THE LIVING ATLAS
Back to the discovery

Source record · 2021

Characterizing possible failure modes in physics-informed neural networks

Aditi S. Krishnapriyan, Amir Gholami, Shandian Zhe, Robert M. Kirby and Michael W. Mahoney

Read the original Publication page
Type
Academic experimental research
Publication
NeurIPS
Version used
Published article; linked access version identified below
Source date
2021
Recorded in Atlas
23 October 2026
License
Original source rights apply; Atlas summaries and diagrams are original.

Publication date; this is not a date of an Atlas replication. No separate validity period has been established for the finding.

How Atlas uses this source

Can AI discover something new?

Sections 2–4: failure modes, loss landscapes, curriculum and sequence learning

The study tests selected convection, reaction and diffusion problems.

Open the source passage
Supports this claim:Physics-informed training can still fail