use https for features.
text: sort by
tags: modified
type: chronology
hide / / print
ref: -2020 tags: current opinion in neurobiology Kriegeskorte review article deep learning neural nets circles date: 02-23-2021 17:40 gmt revision:2 [1] [0] [head]

Going in circles is the way forward: the role of recurrence in visual inference

I think the best part of this article are the references -- a nicely complete listing of, well, the current opinion in Neurobiology! (Note that this issue is edited by our own Karel Svoboda, hence there are a good number of Janelians in the author list..)

The gestalt of the review is that deep neural networks need to be recurrent, not purely feed-forward. This results in savings in overall network size, and increase in the achievable computational complexity, perhaps via the incorporation of priors and temporal-spatial information. All this again makes perfect sense and matches my sense of prevailing opinion. Of course, we are left wanting more: all this recurrence ought to be structured in some way.

To me, a rather naive way of thinking about it is that feed-forward layers cause weak activations, which are 'amplified' or 'selected for' in downstream neurons. These neurons proximally code for 'causes' or local reasons, based on the supported hypothesis that the brain has a good temporal-spatial model of the visuo-motor world. The causes then can either explain away the visual input, leading to balanced E-I, or fail to explain it, in which the excess activity is either rectified by engaging more circuits or engaging synaptic plasticity.

A critical part of this hypothesis is some degree of binding / disentanglement / spatio-temporal re-assignment. While not all models of computation require registers / variables -- RNNs are Turning-complete, e.g., I remain stuck on the idea that, to explain phenomenological experience and practical cognition, the brain much have some means of 'binding'. A reasonable place to look is the apical tuft dendrites, which are capable of storing temporary state (calcium spikes, NMDA spikes), undergo rapid synaptic plasticity, and are so dense that they can reasonably store the outer-product space of binding.

There is mounting evidence for apical tufts working independently / in parallel is investigations of high-gamma in ECoG: PMID-32851172 Dissociation of broadband high-frequency activity and neuronal firing in the neocortex. "High gamma" shows little correlation with MUA when you differentiate early-deep and late-superficial responses, "consistent with the view it reflects dendritic processing separable from local neuronal firing"

hide / / print
ref: -0 tags: computational neuroscience opinion tony zador konrad kording lillicrap date: 07-30-2019 21:04 gmt revision:0 [head]

Two papers out recently in Arxive and Biorxiv:

  • A critique of pure learning: what artificial neural networks can learn from animal brains
    • Animals learn rapidly and robustly, without the need for labeled sensory data, largely through innate mechanisms as arrived at and encoded genetically through evolution.
    • Still, this cannot account for the connectivity of the human brain, which is much to large for the genome; with us, there are cannonical circuits and patterns of intra-area connectivity which act as the 'innate' learning biases.
    • Mice and men are not so far apart evolutionary. (I've heard this also from people FIB-SEM imaging cortex) Hence, understanding one should appreciably lead us to understand the other. (I agree with this sentiment, but for the fact that lab mice are dumb, and have pretty stereotyped behaviors).
    • References Long short term memory and learning to learn in networks of spiking neurons -- which claims that a hybrid algorithm (BPTT with neuronal rewiring) with realistic neuronal dynamics markedly increases the computational power of spiking neural networks.
  • What does it mean to understand a neural network?
    • As has been the intuition with a lot of neuroscientists probably for a long time, posits that we have to investigate the developmental rules (wiring and connectivity, same as above) plus the local-ish learning rules (synaptic, dendritic, other .. astrocytic).
      • The weights themselves, in either biological neural networks, or in ANN's, are not at all informative! (Duh).
    • Emphasizes the concept of compressability: how much information can be discarded without impacting performance? With some modern ANN's, 30-50x compression is possible. Authors here argue that little compression is possible in the human brain -- the wealth of all those details about the world are needed! In other words, no compact description is possible.
    • Hence, you need to learn how the network learns those details, and how it's structured so that important things are learned rapidly and robustly, as seen in animals (very similar to above).

hide / / print
ref: -0 tags: john F kennedy quote opinion thought lie myth date: 04-14-2009 21:13 gmt revision:1 [0] [head]

For the great enemy of truth is very often not the lie—deliberate, contrived, and dishonest—but the myth— persistent, persuasive, and unrealistic. Too often we hold fast to the clichés of our forbears. We subject all facts to a prefabricated set of interpretations. We enjoy the comfort of opinion without the discomfort of thought.

—John F. Kennedy