|
The efficient coding hypothesis for the peripheral auditory system| old_uid | 17665 |
|---|
| title | The efficient coding hypothesis for the peripheral auditory system |
|---|
| start_date | 2019/04/02 |
|---|
| schedule | 14h30-16h30 |
|---|
| online | no |
|---|
| summary | The primary visual cortex (V1) implements the first stage of cortical processing of the visual information, and it is the most studied and the best understood among the visual areas in the brain. In this talk, I will introduce a model for the functional architecture of this area, based on a structure of metric space induced by a correlation kernel between receptive profiles. This kernel carries a geometrical structure consistent with the well-known properties of long-range horizontal connections in V1, and it is compatible with the perceptual principles ruling completion mechanisms.These characteristics are still present even when the kernel is constructed from a non-parameterized bank of filters known numerically. This led to the idea of applying this construction in the context of convolutional neural networks (CNNs): the introduction of structured lateral connections induced by the filters of convolutional layers indeed turns out to improve the performance of CNNs in tasks related to perceptual completion. |
|---|
| responsibles | Sarti, Petitot, Nadal |
|---|
| |
|