Code

Multi-step Multilingual Adaptation closes the reasoning-quality gap between an LLM's training language and lower-resource target languages.

J. Hwang, T. Kumar, S. Lee, A. Agrawal, H. Palangi, A. Kumar, I. Fiete+, P. Liang+. Learn globally, speak locally: Bridging the gaps in multilingual reasoning. arXiv preprint (2025).

Multi-region neural network probing how the hippocampus supports spatially-embedded decision making in rodents and primates.

Yi Xie, J. Hwang, C. Brody, D. Tank, I. Fiete. A Multi-Region Brain Model to Elucidate the Role of Hippocampus in Spatially Embedded Decision-Making. ICML (2025).

Fragmentation-and-Recall (FAR) maps for efficient exploration of large spaces, drawing on observed fragmentation of grid-cell maps in compartmentalized environments.

J. Hwang, Z. Hong, E. Chen, A. Boopathy, P. Agrawal, I. Fiete. Grid Cell-Inspired Fragmentation and Recall for Efficient Map Building. TMLR (2024).

Sequential Morris water-maze task for studying rapid learning without catastrophic forgetting in artificial agents, with reference RL agents that solve it.

R. Wang, J. Hwang, A. Boopathy, I. Fiete. Rapid Learning without Catastrophic Forgetting in the Morris Water Maze. ICML (2024).

Computes the exact inductive bias of common ML architectures, extending the generalization-difficulty framework.

A. Boopathy, W. Yue, J. Hwang, A. Iyer, I. Fiete. Towards Exact Computation of Inductive Bias. IJCAI (2024).

Applies the FAR fragmentation-and-recall idea to curiosity-driven reinforcement learning, mitigating catastrophic forgetting in long-horizon tasks.

J. Hwang, Z. Hong, E. Chen, A. Boopathy, P. Agrawal, I. Fiete. Neuro-Inspired Fragmentation and Recall to Overcome Catastrophic Forgetting in Curiosity. arXiv preprint (2023).

Compares dynamical systems using Koopman operators; input-driven (InputDSA) variants are also included. Works on arbitrary time series, with or without input.

M. Ostrow, A. Eisen, L. Kozachkov, I. Fiete. Beyond Geometry: Comparing the Temporal Structure of Computation in Neural Circuits with Dynamical Similarity Analysis. NeurIPS (2023).

Model-agnostic measure of how hard a learning task is for a given hypothesis class, derived from the Kolmogorov complexity of the optimal predictor relative to the inductive bias.

A. Boopathy, K. Liu, J. Hwang, S. Ge, A. Mohammedsaleh, I. Fiete. Model-agnostic measure of generalization difficulty. ICML (2023).

Characterizes manifolds and performs unsupervised decoding in the head direction system. Python 2.7; please cite the associated paper if you use it.

R. Chaudhuri, B. Gercek*, B. Pandey*, A. Peyrache, I. Fiete. The intrinsic population dynamics of a canonical cognitive circuit. Nature Neuroscience (2019).

Spike-time-dependent plasticity and heterosynaptic competition spontaneously organize networks to produce long, sparse activity sequences — whether the training inputs are sequential or not. Matlab; aligns with Fig. 2 (summed weight bound).

I. Fiete, W. Senn, C. Wang, R. H. R. Hahnloser. Spike time-dependent plasticity and heterosynaptic competition organize networks to produce long scale-free sequences of neural activity. Neuron (2010).

Simulates grid-cell activity from continuous-attractor dynamics, with periodic or aperiodic boundary conditions. Runs on random or recorded rat trajectories (included).

Y. Burak, I. Fiete. Accurate path integration in continuous attractor network models of grid cells. PLOS Computational Biology (2009).

All newer open-source releases tied to individual papers. Each repo has its own README with run instructions.