Biologically Plausible Computational Models of Perirhinal Cortex - PROJECT SUMMARY Humans and other animals are able to seamlessly integrate sensory and mnemonic information. As a medial temporal lobe structure at the apex of high-level sensory cortices, perirhinal cortex (PRC) is ideally situated to support perceptual-mnemonic integration. Indeed, PRC has been shown to play a causal role in diverse behaviors, including familiarity-based recognition, visual object perception, and fear conditioning. Yet this confluence of functions has proven difficult to formalize, and there is considerable debate over the mechanisms that enable PRC to play a role in these diverse behaviors. By integrating traditional neuroscientific methods within a novel deep learning computational framework, this proposal aims to formalize and evaluate computational theories of PRC function. I have three specific aims: Aim 1. My graduate work has already laid the foundation to formalize PRC function. By integrating lesion, electrophysiological, and behavioral data within a deep learning framework, this work resolves decades of seemingly inconsistent experimental findings surrounding PRC involvement in visual object perception. More specifically, I find that a biologically plausible computational proxy for the primate ventral visual stream (VVS) approximates the visual discrimination behaviors of PRC-lesioned (human and non-human) primates, directly from experimental stimuli. Conversely, PRC-intact participants are able to outperform PRC-lesioned behaviors, as well as these computational proxies for the VVS—a finding that implicates PRC in these behaviors. Aim 2. The work proposed during the F99 phase will build upon my previous work to include a computational account of PRC-dependent visual discrimination behaviors. First, I will develop biologically plausible computational models of PRC-intact visual behaviors able to achieve the performance of PRC-intact human participants on concurrent visual discrimination tasks. Then, I will identify which of these PRC-models best fit in- vivo (fMRI) measurements of PRC function. This will provide extensive training in building deep learning models of visual behaviors, alongside neuroimaging expertise, harnessing resources available at my graduate institution. Aim 3. Work proposed in the K00 phase will build upon these perceptual models to include PRC's known mnemonic functions. I intend to gain extensive experience with reinforcement learning (RL) models of mnemonic behaviors. By integrating deep learning and reinforcement learning within a biologically plausible computational framework, I will build towards an integrated model of PRC-dependent perceptual-mnemonic behaviors. Collectively, this proposal offers a novel framework for characterizing typical and atypical perceptual-mnemonic behaviors, promising new insights into the neurobiology of perception and memory, while outlining the training I need to be an independent researcher at the forefront of computational cognitive neuroscience. It is my hope that this framework may provide a foundation for future clinical applications to address memory-related disorders.
