Portrait of Sushrut Thorat

Dr. Sushrut Thorat

Cognitive computational neuroscientist building predictive agents for naturalistic vision and action; focused on brain alignment, developmental curricula, and continual learning.


Appointments

  • Postdoc (Machine Learning) — Institute of Cognitive Science, Osnabrück University
    Advisor: Tim Kietzmann
    2022–2026
  • Postdoc (Cognitive Neuroscience) — MARCS Institute, Western Sydney University
    2026 (incoming)

Research focus

  • Active, naturalistic vision via self-supervised, predictive agents.
  • Brain alignment: mechanistic and representational tests (visual system; scenes & objects).
  • Developmental curricula for robust generalization.
  • Continual learning and functional drift in task-relevant / null subspaces.

Highlights

  • Glimpse Prediction Networks (GPN) — action-conditioned prediction during eye movements yields scene representations aligned with human visual cortex. (arXiv, 2025; under review at Science Advances)
  • Developmental visual diet (DVD) for robust vision — human-like coarse-to-fine training yields more robust, shape-based AI vision. (arXiv, 2025; in press at Nature Machine Intelligence)
  • Functional drift for continual learning — keep readouts stable while representations move in the null space to support new learning. (CoLLAs, 2024)
  • Animacy organization in human VTC — visual categorizability and perceived agency explain independent components of the animacy continuum. (eLife, 2019)

More: projects, code, and writeups (sushrutthorat.com/work).


Education

Recognition & service

  • MSCA Seal of Excellence (92.6%) — postdoctoral proposal on modeling human visual development (2025)
  • Reviewer — journals & conferences including Nature Human Behaviour, Nature Communications, Science Advances, NeurIPS, ICLR, eLife, …
  • Supervision — 14 BSc, 3 MSc, 3 PhD students