🤖 AI Summary
This work addresses the lack of an efficient and flexible JAX-native framework for neural operators and physics-informed foundation models. We propose a unified JAX-based training library that constructs an end-to-end differentiable programming model through symbolic tracing, encompassing domain specification, model invocation, residual computation, supervised loss, and diagnostics. The framework tightly integrates automatic differentiation, compilation optimizations, and physics-constrained modeling, enabling seamless composition of multiple models, fine-grained parameter-level control, and hyperparameter tuning. Crucially, it allows effortless switching among operator regression, mesh-aware residual evaluation, and PDE-constrained training paradigms without code refactoring, substantially enhancing development efficiency and training flexibility. A complete implementation is publicly released.
📝 Abstract
jNO (jax Neural Operators) is a JAX-native library for neural operators and foundation models with unified support for both data-driven and physics-informed training. Its core design is a tracing system in which domains, model calls, residuals, supervised losses, and diagnostics are written in one symbolic language and compiled into one optimization pipeline. This allows users to move between operator regression, mesh-aware residual evaluation, and PDE-constrained training without restructuring the surrounding code. jNO also supports multi-model compositions, fine-grained control at parameter level (model, optimizer, and learning rate), hyperparameter tuning, and JAX-native workflows for translated PDE foundation-model families. The source repository is available at https://github.com/FhG-IISB/jNO.