CoRoPINN: Cognitive region optimized physics-informed neural networks

Jingcong Li, Yu Liao, Ximeng Wang, Jie Deng · 2026

Read CoRoPINN: Cognitive region optimized physics-informed neural networks by Jingcong Li, Yu Liao, Ximeng Wang, Jie Deng (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/plos-pone-0358646.txt.

Description

Jingcong Li and three colleagues improve the neural networks that solve physics equations by teaching them where to concentrate, in PLOS ONE in 2026. Physics informed networks learn a solution while respecting the governing equation at sampled points, but they spread effort evenly and under learn the hard regions. Their system closes a feedback loop. The equation's own error at each point estimates how difficult that point is. That estimate stretches the physical constraint around hard points. It also shifts training weight toward them.

Tested across five network designs on three one dimensional problems, the method took the lowest error on every benchmark. It cut the runner up method's aggregate error by roughly a fifth to three fifths, depending on the problem. Higher dimensional and fluid test cases followed. The principle is portable: a solver that notices its own confusion can spend its budget where confusion lives.

Catalog record

CREATORJingcong Li, Yu Liao, Ximeng Wang, Jie Deng
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE65320
LICENSECC_BY
RIGHTSCC BY
SOURCEPLOS

Source: https://doi.org/10.1371/journal.pone.0358646

Browse the full library · About the Tower of Babel