Sparse Spike Coding

Sparse Spike Coding corresponds to the notion of coding each instance from an ensemble of signals (such as natural scenes) using few elementary events (or spikes).

This technique evolved from work showing:

  1. Algorithms based on correlation based inhibition may provide a sparse coding of signals and images (see Publications/Perrinet03ieee),

  2. Appropriate tuning to natural scenes allows a Sparse Spike Coding of information and that this information is related to basic inferences,

  3. This transformation of spatial coding into neuro-temporal information proved to be efficient and general (for a review see Publications/Perrinet06) and could be extended to temporal signals.

  4. Sparse Coding corresponds to efficient coding (see Publications/Perrinet10shl). In fact, it relates to the Occam Razor: "For the same quality of coding, the simplest one should be privileged." Thus, using fewer elements relative to the dimension of the signal to represent (for instance the class of all linear combinations of edges at different orientations), that is relative to the dimension of the signal itself (for instance the number of pixels for a digital image).

When trying to evaluate and improve the efficiency of Sparse Spike Coding, we integrated this approach in a Sparse Hebbian Learning scheme.


welcome: please sign in