture provides all the flexibility needed to store and extract space, time and memory representations. Over and beyond showing neuroscientific evidence for a fuzzy memory network, I will demonstrate that Predictive algorithms in machine learning can significantly benefit from incorporating such a fuzzy memory network in improving their predictive power.