Sabine Muzellec

Vision, Brain & AI ⭐

🔬 I am a postdoctoral researcher in computational neuroscience in Kohitij Kar’s lab at York University (Toronto, Canada), where I study the computational principles underlying high-level visual processing. My research bridges artificial neural networks, primate neurophysiology, and human behavior to investigate how visual representations emerge and how they differ across neurotypical and autistic perception.

🧠 Methodologically, my work integrates ANN perturbations, biologically inspired circuit modeling, representational alignment techniques, primate electrophysiology, and human behavioral experiments.

🛠️ A central focus of my work is the development of perturbed and biologically constrained ANN models that incorporate circuit-level features such as neural noise and excitatory–inhibitory balance. In parallel, I develop methodological tools that more rigorously compare ANN representations with brain activity. These include reverse predictivity, which identifies ANN units that fail to align with macaque IT neurons, and MAPS, a framework for finding the ANN explanations that best match human explanations.

🧩 Beyond visual perception, I am also interested in image memorability. Through model-based image synthesis and behavioral experiments, I study individual and clinical differences in memory performance and test how targeted manipulations of image features can enhance memorability.

🎓 I obtained my PhD in computer science in 2025, co-supervised by Rufin VanRullen (CerCo, Toulouse) and Thomas Serre (Brown University). I developed complex-valued neural network models to study neural synchrony as a mechanism for visual binding. This work introduced new architectures that induce phase-based synchrony in ANNs and demonstrated how temporal coordination can improve robustness, feature grouping, and human-like generalization, while also comparing synchrony signatures in models and primate IT cortex.

🌱 In the long term, my goal is to contribute to computational psychiatry by developing mechanistic models of diverse neural and psychiatric conditions. I aim to build computational frameworks that capture different sensory and cognitive processing profiles, support clinical translation through improved diagnostic tools and hypothesis testing, and enable model-based stimulus manipulations that can help shift behavioral responses toward more typical patterns. This work also advances fundamental understanding of high-level visual computation and its variability across individuals.