This paper presents an any-dimensional PDE solver that trains machine learning models in lower dimensions and applies them directly to higher-dimensional PDEs in a zero-shot manner, under mathematical conditions based on symmetries.
See LICENSE filePDE solverany-dimensional learningzero-shot transfer
This paper studies out-of-distribution regression in scientific machine learning, focusing on conditions for stable extrapolation in high-dimensional function and operator learning. It analyzes how assumptions on target functions or operators affect extrapolation beyond the training domain and reviews existing bounds based on distribution-discrepancy penalties.
See LICENSE filescientific machine learningout-of-distribution generalizationoperator learning
This paper studies preconditioned training for physics-informed neural operators across general physics simulations, using governing equations rather than solver-generated datasets to reduce offline data-generation cost and enable fresh samples during optimization.
See LICENSE filephysics-informed neural operatorpreconditioned trainingscientific machine learning
SHRAV is an architecture-independent computational framework for physical modeling and inverse design, featuring a state-continuation core with declared reuse boundaries and explicit roles for learned evolution.
See LICENSE fileinverse designphysical modelingstate continuation
Investigates the relationship between residual scale and learnability in Δ-learning, demonstrating through molecular graph neural network evaluation on total energy targets that smaller residual targets do not necessarily yield better learning outcomes.
See LICENSE fileDelta learningresidual learningmolecular graph neural network
GravINNs embeds post-Newtonian equations of motion directly into neural network training to learn binary orbital dynamics, including single-orbit surrogates and extensions across a four-dimensional parameter space.
See LICENSE filegravity-informed neural networkspost-Newtonian approximationbinary dynamics
PosteriorBench introduces a benchmark for evaluating how well generative inverse solvers capture true posterior distributions in scientific inverse problems, addressing limitations of pointwise accuracy metrics.
See LICENSE fileinverse problemsgenerative modelsuncertainty quantification
A 36M-parameter neural network solver for partial differential equations based on U-Net architecture, enabling data-driven solving and prediction of PDE systems