Jesseba Fernando

Network Science Institute @ Northeastern University.

about me

IMG_2895.jpg

I’m Jesseba, a Network Science PhD student at Northeastern University. I’m advised by Dr. Sam Scarpino. I study how learning changes networks in both the brain and machines, and use methods from network science and information theory.

In the brain, I study how the functional roles of neurons change as animals learn, and whether these changes can predict future behavior. In transformers, I look at the representations as they progress layer by layer through the residual stream and ask how the dynamics of the computations relate to the network’s structure. Composition brings them together since both systems achieve generalization by recombining what they already know. This flexibility comes at a cost because a system that composes freely loses its ability to tell the inputs apart. I want to understand how each system negotiates that tradeoff, and how the coding strategy it chooses shapes what it can learn.

Before joining Northeastern, I spent five years in Mark Andermann’s lab at Harvard Medical School, studying how hunger and other motivational states change what visual cortex responds to. I then worked with Bill Lotter at Dana-Farber Cancer Institute on how medical imaging models behave when the data they receive differs statistically from the data they were trained on.

In my free time, I co-write Of Two Minds, a newsletter with Grigori Guitchounts.

When I travel I bring more camera than I need, usually to places with bad roads. Some of it ends up here.

news

Jun 12, 2026 Session chair for the neural theory session at Seventh International Conference on Mathematics of Neuroscience and AI.
Mar 16, 2026 Co-organizing the CoSyNe 2026 workshop on Renormalization Principles in Neural Systems with Andrea Santoro and Giovanni Petri.
Jan 16, 2026 Awarded the AccelNet-MultiNet Fellowship. Collaborating with Giovanni Petri, Andrea Brovelli, and Alain Barrat on neural dynamics in Marseille and London this summer.

publications

2026

  1. Bound by semanticity: universal laws governing the generalization-identification tradeoff
    Marco Nurisso, Jesseba Fernando, Raj Deshpande, Alan Perotti, Raja Marjieh, Steven M. Frankland, and 6 more authors
    International Conference on Learning Representations, 2026
    URL arXiv
  2. Dynamics of the Transformer Residual Stream: Coupling Spectral Geometry to Network Topology
    Jesseba Fernando, and Grigori Guitchounts
    Neural Information Processing Systems, 2026
    arXiv

2025

  1. Transformer Dynamics: A neuroscientific approach to interpretability of large language models
    Jesseba Fernando, and Grigori Guitchounts
    arXiv preprint, 2025
    URL arXiv

2024

  1. Cortical reactivations predict future sensory responses
    Nghia D Nguyen, Andrew Lutas, Oren Amsalem, Jesseba Fernando, Andy Young-Eon Ahn, Richard Hakim, and 5 more authors
    Nature, 2024
    URL
  2. Beyond Structured Attributes: Image-Based Predictive Trends for Chest X-Ray Classification
    Katharina V Hoebel†, Jesseba Fernando†, and William Lotter
    Proceedings of Machine Learning Research Volume 250:610–640 , 2024
    URL

2023

  1. Brainstem serotonin neurons selectively gate retinal information flow to thalamus
    Jasmine DS Reggiani, Qiufen Jiang, Melanie Barbini, Andrew Lutas, Liang Liang, Jesseba Fernando, and 5 more authors
    Neuron, 2023
    URL

2022

  1. Visual association cortex links cues with conjunctions of reward and locomotor contexts
    Kelly L McGuire, Oren Amsalem, Arthur U Sugden, Rohan N Ramesh, Jesseba Fernando, Christian R Burgess, and 1 more author
    Current Biology, 2022
    URL