Sequence learning, prediction, and replay in networks of spiking neurons

Science Related Papers
1

Our work on extending the temporal memory (TM) component of HTM is now available at [2111.03456] Sequence learning, prediction, and replay in networks of spiking neurons. In this work, we demonstrate that the principle mechanisms underlying sequence learning, prediction, and replay in the TM model can be implemented using continuous-time dynamics with known biological ingredients such as spiking neurons, dendritic action potentials, spike-timing-dependent plasticity, and homeostasis.
By strengthening the link to biology, our implementation permits a more direct evaluation of the TM-model predictions based on experimentally accessible quantities and behavioral data.
All code is available at Sequence learning, prediction, and replay in networks of spiking neurons | Zenodo

Looking forward to your feedback.

9 Likes
2

This work is now published in PLOS CB: Sequence learning, prediction, and replay in networks of spiking neurons

6 Likes
3

This popped up in my twitter feed:

7 Likes
4

Interesting paper. Certainly supports mental time travel and metaphorical scaffolding of grammar acquisition. It may even contain the requisite mechanisms, I need to read carefully. Thanks.

1 Like
5

I think prediction is too broad a term to be meaningful. All representations are “predictions”, it’s only a matter of when and where. In terms of a neuron, I think the supposed predictions in apical dendrites are better described as context.

2 Likes