Ali Hummos

Ali Hummos

Email: Ahummos [at] MIT [dot] edu
Google Scholar

I am a psychiatrist and computational neuroscientist, currently a postdoctoral researcher at MIT working with Mehrdad Jazayeri. My mission is to bridge the gap between neuroscience, artificial intelligence, and psychiatry by building brain-inspired models that enhance our understanding of mental health and disease.

My philosophy stems from the belief that understanding human cognition and mental health requires models that mimic the brain's capacity to process rich temporal environments, adapt rapidly, and operate within constrained resources. While artificial intelligence has made remarkable strides, it has diverged from its roots in computational neuroscience. Current AI systems often neglect the dynamic, agentive, and context-sensitive aspects central to human cognition. This divergence leaves critical questions unanswered: How do we form and maintain a sense of self and what are the benefits of questioning it so harshly? How do mechanisms of perception and imagination influence reality and its breakdown in psychosis? What drives self-sustaining feedback loops in emotional and cognitive states, such as those seen in trauma, depression, and anxiety?

My work aims to address these gaps by developing networks that emulate essential brain dynamics, such as segmenting temporal experiences and switching between cognitive states. Ultimately, I seek to embed these models in multi-agent environments to study the interactions and environmental factors that contribute to mental illness.

News

  • Spotlight at NeurIPS 2024: Our paper "Flexible task abstractions emerge in linear networks with fast and bounded units" explores task switching and modular weight specialization in neural networks. Read more.
  • Upcoming Conference: Attending RLDM 2025!

Projects

  • Neural network models of thalamocortical circuits: Understanding how task discovery and continual learning can emerge from neural architectures.
  • Brain-inspired algorithms: Developing algorithms inspired by brain dynamics to solve machine learning challenges like task identification in streams of unlabeled data.
  • Context switching in schizophrenia: Collaborating with Matthew Nassar to explore cognitive dynamics in schizophrenia patients.

Recent Publications