- - - -

TL;DR

Connections between neighboring cortical neurons allow neural networks to compute variance of sensory input.
Crucial for Bayesian computations in the brain.

Context

Imagine hiking through a forest. Wind rustles leaves. Strange noise catches your attention. Squirrel ? Bird ? Lion ? Wait for instagram picture, or flee for survival ? Daily problem: making sense of unreliable sensory input.

This unreliability: inverse of information variance. Constant problem in vision. Images: numerous lines, “edges,” like puzzle pieces.

Puzzle pieces vary. Problem for the primary visual cortex. Human behavior known to take uncertainty as a decision factor [1,2].

Recent research: showed uncertainty-dependent activity in macaque primary visual cortex. Complex responses, mirroring human decision-making [3].
Why and how responses exists ? Only theories until this paper.

Method

Our research: examine neurons in primary visual cortex. Construct comprehensive interaction model. Visual stimuli: Motion Clouds, mimic natural image complexity [4].

Results: single neurons

Discovered phenomenon: neurons in primary visual cortex respond distinctly to image uncertainty. Two main types: “flat” (unchanged by increased variance), “non-linear” (response quickly destroyed by increased variance).

Different timings: “flat” neurons respond slowly, “non-linear” neurons respond quickly.

Labeled groups: “flat = “resilient” as opposed to “non-linear” = “vulnerable,” clustered based on neural metrics.

Results: population of neurons

Next question: what do these neurons do? Used neural decoding: guess neurons’ “seeing” based on responses.

Both resilient and vulnerable neurons guessed average orientation well.

Resilient neurons excelled in guessing both average orientation and complexity. Not vulnerable.

Specific neurons found in different cortex layers, connecting differently. Provided anatomical basis for two neuron groups.

Results: model

Simulated differential anatomical basis using a computer model. Tweaking “recurrence” mimicked behavior of resilient and vulnerable neurons.

Overall

Recurrence explains how neurons encode (or not) input variance. Findings support understanding of brain encoding complexity, not just average features, thanks to neuronal connectivity.

Relevance

Crucial step in understanding how cortex handles “visual puzzles.” Allows brain to perform complex probabilistic computations. Model gaining popularity in neuroscience [5].

References

[1] Von Helmholtz, H. (1925). Helmholtz’s treatise on physiological optics (Vol. 3). Optical Society of America.

[2] Barthelmé, S., & Mamassian, P. (2009). Evaluation of objective uncertainty in the visual system. PLoS computational biology, 5(9), e1000504.

[3] Hénaff, O. J., Boundy-Singer, Z. M., Meding, K., Ziemba, C. M., & Goris, R. L. (2020). Representation of visual uncertainty through neural gain variability. Nature communications, 11(1), 2513.

[4] Leon, P. S., Vanzetta, I., Masson, G. S., & Perrinet, L. U. (2012). Motion clouds: model-based stimulus synthesis of natural-like random textures for the study of motion perception. Journal of neurophysiology, 107(11), 3217-3226.

[5] Spratling, M. W. (2016). A neural implementation of Bayesian inference based on predictive coding. Connection Science, 28(4), 346-383.

On a personal note

First first-authored peer-reviewed paper !!
Very long adventure — experiments started in 2020, pre-COVID-19. Wrote 4 versions of paper from scratch. Grateful to reviewers, editors at Nat. Com. Bio. for publishing opportunity.