emer / axon

Axon is a spiking, biologically-based neural model driven by predictive error-driven learning, for systems-level models of the brain
BSD 3-Clause "New" or "Revised" License
19 stars 7 forks source link

fsfffb ClampExtMin is dependent on overall layer sparsity -- default of .05 causes sparse output layers to fail to clamp #314

Open rcoreilly opened 1 year ago

rcoreilly commented 1 year ago

Param:

    ClampExtMin float32 `viewif:"On" def:"0.05" desc:"minimum GeExt value required to drive external clamping dynamics (if clamp is set), where only GeExt drives inhibition.  If GeExt is below this value, then the usual FS-FFFB drivers are used."`

based on:

    GeExts float32 `desc:"all extra GeExt conductances added to neurons, normalized by pool size"`

the "normalized by pool size" thing is the issue. For LVis, the output layer is 1/100 = .01 activity level, and with weak GeExt clamping levels, the actual GeExts levels is likely fairly low on average (todo look at actual values).

In this model, this param is not useful so I just set it to 0, but it would be good to revisit (base thr on non-normalized value?) and have a better default perhaps.