Tsinghua Huang Gao Team Wins Outstanding Paper Award at ICML 2026; Time-Tested Award Goes to Classic Algorithm A3C

07/06 09:20

On July 6, it was announced that the International Conference on Machine Learning (ICML) 2026 was held in Seoul, South Korea, where the annual award-winning papers were revealed. The paper by Tsinghua University's Huang Gao team in collaboration with Alibaba, titled 'The Flexibility Trap: Rethinking the Value of Arbitrary Order in Diffusion Language Models,' won the Outstanding Paper Award. The research reveals that the flexibility of arbitrary generation order in diffusion language models limits the model's potential in general reasoning tasks such as mathematics and programming. By abandoning arbitrary order and adopting the traditional left-to-right generation, the method not only becomes simpler but also significantly improves reasoning accuracy. Another Outstanding Paper Award was given to researchers from MIT and Yale University for their study proposing a high-accuracy sampling algorithm for diffusion models (High-Accuracy Sampling for Diffusion Models and Log-Concave Distributions), achieving exponential optimization in the number of steps (or sampling complexity) required to reach the target sampling accuracy. Additionally, a position paper titled 'Position: The Alignment Community is Unintentionally Building a Censor’s Toolkit,' co-authored by researchers from Ludwig Maximilian University of Munich and independent researchers, highlighted the dual-use risks of current AI alignment technologies, which can easily be manipulated into tools for information censorship. The Time-Tested Award was presented to the Google DeepMind team for their classic reinforcement learning algorithm 'Asynchronous Methods for Deep Reinforcement Learning,' published in 2016. This research introduced the asynchronous advantage actor-critic (A3C) architecture, significantly enhancing the training efficiency of deep reinforcement learning and marking the beginning of an era where intelligent agents can be efficiently trained using standard multi-core CPUs.

БычийМедвежийЛайкПоделиться

Заявление об освобождении от ответственности：Материалы выше не представляют собой позицию HTX.，HTX не дает никаких торговых рекомендаций.。