Automatic detection of spiking motifs in multi-unit raster plots

Recently, there has been growing interest in the detection of precise spike motifs in multi-unit raster plots, for a review, see https://laurentperrinet.github.io/publication/grimaldi-22-polychronies . In this internship, we will introduce an extension of such detection models by providing a generative model for raster plot synthesis. An optimal detection procedure will then be derived from this model. This takes the form of a logistic regression coupled with a temporal convolution. We will evaluate the ability of this model to detect spike patterns in synthetic data. Since this model is differentiable, we may derive an unsupervised learning method in the form of gradient descent on the loss function of an auto-encoder model for the raster using the spike patterns. This unsupervised learning method should be able to recover the synthetically generated spike patterns, and we plan to apply it to neurobiological data.

The student must have a knowledge of the mathematics used in standard machine learning (algebra, probability) and a keen curiosity for understanding the nuts and bolts underlying neural processes.

The INT is a neuroscience institute covering a wide range of fields, from bimolecular mechanisms to cognition, and this in translational to fundamental perspectives. It has a growing computational community regrouped in the Computational Neuroscience Center @ INT (CONECT, see https://conect-int.github.io ).