search

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
18 stars 8 forks source link

write a test for pooled inhibition -- started #239

Closed rcoreilly closed 1 year ago

rcoreilly commented 1 year ago

shows up in the bench_objrec model. None of our tests use pool 4D layers -- will write one for this.

rcoreilly commented 1 year ago

false alarm -- nmda variation again it seems. the real culprit was dwt submean.

rcoreilly commented 1 year ago

really need this test -- it looks like Lvis (which massively uses pooled inhibition) is behaving very differently on GPU in layer V2 which depends critically on pooled inhib.

rcoreilly commented 1 year ago

fix for divergence in v2: 9448f4d94da2326b2c3c4a791bbf662f28cfcf3e

Now Lvis CPU vs. GPU are nearly identical, except for the the critical output layer when increasing NData higher. crazy.

did thorough test in pool_test with large layers and ndata = 8 and GABAB, NMDA set to 0 (to avoid contamination) -- Gi does not diverge, but other things are a bit diff still (INet) -- this is just an unavoidable issue apparently: #237

rcoreilly commented 1 year ago

found the issue affecting Lvis -- this should be all good now and test is robust with large scale test.