Jie Fu

Finite populations of MDP agents can now be controlled under stochastic reach-avoid chance constraints by directly managing the variance of their empirical density.

Nonlinear recurrent memories struggle with sequential test-time training, but layerwise delta updates unlock significant gains in language modeling and retrieval.

Iterative refinement of latent states using a small recurrent reasoner can drastically enhance reasoning performance in frozen LLMs, outperforming traditional methods at a fraction of the compute cost.

SWD achieves high-fidelity circuit extraction with less than 1% of the data used by traditional methods, revolutionizing interpretability in pretrained transformers.

Circuit extraction claims can be misleading, as the perceived mechanisms behind model behavior are heavily influenced by how circuits are reported and compared.

Location-based services are vulnerable to privacy breaches, but a new differential privacy framework can safeguard kNN queries without losing utility.