Building tools to evaluate learning systems on dynamical prediction and identification tasks.
This line of work develops benchmarks and perturbation-based tests for forecasting and system identification models, with an emphasis on how scaling and generalization behave in dynamical settings. The work is centered on DynaDojo and is something I’ve been building with Max Kanwal.
@article{bhamidipaty2023dynadojo,project={dynadojo},publication_sort={2023-02},title={Dynadojo: An Extensible Platform for Benchmarking Scaling in Dynamical System Identification},author={Bhamidipaty, Logan M and Bruzzese, Tommy and Tran, Caryn and Ratl Mrad, Rami and Kanwal, Maxinder S},journal={Advances in Neural Information Processing Systems},volume={36},pages={15519--15530},url={https://proceedings.neurips.cc/paper_files/paper/2023/file/32093649cbbcff773d9a991d8c30a7fe-Paper-Datasets_and_Benchmarks.pdf},year={2023},month=dec,}
@inproceedings{kanwal2025measuring,project={dynadojo},publication_sort={2025-01},title={Measuring Memorization and Generalization in Forecasting Models via Structured Perturbations of Chaotic Systems},author={Kanwal, Max and Tran, Caryn},booktitle={ICML 2025 Workshop on Methods and Opportunities at Small Scale},url={https://openreview.net/pdf?id=DHo0eqNkvm},year={2025},month=jan,}
Advances in Neural Information Processing Systems, Dec 2023
In ICML 2025 Workshop on Methods and Opportunities at Small Scale, Jan 2025