|
Learning in populations of spiking neurons: how to find strength in numbers| old_uid | 7790 |
|---|
| title | Learning in populations of spiking neurons: how to find strength in numbers |
|---|
| start_date | 2009/12/08 |
|---|
| schedule | 10h30 |
|---|
| online | no |
|---|
| summary | It is widely believed that the aggregation of many individual neurons into a population is a key mechanism for achieving robust information processing in the brain. Thanks to averaging, fluctuations present in the single neurons have only little influence on the population response. However, in the context of reinforcement learning, this averaging has its flip side. Since the single neuron performance is only loosely related to the population response, a global reinforcement signal based on the population response can only unreliably assess the performance of any single neuron. Reinforcement feedback is therefore difficult to be interpreted at the level of the single neuron (or even the single synapse) and, for standard reinforcement algorithms, slows down dramatically with increasing population size. We suggest a novel, biologically realistic form of synaptic plasticity, which combines the reward feedback with the population feedback, and which overcomes the degradation of learning with increasing population size. Work done together with Robert Urbanczik. |
|---|
| responsibles | <not specified> |
|---|
| |
|