Back

Neurons from pre-motor areas to the Mushroom bodies can orchestrate latent visual learning in navigating insects.

Wystrach, A.

2023-03-10 neuroscience
10.1101/2023.03.09.531867 bioRxiv
Show abstract

Spatial learning is peculiar. It can occur continuously and stimuli of the world need to be encoded according to some spatial organisation. Recent evidence showed that insects categorise visual memories as whether their gaze is facing left vs. right from their goal, but how such categorisation is achieved during learning remains unknown. Here we analysed the movements of ants exploring the world around their nest, and used a biologically constrained neural model to show that such parallel, lateralized visual memories can be acquired straightforwardly and continuously as the agent explore the world. During learning, left and right visual memories can be formed in different neural comportments (of the mushroom bodies lobes) through existing lateralised dopaminergic neural feedback from pre-motor areas (the lateral accessory lobes) receiving output from path integration (in the central complex). As a result, path integration organises visual learning internally, without the need to be expressed through behaviour; and therefore, views can be learnt continuously (without suffering memory overload) while the insect is free to explore the world randomly or using any other navigational mechanism. After learning, this circuit produces robust homing performance in a 3D reconstructed natural habitat despite a noisy visual recognition performance. Overall this illustrates how continuous bidirectional relationships between pre-motor areas and visual memory centres can orchestrate latent spatial learning and produce efficient navigation behaviour.

Matching journals

The top 2 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.