Volkan Cevher

This work shows that Kimi Delta Attention (KDA) can realize 2D rotations by combining a single delta-rule transformation with a second reflection supplied by its channel-wise gate, and characterize the expressivity of CKDA and prove that every orthogonal diagonal-plus-rank-one matrix is exactly a CKDA transition matrix.

Circuit discovery in LLMs is fraught with variability, and a new method reveals that this complexity may be an inherent challenge rather than a flaw in existing techniques.

Soft attention-discounted reranking can boost masked diffusion model performance by over 10 percentage points while maintaining efficient decoding.

Squeeze 2x more speed from your conditional flow matching models by optimizing data-noise coupling across minibatches.

Forget state-action spaces: this work achieves efficient multi-agent imitation learning by concentrating on feature-level representations in linear Markov games.

Even perfect imitation of expert behavior in multi-agent settings can lead to highly exploitable policies, unless experts play dominant strategies or exhibit best-response continuity.

Stop wrestling with complex setups: this new benchmark makes evaluating machine unlearning algorithms as easy as running a script.