Yong-Liang Shen

Even the best vision-language models struggle with predictive spatial reasoning, scoring less than 60% on a new benchmark designed to test anticipating unseen outcomes.

Visual-native skill cards, directly accessible to VLMs, unlock a self-reinforcing loop of policy improvement and skill accumulation, drastically boosting sample efficiency.

Explicit 3D geometry, rendered as text, unlocks superior spatial reasoning in RGB-only VLMs by guiding trajectory correction during distillation.

Computer-use agents can bypass brittle, hand-engineered APIs entirely by learning when to substitute mouse clicks with native shell commands, yielding a 14.8-point leap on OSWorld.

This work proposes RetireOPD (Self-Retiring On-Policy Distillation), which first optimizes a decoupled, skill-conditioned teacher with environment rewards and then trains a skill-free student jointly with RL and OPD.

This work proposes EvoSkill-GUI, a training-free framework in which each skill is a structured multi-file package containing retrieval metadata, executable plans, backup localization, failure-recovery rules, accessibility utilities, and failure cases.

Standard translation metrics and 235B-parameter LLM judges actively reward the worst social media translations due to cultural blindness, but progressively decaying masked cultural scaffolding enables an 8B model to rival Gemini-3.1-Pro.

PaperGym achieves a remarkable 73.48 on ResearchQA, outperforming larger models and redefining how AI can generate and evaluate research plans.

TTPO achieves label-supervised performance without any ground-truth labels, outperforming traditional methods on key benchmarks.

Generating over 203,000 unique web interaction trajectories, BrowserForge significantly boosts model performance on real-world tasks by leveraging the vastness of the open web.

Gaussian guidance reveals that optimal trajectory retention depth varies across tasks, leading to a substantial performance boost in reinforcement learning.

Recursive belief updates in AgentOPSD reveal pivotal decision points, leading to a 89.1% success rate on complex RL tasks.

SkillRise achieves up to 8.5 percentage points better performance than leading methods by effectively reusing transferable skills across related tasks.

Relay-OPD not only corrects reasoning missteps in real-time but also slashes training time by over half, setting a new standard for on-policy distillation efficiency.

VLA-Corrector allows VLA models to adaptively replan actions in real-time, drastically reducing compounding errors in dynamic environments.

Decoupling perception from reasoning in visual tasks leads to a remarkable 93.2% accuracy on V-Star, showcasing a new paradigm for fine-grained visual reasoning.

Current translation metrics miss the mark on capturing the cultural nuances and emotional resonance crucial for effective social media UGC translation.

LLMs can reason better when they're not forced to answer in English, and a new RL method leverages this quirk to boost performance across reasoning tasks.

LLMs can learn to recognize when they lack sufficient information for reasoning and proactively ask for clarification, leading to more reliable and concise answers.

Offloading memory and computation to a copilot lets a 7B parameter GUI agent outperform larger models on long-horizon tasks, suggesting a path to more efficient and capable GUI automation.

Uncertainty-driven zoom-in boosts GUI grounding accuracy by up to 13.4% without any retraining, showing that targeted attention to model uncertainty can significantly improve performance.

Forget noisy pseudo-labels: SpatialEvo unlocks self-supervised 3D spatial reasoning by generating perfectly accurate training data directly from scene geometry.

Finally, a unified open-source framework lets you train, evaluate, and deploy GUI agents across real devices and chat platforms, closing the gap between research and real-world application.

Even frontier models like Claude Sonnet 4.6 stumble when asked to infer user preferences and proactively assist in mobile tasks, achieving less than 50% success despite excelling at explicit task execution.

Multimodal models can "see" the image but still fail at reasoning because the visual input distracts the routing mechanism from activating the right experts.

LLM agents can internalize skills via in-context RL, achieving zero-shot autonomous behavior without the token overhead and retrieval noise of traditional methods.

LLMs can escape the trap of confidently wrong reasoning by co-evolving a generator and verifier from a single model, bootstrapping each other to break free from flawed consensus.

By adversarially co-evolving code and test LLMs, Code-A1 achieves code generation performance on par with human-annotated training, while simultaneously boosting the LLM's ability to find bugs.