Beyond Compression: Training Latent Representations for Stable Long-Horizon Rollout in Neural Surrogate Solvers
Research on improving long-horizon rollout stability in neural surrogate solvers, reducing error by 40% while matching full-resolution models on physics benchmarks.
Research on improving the stability of long-horizon rollout in neural surrogate solvers through training latent representations, with results showing significant error reduction and performance improvement on physics benchmarks.
- Research on training latent representations to improve long-horizon rollout stability in neural surrogate solvers
- Interventions reduce long-rollout error by approximately 40% while matching full-resolution models on physics benchmarks
- Approaches require 2 orders of magnitude fewer floating point operations and half the GPU memory
- Applied to mesoscale crystal-plasticity simulations of high-cycle fatigue
- Products
- Neural Surrogate Solvers
- Countries
- China
- AI models
- Neural Surrogate Solvers
From the original article
1 min read at arxiv.orgLatent neural surrogate solvers, or latent dynamics models, accelerate simulations of time-dependent physical systems by evolving a compressed latent space rather than resolving full-resolution fields directly. In principle this reduces computational cost and simplifies learning, but in practice errors often accumulate rapidly during long autoregressive rollouts, limiting predictive utility.
By Andreas E. Robertson · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.