Projects & publications

Research

NerDS Lab · Self-supervised learning

Cross-subject representation learning

Evaluating cross-subject generalization and zero-shot transfer across 142 imaging sessions from six animals, testing whether learned representations preserve biological identity in unseen subjects.

NerDS Lab · Neural decoding

Zero-shot behavior decoding

Extending ZEN to large-scale electrophysiology data and learning shared representations across subjects and recording conditions. Matched a supervised fine-tuning baseline on running-speed decoding using 75% fewer target-subject trials.

NerDS Lab · Neural representations

Representation dynamics

Studying how learned representations change over time and across behavioral states, sessions, and subjects, using optimal transport to align embedding spaces and evaluate cross-session generalization.

  1. Active Inference with Reusable State-Dependent Value Profiles

    Poschl, J.

    Biological Cybernetics · 2026

  2. ZEN: A Framework for Zero-Shot, Across-Subject Neural Decoding

    Arora, V., Lachi, D., Mahato, S. P., Andre, A., Poschl, J., Chapuis, G., Whiteway, M. R., Dyer, E. L.

    NeurIPS · 2026 · Under review

  3. Word Predictability on Code-switching Points in Cantonese–English Discourse

    Chan, A., Li, Y., Poschl, J.

    Proceedings of the Society for Computation in Linguistics (SCiL) · 2026

  1. FormAI

    Phone-based athletic form analysis combining pose estimation, temporal phase classification, and biomechanical rules to deliver personalized motion feedback.

    Computer vision · Biomechanics