Yifan Zhang

Looping Transformers can achieve better performance with shared parameter updates, revealing that scaling rules must adapt to parameter visits for stable recurrent depth.

RefineEvo transforms heuristic design from static trial-and-error into a dynamic, experience-driven process that significantly boosts solution quality and efficiency.

SCOPE's structured feedback mechanism boosts code generation accuracy, achieving a notable 39.4% pass rate on LiveCodeBench V6—outperforming existing methods by a significant margin.

SearchEyes achieves state-of-the-art performance in multimodal search by unifying training data, environments, and rewards into a cohesive simulated world.

Explicitly modeling rare extreme events in time series forecasting can significantly enhance predictive accuracy, as shown by Exformer's superior performance over traditional models.

Transforming continuous geometric signals into structured discrete tokens leads to state-of-the-art performance in multi-animal tracking, even in the most challenging scenarios.

Zero-shot reward function design using CRWM cuts down design latency while achieving state-of-the-art performance in robotic skill acquisition.

VERITAS reveals that incorporating verifier feedback can boost theorem proving success rates by over 10%, challenging the efficacy of traditional binary pass/fail approaches.

Fine-tuning on the new OmniVideo-100K dataset boosts model performance by over 20% in audio-visual reasoning tasks, revealing the power of structured scripts in enhancing multimodal understanding.

State-of-the-art performance in social intelligence reasoning is achieved by ensuring long-tail events are prioritized over head events through innovative knowledge distillation techniques.

LLMs still have a long way to go when it comes to interactive problem-solving, as revealed by a new benchmark that tests reasoning under budget constraints.