news

Oct 09, 2026 I gave a lab meeting to the tinlab and the broader Linguistics and AI communities at Boston University, titled "Many Solutions to the Same Task: Identifiability in Trained Neural Networks". It was a great pleasure chatting with everyone after the talk!
Oct 01, 2026 Our new paper "Traversing the solution space of neural networks with Hessian Null Space Continuation" is on arXiv! See the paper here and the project website here 💙
Sep 28, 2026 Our paper "Effectiveness of Curriculum Learning Depends on Reward Sparsity and Competing Optima" has been accepted at NeurIPS 2026! 🇦🇺
Aug 05, 2026 I'll be giving a talk on our recent work developing Hessian Null Space Continuation, a scalable method for traversing functionally equivalent networks with distinct internal mechanisms, at the New England Mechanistic Interpretability (NEMI) workshop! See our abstract on its website.
Apr 06, 2026 I gave a talk titled "System identifiability in artificial and biological neural circuits" at the Emergence of Intelligent Phenomena journal club at Harvard & the Kempner Institute!
Feb 13, 2026 I gave a talk on InputDSA at the Harvard Medical School Friday Seminar Series!
Jan 26, 2026 Our InputDSA was accepted to ICLR 2026!
Dec 21, 2025 🎉 Our InputDSA was selected for a talk at COSYNE 2026 (~2.5%)! If you wanna chat about neural dynamics, interpreting RL agents, representional alignment, learning dynamics and curriculum at the conference, please reach out! 🇵🇹
Dec 01, 2025 In San Diego for NeurIPS 2025, presenting our Degeneracy in RNNs work at the main conference, and as a Spotlight talk at the NeurReps workshop!
Oct 29, 2025 Our new paper InputDSA: Demixing then Comparing Recurrent and Externally Driven Dynamics is now on arXiv! Thanks to my amazing collaborators from the Rajan and Fiete Group, especially my co first-author Mitchell Ostrow.
Sep 28, 2025 ✨ Our paper Measuring and Controlling Solution Degeneracy across Task-Trained Recurrent Neural Networks has been accepted as a Spotlight at NeurIPS 2025!