Learning Implicit Bias in Generative Spaces for Accelerating Protein Dynamics Emulation
Kaihui Cheng, Zhiqiang Cai, Wenkai Xiang, Zhihang Hu, [3pt] Siyu Zhu, Tzuhsiung Yang, Yuan Qi, [6pt] Fudan University, [2pt] Corresponding authors · 2026
Read Learning Implicit Bias in Generative Spaces for Accelerating Protein Dynamics Emulation by Kaihui Cheng, Zhiqiang Cai, Wenkai Xiang, Zhihang Hu, [3pt] Siyu Zhu, Tzuhsiung Yang, Yuan Qi, [6pt] Fudan University, [2pt] Corresponding authors (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-01833.txt.
Description
Cheng and co-authors speed up fake protein motion from learned models. Generative emulators mimic protein dynamics at low cost. But they inherit the spread of their training data. They tend to revisit known shapes, not rare ones. Long runs make that bias worse.
Their method adds a history-aware bias in the model's space. The bias pushes sampling away from past shapes. A support term keeps new shapes in valid zones. A repair step projects stray samples back to real shapes. Tests use the DynamicPDB-80 data set. The authors report diversity up by 35 percent. On 12 fast-folding proteins, coverage comes far faster. Gains reach up to 37 times with the repair step.
Catalog record
| CREATOR | Kaihui Cheng, Zhiqiang Cai, Wenkai Xiang, Zhihang Hu, [3pt] Siyu Zhu, Tzuhsiung Yang, Yuan Qi, [6pt] Fudan University, [2pt] Corresponding authors |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 39 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.01833