Who am I?

I am a NeuroAI researcher working at the intersection of Artificial Intelligence (AI) and Computational Neuroscience. I was trained at , where I studied Mathematics (BSc), Computer Science (BSc, MSc), and Machine Learning (MSc). I began working on computational models of cognition during my PhD at , in the Mnemosyne team, under the mentorship of Nicolas Rougier and Xavier Hinaut. There, I developed recurrent neural network models of working memory. I then joined as a postdoctoral scholar in the Stanford Cognitive & Systems Neuroscience Laboratory (SCSNL), under the mentorship of Vinod Menon. My work there focused on computational models of mathematical cognition. For more details, see my résumé.

What defines my research?

• A core research question: How do neural circuits give rise to complex behavior?
• A core research approach: Develop minimalist biologically and developmentally inspired AI models to provide mechanistic explanations of how the brain generates behavior, and what neural disruption can disrupt behavior.

Selected publications

1. Strock, A., Nghiem, T.-A. E., Trouvain, N., Mistry, P. K., Hinaut, X., & Menon, V. (2025). Recurrent neural network models reveal unified mechanisms generating event-related potentials from MMN to P300. BioRxiv.
   ABSTRACT

   PREPRINT

   CODE

   The brain’s ability to detect behaviorally relevant stimuli from sensory inputs is fundamental to cognition, yet the neural mechanisms linking synaptic processes to event-related potential (ERP) signatures remain unclear. Here, we develop recurrent neural network (RNN) models of ERP responses demonstrating that short-term synaptic depression – a ubiquitous plasticity mechanism – provides a unified computational framework for mismatch negativity (MMN) and P300 responses across passive and active oddball paradigms. Our models reveal that neural populations spontaneously organize stimulus representations into probability-dependent geometric manifolds, where rare events occupy expanded state space. Hierarchical connectivity creates 9-fold signal amplification with enhanced noise robustness, explaining P300’s functional advantages over sensory responses. Critically, the same synaptic mechanism accounts for attentional modulation of behaviorally relevant stimuli, providing the first unified explanation bridging automatic and controlled attention. This framework offers quantitative predictions for how synaptic and connectivity disruptions manifest as altered MMN and P300 characteristics in neuropsychiatric disorders including schizophrenia and autism.

2. Strock, A., Mistry, P. K., & Menon, V. (2025). Personalized deep neural networks reveal mechanisms of math learning disabilities in children. Science Advances, 11(23). https://doi.org/10.1126/sciadv.adq9990
   ABSTRACT

   PREPRINT

   URL

   CODE

   Learning disabilities affect a substantial proportion of children worldwide, with far-reaching consequences for their academic, professional, and personal lives. Here we develop digital twins—biologically plausible personalized deep neural networks (pDNNs)—to investigate the neurophysiological mechanisms underlying learning disabilities in children. Our pDNN reproduces behavioral and neural activity patterns observed in affected children, including lower performance accuracy, slower learning rates, neural hyperexcitability, and reduced neural differentiation of numerical problems. Crucially, pDNN models reveal aberrancies in the geometry of manifold structure, providing a comprehensive view of how neural excitability influences both learning performance and the internal structure of neural representations. Our findings not only advance knowledge of the neurophysiological underpinnings of learning differences but also open avenues for targeted, personalized strategies designed to bridge cognitive gaps in affected children. This work reveals the power of digital twins integrating artificial intelligence and neuroscience to uncover mechanisms underlying neurodevelopmental disorders.

3. Strock, A., Hinaut, X., & Rougier, N. P. (2020). A Robust Model of Gated Working Memory. Neural Computation, 32(1), 153–181. https://doi.org/10.1162/neco_a_01249
   ABSTRACT

   PREPRINT

   URL

   CODE

   Gated working memory is defined as the capacity of holding arbitrary information at any time in order to be used at a later time. Based on electrophysiological recordings, several computational models have tackled the problem using dedicated and explicit mechanisms. We propose instead to consider an implicit mechanism based on a random recurrent neural network. We introduce a robust yet simple reservoir model of gated working memory with instantaneous updates. The model is able to store an arbitrary real value at random time over an extended period of time. The dynamics of the model is a line attractor that learns to exploit reentry and a nonlinearity during the training phase using only a few representative values. A deeper study of the model shows that there is actually a large range of hyperparameters for which the results hold (e.g., number of neurons, sparsity, global weight scaling) such that any large enough population, mixing excitatory and inhibitory neurons, can quickly learn to realize such gated working memory. In a nutshell, with a minimal set of hypotheses, we show that we can have a robust model of working memory. This suggests this property could be an implicit property of any random population, that can be acquired through learning. Furthermore, considering working memory to be a physically open but functionally closed system, we give account on some counterintuitive electrophysiological recordings.