Learning to combine top-down context and feed-forward representations under ambiguity with apical and basal dendrites

One of the most striking features of neocortical anatomy is the presence of extensive top-down projections into primary sensory areas. Notably, many of these top-down projections impinge on the distal apical dendrites of pyramidal neurons, where they exert a modulatory effect, altering the gain of responses. It is thought that these top-down projections carry contextual information that can help animals to resolve ambiguities in sensory data. However, it has yet to be demonstrated how such modulatory connections to the distal apical dendrites can serve this computational function. Here, we develop a computational model of pyramidal cells that integrates contextual information from top-down projections to apical compartments with sensory representations driven by bottom-up projections to basal compartments. When input stimuli are ambiguous and relevant contextual information is available, the apical feedback modulates the basal signals to recover unambiguous sensory representations. Importantly, when stimuli are unambiguous, contextual information which is irrelevant or opposes sensory evidence is appropriately ignored by the model. By generalizing the task to temporal sequences, we further show that our model can learn to integrate contextual information across time. Using layer-wise relevance propagation, we extract the importance of individual neurons to the prediction of each category, revealing that neurons that are most relevant for the overlap of categories receive the largest magnitude of top-down signals, and are necessary for solving the task. This work thus provides a proof-of-concept demonstrating how the top-down modulatory inputs to apical dendrites in sensory regions could be used by the cortex to handle the ambiguities that animals encounter in the real world.

PDF Abstract
No code implementations yet. Submit your code now

Datasets


  Add Datasets introduced or used in this paper

Results from the Paper


  Submit results from this paper to get state-of-the-art GitHub badges and help the community compare results to other papers.

Methods


No methods listed for this paper. Add relevant methods here