Summary
A forum post on zhichai.net shares the paper 'Eradicating Negative Transfer in Multi-Physics Foundation Models via Sparse Mixture-of-Experts' (arXiv 2605.15179) by Ellwil Sharma and Arastu Sharma. The paper addresses negative transfer in Scientific Machine Learning (SciML) foundation models, where co-training on disparate PDE regimes—such as broadband open-channel fluid dynamics and boundary-dominated porous media flows—causes gradient conflict, unstable optimization, and plasticity loss in dense neural operators. The authors propose Shodh-MoE, a sparse-activated latent transformer architecture for multi-physics transport. It operates on compressed 16^3 physical latents from a physics-informed autoencoder featuring an intra-tokenizer Helmholtz-style velocity parameterization that restricts decoded states to divergence-free velocity fields. The post includes metadata, an AI-translated Chinese abstract, and the original abstract excerpt.
Paper Overview
Field: ML
Authors: Ellwil Sharma, Arastu Sharma
Published: 2026-05-14
arXiv: 2605.15179
Original Abstract (excerpt)
Scaling Scientific Machine Learning (SciML) toward universal foundation models is bottlenecked by negative transfer: the simultaneous co-training of disparate partial differential equation (PDE) regimes can induce gradient conflict, unstable optimization, and plasticity loss in dense neural operators. In particular, broadband open-channel fluid dynamics and boundary-dominated porous media flows impose incompatible spectral and geometric demands on a single dense parameter path.
We introduce Shodh-MoE, a sparse-activated latent transformer architecture for multi-physics transport. Shodh-MoE operates on compressed 16^3 physical latents produced by a physics-informed autoencoder with an intra-tokenizer Helmholtz-style velocity parameterization, restricting decoded states to divergence-free ve... *(abstract truncated in source)*
---
*Auto-collected on 2026-05-15.*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/177620065