Possible neural correlate of the TBT object layer learning mode


In TBT, the object layer (a.k.a. output layer; L2/3) learns to represent an object using a stable SDR. As part of that, there’s a learning mode where randomly chosen neurons are held stable in the object layer. Those neurons learn a bunch of different features at locations on the object (from the input layer; L4).

Neural Correlate

Here, when some neurons are chosen to learn the new object, those neurons don’t actually fire. Instead, they have apical dendritic spikes. That causes their basal dendrites to learn, strengthening inputs from L4 [1].

The apical dendritic spikes only last 0.1 seconds, but they are caused by higher-order thalamus [1] and could be repetitively evoked by that [2-4].

It’s nice that cells don’t have to fire to learn a new object, because that makes it possible to learn new objects while simultaneously trying to disambiguate known ones. However, the lateral connections would require firing to learn voting, so that learning would have to happen later.


This might have roles besides learning new objects. The apical dendritic spikes make it much easier for the cells to fire [2]. So apical dendritic spikes might be required to fire in some cases, e.g. upon detecting an object (before there’s lateral input).

The apical dendritic spikes are caused by higher-order thalamus, which is driven by L5tt cells. In other posts (a, b), I argued that L5tt cells burst to send an anchoring / detection signal to higher-order thalamus.

Detection could be when it chooses apical dendritic segments to represent a new object. If it turns out to be a known object, perhaps it can cancel the synaptic changes. Also, it’d choose dendritic segments based on the anchoring signal, so somehow related to reference frames.

[1] Higher-Order Thalamocortical Inputs Gate Synaptic Long-Term Potentiation via Disinhibition
[2] High-order thalamic inputs to primary somatosensory cortex are stronger and longer lasting than cortical inputs
[3] Learning-Dependent Enhancement of Persistent Activity in the Neocortex
[4] Cortical Sensory Responses Are Enhanced by the Higher-Order Thalamus
[5] Anatomically and functionally distinct thalamocortical inputs to primary and secondary mouse whisker somatosensory cortices
(The part of POm which’d cause the apical dendritic spikes is strictly higher-order, meaning it receives no sensory input [5].)

1 Like