research

Connectome-constrained world models

This is a placeholder post so you can see how a research write-up renders. Replace the file in _posts/ with your own. The first paragraph is a good place to state the question in one or two plain sentences.

The idea

A world model learns to predict how an environment evolves so an agent can plan inside its own imagination. A connectome is the full wiring diagram of a nervous system. The thesis asks a simple question: if we constrain the world model’s connectivity to match a biological connectome, what do we gain?

What I did

  • Built a baseline world model in JAX.
  • Replaced the dense recurrent core with a sparse graph derived from a published connectome.
  • Compared sample efficiency and stability across a handful of control tasks.
# Illustrative — the connectome enters as a fixed sparsity mask.
W = init_weights(n, n) * connectome_mask

What I found

Write your results here. A short table or a single figure usually does more than three paragraphs of prose.

What’s next

End with the open threads — it signals you know where the work goes, which is exactly what research-engineering interviewers look for.