Read on Twitter
0/Another great interview on Brain Inspired with @dileeplearning. https://twitter.com/pgmid/status/1344038876562685952
1/I haven't been terribly careful with distinguishing between predictive coding and predictive processing in the past, but perhaps it would be good to start being more careful with that terminology.
@PaulPcf22
@PaulPcf22
2/I've always been a bit concerned with Bastos-style models in that they assume neurons with fairly simple integration properties. https://pubmed.ncbi.nlm.nih.gov/23177956/
3/Predictive processing models from @Numenta involve more complex units; e.g. neurons capable of switching into an active/ready (laterally inhibiting) state if they receive (sparse) coincident inputs to apical portions of their dendritic arbors. https://www.frontiersin.org/articles/10.3389/fncir.2016.00023/full
4/O'Reilly's model of pulvinar as blackboard is mentioned in the interview with @pgmid, although I wonder if thalamus plays more of helping role in establishing this functionality via promoting synchrony/precision-weighting of cortical ensembles. https://twitter.com/adamsafron/status/1324104549267898373
5/If canonical microcircuits depend on non-linear dendritic integration, then might we become confused in attempting to characterize cortical computation without taking these computational properties into account?
@tyrell_turing
@tyrell_turing
6/One thing I like about vanilla predictive coding (only prediction-errors are passed upwards) is that it would both provide energetic efficiency (crucial considering the metabolic expense of brains) and induce sparsity.
@SubutaiAhmad @JeffCHawkins
https://arxiv.org/abs/1903.11257
@SubutaiAhmad @JeffCHawkins
https://arxiv.org/abs/1903.11257
7/Ascending stream gamma primarily involving superficial pyramidal neurons as basis for low/mid-level top-down biasing/pattern-completion? https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1008033
8/Gamma as both predictive-enhancement, and possibly also "explaining away" for neurons within the scope of the synchronous complex, as described by "divisive normalization" models? https://www.jneurosci.org/content/39/37/7344
8.1/Interview with @pgmid discussing Adaptive Resonance Theory with Stephen Grossberg: https://twitter.com/adamsafron/status/1321191111495397381
9/Theta as always excitatory, involving discrete acts (both overt/behavioral and covert/mental), tending to increase after events, and potentially working via cross-frequency phase coupling with gamma? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3359652/
9.1/Surprising events, p300 waveforms, and switching hippocampal modes from suppressing to enhancing? https://onlinelibrary.wiley.com/doi/abs/10.1111/psyp.13400
10/Alpha and beta as always inhibitory, to the extent that inhibition of some areas don't lead to excitation of neighboring areas via releasing lateral inhibition (broadly construed)? https://www.pnas.org/content/117/49/31459
10.1/ @dileeplearning mentioned in the interview that thalamic gain/attentional amplification could be thought of as a kind of explaining away. Very curious to understand that better.
