← Back to feed

Data Free Training of Diffusion Models for Lattice Field Theories

L5 · ResearcherResearcharXiv· 10/13/2026

Addresses fundamental sampling bottlenecks in computational physics with novel ML training approaches

AI Summary

Researchers developed a data-free method for diffusion models in lattice field theories using self-generated samples and target score identities.

Excerpt

Sampling equilibrium field configurations remains a major computational bottleneck in lattice field theory, motivating the development of generative samplers that can produce independent configurations. Diffusion models are promising for this purpose, but their standard denoising score matching objective requires samples from the target distribution, seemingly reintroducing the original sampling problem. Here, we show that diffusion models can instead be trained from their own generated samples

Read Original
0 upvotes · 0 downvotes · 1 min read

Related Articles

L5 · ResearcherResearcharXiv
Tev: Scaling Long-Horizon Agentic Environments with Dense Progress Rewards

Researchers introduce Tev, a novel training framework with dense progress rewards that enables 27B parameter models to match GPT-6 and Claude Opus 5 performance on long-horizon agentic tasks.

L5 · ResearcherResearcharXiv
AesCode: Aesthetic Code Generation with Decoupled Cross-Modal Rewards

Researchers introduce AesCode, a method that improves aesthetic code generation for visual artifacts by using decoupled cross-modal rewards from both code verifiers and visual language models.

L5 · ResearcherResearcharXiv
ChunkAct: Action Chunking for Agentic Robotics

ChunkAct introduces action chunking to agentic robotics, reducing LLM calls by 2.9x while improving task success by 5.0 points on LIBERO-Pro benchmark.

L5 · ResearcherResearcharXiv
GUIDE: Computationally Augmented Consensus for Trustworthy Clinical Practice Guidelines

GUIDE is a multi-agent AI system that automates clinical guideline generation, achieving 88.85% literature recall and reducing expert workload by over 70% while maintaining expert-quality recommendations.

L5 · ResearcherResearcharXiv
What Matters in Quality-Aware VLA Pretraining: A Benchmark and Empirical Study

Researchers introduce AXIS-Bench for evaluating quality-aware vision-language-action pretraining methods, showing that episode conditioning improves robot success rates from 9.6% to 31.7% in real-world experiments.