Does this happen in actual pDp? Optogenetic reduction of feedback inhibition (PINdlx), did reveal runaway correlations - by mechanisms consistent with our simulations.

But: co-tuned feedback inhibition in balanced EI assemblies can prevent such “runaway correlations”! Recruitment of co-tuned I neurons disproportionately inhibits “high-activity” assemblies, which once again lowers pattern correlations (here: r ≈ 0.5).

W/o inhibition, recurrent E connect. in assemblies causes nonlinear activity amplification. This depends on input strength! Stronger activity (“before”) causes stronger amplification (“after”). Thus, r becomes high when stimuli activate common sets of “high-activity” assemblies.

What is going on? Here is a toy example: Two odors (e.g. “orange” and “lemon”) are encoded as two specific, but overlapping activation patterns across Dp assemblies (simulated; correlation coefficient r ≈ 0.5).

These “runaway correlations” - between activity patterns evoked by virtual odors - were “unmasked” by partial suppression of inhibition (here: above the diagonal). They occur independently of runaway activity and can impair pattern classification.

In its extension, it suggests that structured recurrent connectivity within assemblies can generate high correlations between subsets of input patterns when excitation is not precisely balanced by (feedback) inhibition.

In our new manuscript, we now combine insights from (1) a biologically-constrained network model and (2) optogenetic perturbation experiments in pDp.

Paper alert! We combine modeling and experiments to study “balanced” EI assemblies of co-tuned excitatory and inhibitory neurons in a memory network. TL; DR: EI assemblies suppress “runaway activity” & “runaway correlations” - which interfere w/ memory capacity & discriminability

It's a great honor to be selected among so many outstanding #postdoc papers from @FMIscience! Kudos to co-authors @BionicCamel, Chie Satou, Shinichi Higashijima, and Rainer Friedrich. Paper digest: bit.ly/33nZ9A1.