Modeling of when


My understanding of the HTM theory and of the references therein to the underlying neurophysiology, is that there is no modeling of the correct time delay until the arrival of the expected stimuli, given the previous basal or apical NMDA spike, i.e, there seems to be a model for the where and the what, given the recent experience, but not for the when, something I would consider crucial in understanding how the processing of music (and every other human sensitive experience I would say) happens. Can anyone correct me or point me to the proper papers please ?

Thank you

1 Like

At the most basic level, the HTM/SDR model encodes transition, necessary to build sequences.

There are numerous timekeepers in the older brain structures that are more than likely sampled the same way as a sense just like sight and sound.[1]

So - not just in one place and more than just a single mechanism.

[1] Scientific Background - The Brain’s Navigational Place and Grid Cell System.

See page 5, middle right side of the page.

Thanks Mark and Matt.
After your answers, I also found these references useful

I actually came to this question from asking myself what could be the implications of applying the sensorimotor basis of the HTM model to modeling higher functions like causality-based reasoning (induction, deduction) - causality, not sensor input, being the proposed synaptic biasing driver of rational thought - and if the universality of the neural computation algorithm proposed by HTM across regions would therefore force me to consider integrating the biasing effects of precise timing in the formation of causality-driven predictions (something weird…but philosophically interesting). Anna C. Nobre, in her article (first link) however alludes to the expected loss of timing precision as we move towards higher cortical regions (even as she confirms Jeff’s statement about the top-down nature of temporal biases).

It took some time but I hope it helps:
Time Finds Its Place in the Hippocampus

Thanks Mark :slight_smile:

I had some problems with the fact that this paper used too much statistical modeling to extract time metrics from what could just be an integrative way of modeling speed by the brain, instead of relying on more direct experimental ways to isolate this dimension, but they do state that other experiments tried to do that exactly and faced different sources of pollution. And it does seem natural, from a behavioral and perceptive point of view that we model time closely related with space.

Anytime you model movement through space you are mixing time and pace. (deliberate choice of wording)
The little treadmill experiment removes shifting visual references as a cue of movement in space.
Varying the speed changes the number of footsteps per unit time.
I don’t know that you could do a more elegant experiment to isolate the influence of time on path formation.

This in no way changes my assertion that there are multiple time-keepers in the brain.

The cortex learns these sequences in a different location; patent HM could not make new memories but could remember events prior to his surgery. This strongly suggests that the hippocampus encodes time and space but that this is then transferred to the cortex to make it available in the conscious experience.