Unsupervised Detection of Recurrent Patterns in Neural Recordings with Constrained Filters

Roman Koshkin, Tomoki Fukai

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: spike sequences, place cells, recurrent pattern detection

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: Detecting sequential patterns in neural data with backprop-optimized 2D filters

Abstract: Structured spontaneous neural activity, characterized by the expression of repetitive patterns, is crucial for memory, learning and spatial navigation. However, investigating the functional role of these patterns has been challenging due to a lack of scalable methods for detecting them in large-scale recordings. To address this challenge, we propose an unsupervised approach that utilizes backpropagation to optimize the parameters of a predefined number of spatiotemporal filters, which serve as pattern detectors. We demonstrate the scalability and efficiency of our approach for detecting place cell sequences in biologically plausible synthetic and real datasets obtained from the mouse hippocampus. Our speed benchmarks demonstrate that our method significantly outperforms prior art, enabling the study of spontaneous activity in larger recordings.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 7388