Learning sparse codes and predictive context with compartmentalised inputs

There has been renewed interest in dendrites in computational neuroscience. The corresponding concept in artificial neural networks is that of compartmentalised inputs: integrating different pathways in distinct locations within each unit, with potentially different learning rules. Here we show how a single layer of neurons with multiple compartments can learn sparse codes and their predictive context using a local unsupervised rule, and how this could be used as a building block for cognitive architectures.

Date: 09/03/2018
Time: 16:00
Location: LB252

Share this post on: Twitter| Facebook| Google+| Email