Kai-Wei Chang

Uncovering crucial supporting lemmas for theorem proving by traversing multi-hop dependencies in a formal library's declaration graph.

Long-horizon agents can now achieve GPT-6 Astra and Claude Opus-level performance on complex tasks by learning from dense, progressive rewards derived from latent world specifications.

A simple post-hoc geometric calibration method recovers lost out-of-domain capabilities and boosts in-domain performance after SFT, even on large, diverse models.

AudioICL-Bench is introduced, a diagnostic benchmark whose per-episode rules are resampled so that no correct answer is recoverable from prior knowledge, and its nine tasks are organized along two axes that separate what must be learned from demonstrations from what must be perceived in the signal, enabling failures to be attributed to either source.

Frontier multimodal models that master complex single-image visual reasoning collapse to sub-35% accuracy when asked to detect basic low-level noise and texture differences between two images.

Sparsifying communication topologies in LLM multi-agent systems can lead to significant efficiency gains without sacrificing accuracy, challenging conventional design approaches.

Federated learning can outperform centralized models in multi-agent safety without compromising data privacy, achieving a remarkable 43% reduction in attack success rates.

Multi-agent refinement can enhance diagram quality across multiple iterations, countering common pitfalls like quality drift and forgetting.

OpenVLThinkerV2 leapfrogs existing open-source and proprietary multimodal models by using a novel Gaussian-based RL objective that ensures gradient equity across diverse visual tasks.

Ditch the polarity labels: SemEval-2026's DimABSA task reveals how modeling sentiment along valence-arousal dimensions unlocks nuanced understanding in both aspect-based sentiment analysis and stance detection.

Forget hand-designed agent communication topologies: Agent Q-Mix learns decentralized communication strategies that boost accuracy and token efficiency in LLM multi-agent systems.

Scaling VLMs won't magically unlock reasoning skills; you need to address the reporting bias in training data that suppresses tacit information.

Forget hand-annotated data: Magnet distills multi-turn tool-use skills into LLMs by automatically generating training trajectories that outperform even Gemini 1.5 Pro.