About
I'm a researcher working on the mechanistic interpretability and control of neural sequence models — understanding how these systems compute, and using that understanding to steer their behaviour safely.
I'm a PhD candidate in Applied Mathematics at the University of Western Ontario, working with Lyle Muller and Ján Mináč across the Muller Lab and the Fields Lab for Network Computation at the Fields Institute. As neural networks are deployed in higher-stakes settings, I think the systems we trust should be transparent by construction: understandable from their structure, not only explained after the fact.
Before the PhD I worked as a strategy consultant advising major North American banks.
Publications & research
Explicit operator explains end-to-end computation in the modern neural networks used for sequence and language modeling
Shikder, A. N.* and Dey, R.* et al.(* equal contribution)
Under review · Physical Review X
arXiv →
Exact expression for maximum Lyapunov exponent during transient dynamics in networks with expressive computational capabilities
Powanwe, A.*, Liboni, L.*, Shikder, A. N., et al.
Under review · Physical Review Letters
arXiv →
S¹: A Precisely Interpretable State Space Model via an End-to-End Closed-Form Oscillatory Operator
Dey, R.* and Shikder, A. N.*, et al. (* equal contribution)
Under review · NeurIPS
Complex Walks for Interpretable Associative Memory in State-Space Networks
Shikder, A. N., et al.
In preparation