Abstract

An important question about the neural code asks how task-relevant variables are represented in neural populations. Neurons may be purely selective to only one task-relevant variable, or they may exhibit a mixture of selectivities. At its core, the work of Stefano Fusi and colleagues has sought to establish the special significance of nonlinear mixed selectivity neurons, arguing that they underlie flexible and generalisable neural representations in complex cognitive tasks. Specifically, they show that nonlinear mixed selectivity allows for higher dimensional representations (abstractions) of task variables and demonstrate that linearly separable, clustered geometries arise in both biological and artificial neural networks that enable high-level performance. This talk will discuss the motivation for studying nonlinear mixed selectivity, the theoretical framework proposed by Fusi and colleagues, and an array of supporting human and non-human primate experimental evidence. Papers to be discussed: Rigotti, M. et al. The importance of mixed selectivity in complex cognitive tasks. Nature 497, 585–590 (2013). Fusi, S., Miller, E. K. & Rigotti, M. Why neurons mix: high dimensionality for higher cognition. Current Opinion in Neurobiology 37, 66–74 (2016). Bernardi, S. et al. The Geometry of Abstraction in the Hippocampus and Prefrontal Cortex. Cell 183, 954-967.e21 (2020). Boyle, L. M., Posani, L., Irfan, S., Siegelbaum, S. A. & Fusi, S. Tuned geometries of hippocampal representations meet the computational demands of social memory. Neuron 112, 1358-1371.e9 (2024). Courellis, H. S. et al. Abstract representations emerge in human hippocampal neurons during inference. Nature 632, 841–849 (2024).