|
Spiking neural models of perception| old_uid | 5318 |
|---|
| title | Spiking neural models of perception |
|---|
| start_date | 2008/10/02 |
|---|
| schedule | 17h |
|---|
| online | no |
|---|
| summary | Neurons encode and process information with spikes. In classical connectionism, the information is conveyed by the firing rate. Spiking neuron models offer an additional dimension to the rate: synchrony. Synchronous spike trains are more effective than uncorrelated ones in driving the responses of target neurons. Because neurons can encode their inputs in a sequence of precisely timed spikes, input similarity translates into synchronous spiking, which can be easily detected and learned by afferent neurons. In this framework, elementary percepts such as lines or periodic sounds arise naturally because invariances in the stimulus result in synchronization in neuron groups (translating the line in its direction, delaying the sound by a multiple of its period). In contrast, in classical rate theory, these elementary percepts must be learned from the statistics of the environment. The unity of objects emerges naturally in spiking models from the fact that coherent motion or transformation translates into synchronized spiking in the receptor neurons. In this way objects “sign” the spike trains they elicit. |
|---|
| responsibles | Cousineau, Barthelme |
|---|
| |
|