Photo of Shivang Rawat

Shivang Rawat

AI Research Scientist, The Sequel Institute (Telepath), New York

I study how recurrent neural networks, biological and artificial, can stay stable, learn efficiently and keep learning over time. My work draws on dynamical systems, stability theory and stochastic processes.

I received my Ph.D. in Mathematics from the Courant Institute at New York University in 2025, advised by Prof. Stefano Martiniani. Before that, I studied chemical engineering at the University of Minnesota (M.S.) and IIT Roorkee (B.Tech.).

Research

I develop theory for recurrent computation in brains and machines. A recurring theme is divisive normalization: it keeps recurrent circuits stable even when their recurrent weights are strong, and it shapes how cortical areas communicate. I also study how deep continuous-time networks learn, and how long timescales emerge in stochastic neural dynamics.

Publications

  1. NeurIPS 2026

    Prospective Coding Improves Learning in Deep Continuous-Time Recurrent Networks

    S. Rawat, M. Morello, F. Morone, D. J. Heeger

    Advances in Neural Information Processing Systems (NeurIPS), 2026 Spotlight

    Introduces Recursive Quadrature Filters and a parameter-free prospective input update that mitigates depth-dependent gradient attenuation in deep continuous-time recurrent networks and state space models.

  2. PNAS 2026

    Stabilization of recurrent neural networks through divisive normalization

    F. Morone*, S. Rawat*, D. J. Heeger, S. Martiniani

    Proceedings of the National Academy of Sciences, 123(30), e2601841123, 2026

    Divisive normalization keeps an arbitrary recurrent network stable even when the spectral radius of its recurrent weights exceeds 1, and the breakdown of normalization coincides with the critical slowing down that precedes instability.

  3. eLife 2026

    Hierarchical Neural Circuit Theory of Normalization and Inter-areal Communication

    A. Pal, S. Rawat, D. J. Heeger, S. Martiniani

    eLife, 111297.1, 2026

    A hierarchical circuit theory with feedback that implements divisive normalization across cortical areas and accounts for observed power and coherence spectra and low-dimensional communication subspaces.

  4. NeurIPS 2024

    Unconditional stability of a recurrent neural circuit implementing divisive normalization

    S. Rawat, D. J. Heeger, S. Martiniani

    Advances in Neural Information Processing Systems (NeurIPS), 37, 14712–14750, 2024

    Links divisive normalization to the stability of ORGaNICs, a biologically plausible cortical circuit model, so that it can be trained with backpropagation through time without gradient clipping.

  5. PRR 2024

    Element-wise and recursive solutions for the power spectral density of biological stochastic dynamical systems at fixed points

    S. Rawat, S. Martiniani

    Physical Review Research, 6(4), 043179, 2024

    Inverse-free, element-wise and recursive solutions for the noise power spectral density of stochastic linear systems, illustrated on models from neuroscience and evolutionary game theory.

Preprints

  1. arXiv 2026

    A minimal mechanism to generate long timescales without fine tuning

    K. McClain*, S. Rawat*, M. Morrell, S. Martiniani, D. J. Heeger, F. Morone

    arXiv:2608.21586, 2026

    A dynamical feedback gain lets a linear recurrent network self-organize to the edge of stability, so long timescales and scale-free correlations arise without fine tuning.

  2. arXiv 2026

    Exact Variance and Fano Factor for Arbitrary Level Crossings in Stationary Gaussian Processes

    S. Rawat, F. Morone, D. J. Heeger, S. Martiniani

    arXiv:2605.25278, 2026

    Exact formulas for the variance and Fano factor of level crossings in smooth stationary Gaussian processes, going beyond the Kac–Rice mean rate to show when crossings cluster or become regular.

Thesis

  1. Ph.D. 2025

    Divisive Normalization as a Mechanism for Stable and Efficient Recurrent Neural Computation

    Ph.D. thesis, Courant Institute of Mathematical Sciences, New York University, 2025

* Equal contribution

Talks & posters

Honors & awards

  • 2025

    Wilhelm T. Magnus Memorial Prize, for significant contributions to mathematical sciences

  • 2024

    APS DBIO Travel Award

  • 2022

    MacCracken Fellowship, New York University

  • 2019

    Arnie Fredrickson Fellowship, University of Minnesota

  • 2018

    DAAD-WISE Scholarship for an undergraduate summer internship

  • 2015

    KVPY (Kishore Vaigyanik Protsahan Yojana) Scholar

Service

  • 2027

    Reviewer, ICLR

  • 2025–26

    Reviewer, NeurIPS