[Lab Seminar][Neural code][Choe] Good decisions require more than information

Jul 2, 2021

Lab seminar: September 15, 2021

This perspective piece in Nature Neuroscience talks about a really interesting recent result regarding the so-called “efficient coding” principle for neural coding.

In computational neuroscience, there’s been a long tradition of emphasizing the efficiency of neural coding: information maximization, mutual information minimization, redundancy reduction, factorial code, decorrelation, etc. etc.

The main idea is, to maximize the “information” carrying capacity, neuronal activity should not be correlated. Anything that leads to correlated firing patterns are considered suboptimal.

However, in the paper described in this perspective article, good task-level decision in a mouse is associated with “more” correlated activity than less, which is directly against the efficient coding principle.

Their conclusion is that efficient coding principle may apply well to early sensory processing, but for high-level tasks, this may have adverse effects.

The following quote summarizes this well: “Ultimately, the findings of Valente et al. emphasize that what matters most is the animal’s task performance; other goals (like maximizing sensory information) are secondary.”

https://www.nature.com/articles/s41593-021-00883-9

Here’s the main ref: https://www.nature.com/articles/s41593-021-00845-1

← Back to all articles         Quick Navigation:    Next:[ j ] – Prev:[ k ] – List:[ l ]