The geometry of representational drift in natural and artificial neural networks

Neurons in sensory areas encode/represent stimuli. Surprisingly, recent studies have suggested that, even during persistent performance, these representations are not stable and change over the course of days and weeks. We examine stimulus representations from fluorescence recordings across hundreds of neurons in the visual cortex using in vivo two-photon calcium imaging and we corroborate previous studies finding that such representations change as experimental trials are repeated across days. This phenomenon has been termed “representational drift”. In this study we geometrically characterize the properties of representational drift in the primary visual cortex of mice in two open datasets from the Allen Institute and propose a potential mechanism behind such drift. We observe representational drift both for passively presented stimuli, as well as for stimuli which are behaviorally relevant. Across experiments, the drift differs from in-session variance and most often occurs along directions that have the most in-class variance, leading to a significant turnover in the neurons used for a given representation. Interestingly, despite this significant change due to drift, linear classifiers trained to distinguish neuronal representations show little to no degradation in performance across days. The features we observe in the neural data are similar to properties of artificial neural networks where representations are updated by continual learning in the presence of dropout, i.e. a random masking of nodes/weights, but not other types of noise. Therefore, we conclude that a potential reason for the representational drift in biological networks is driven by an underlying dropout-like noise while continuously learning and that such a mechanism may be computational advantageous for the brain in the same way it is for artificial neural networks, e.g. preventing overfitting.Author summary: Recently, it has been shown that the neuronal representations of sensory information in the brain can vary, even during seemingly stable performance. Why such “representational drift’’ occurs in the brain is currently unknown. In this work, using experimental data that images thousands of neurons across many mice, we precisely quantify how certain representations change over time with geometric tools used to understand high-dimensional data. Across two datasets where mice are either passively viewing a movie or actively performing a task, we find the representational changes have strikingly similar geometric properties. We then induce representational changes in an artificial neural network by injecting it with several distinct types of noise while it continues to adjust its components to maintain stable performance. Comparing the properties of its representational drift to what we observed in experiment, only a specific category of noise, known as “dropout’’, matches the geometry we observed in experiments. This hints at a potential biological mechanism underlying representational drift: a random suppression of certain neuronal components and a subsequent compensating change in other components. Additionally, dropout is well-known for helping artificial neural networks learn better, potentially hinting at a computational advantage to drift in the brain.