Tong Zhang

AsyncWebRL achieves a staggering 2.9× increase in training throughput while setting a new state-of-the-art performance for web agents on challenging tasks.

Fewer than two edits can enable 76.2% of unsafe images to bypass safety classifiers while retaining their malicious intent, exposing critical vulnerabilities in current moderation systems.

Verified workflows in Lean4Agent outperform unverified ones by nearly 12%, showcasing the power of formal methods in enhancing LLM agent reliability.

OpenWebRL-4B sets a new benchmark for open-source visual web agents, achieving impressive success rates with minimal initial data while outperforming larger-scale competitors.

FLAME uncovers a hidden statistical energy gap in AI-generated images, enabling precise localization of forgeries that traditional methods miss.

One model to control them all: Qwen-VLA achieves impressive zero-shot generalization across diverse robotic tasks and embodiments by unifying vision-language-action modeling.

Stop hand-crafting pseudo-labels: this framework learns to generate and select them for semi-supervised segmentation, boosting performance on RefCOCO, RefCOCO+, and RefCOCOg.

Forget patch-based image tokenization: channel-wise quantization unlocks better codebook utilization and text-to-image generation by representing images as discrete levels of visual detail.

Achieve state-of-the-art low-light image enhancement with real-time inference using an extremely lightweight and unsupervised framework.

Looping language models isn't just for single agents anymore: Recursive Multi-Agent Systems (RecursiveMAS) show that agent collaboration itself can be scaled through recursion, yielding faster and more efficient problem-solving.

GUI agents can achieve significantly stronger task-solving capabilities through carefully designed post-training and data curation, without relying on costly online data collection.