llm-kb
← Back to research
Paper

New paper shows score matching can fail to ensure stable reverse-time diffusion sampling

A new arXiv paper demonstrates that small forward-marginal error from score matching does not guarantee numerical stability of the reverse-time diffusion sampler. The authors construct a score field with arbitrarily small forward-marginal L2 error whose Euler–Maruyama discretizations diverge in every positive moment, despite the continuous reverse process being well-behaved. This highlights a gap between score matching objectives and practical sampling stability.

0 engagement·1 source·Thu, Jul 9, 2026, 05:55 PM
The paper, titled 'Score Accuracy Along the Forward Diffusion Does Not Certify Numerical Stability in Diffusion Sampling,' posted on arXiv on 2026-07-09, shows that score matching controls average error under the forward marginals, but a discretized reverse-time sampler evaluates the learned score along its own trajectory. The authors construct a single smooth score field with arbitrarily small forward-marginal L2 error. The learned reverse-time process is nonexplosive, has moments of every order, and can be arbitrarily close to the exact reverse-time process in path-space total variation. Yet its Euler–Maruyama discretizations converge in probability while every positive moment diverges. This result is significant for practitioners using diffusion models, as it implies that low score-matching loss does not guarantee stable sampling, potentially leading to divergent sample quality or numerical issues in practice.

Entities

Score matching(concept)Reverse-time diffusion sampler(concept)Euler–Maruyama discretization(concept)Score Accuracy Along the Forward Diffusion Does Not Certify Numerical Stability in Diffusion Sampling(paper)score matching(concept)reverse-time diffusion sampler(concept)

Related

PaperThu, Jul 9, 2026, 05:26 PM

Researchers propose memory agent to combat behavioral state decay in long-horizon tasks

A new arXiv paper identifies 'behavioral state decay' as a failure mode where decision-relevant information gets buried in long trajectories. The authors propose a separate memory agent that actively updates a structured memory bank alongside an unmodified action agent, rather than relying on passive retrieval.

0 engagement·1 source·arxiv
arXiv
CommunitySun, Jul 12, 2026, 04:22 PM

Community observes stalled progress on Deep Research products since 2025 launch

A Reddit discussion notes that Deep Research products, which launched in February 2025 as a step change, have seen only incremental improvements since. Known weaknesses like hallucinated facts and poor uncertainty calibration persist in benchmarks over a year later, requiring users to verify outputs and reducing time savings.

43 engagement·1 source·reddit
PaperThu, Jul 9, 2026, 05:01 PM

arXiv paper benchmarks LLM judges for citation quality in deep-research systems

A new arXiv paper studies the calibration of LLM judges used as reward models in reinforcement learning for citation quality in deep-research systems. The work evaluates how capable and biased an LLM judge must be to reliably score rubric criteria like source relevance and factual support for attribution-citation pairs. This matters for practitioners building RL-based systems that depend on automated citation verification.

0 engagement·1 source·arxiv
arXiv
PaperTue, Jul 7, 2026, 05:43 PM

Researchers identify modality interference as root cause of full-duplex SLM degradation

A new paper on arXiv (July 7, 2026) presents a fine-grained analysis of optimization dynamics in full-duplex Spoken Language Models (SLMs), identifying severe modality interference as the root cause of knowledge degradation and compromised semantic integrity. The work aims to enable more natural and intelligent full-duplex SLMs.

0 engagement·1 source·arxiv
arXiv
PaperTue, Jul 7, 2026, 05:09 PM

DynaKRAG: A unified framework for learnable evidence control in multi-hop RAG

Researchers propose DynaKRAG, a framework that learns a state-conditioned policy to dynamically choose among evidence operations (retrieval, reformulation, critique, sufficiency checking) in multi-hop retrieval-augmented generation. This moves beyond fixed pipelines, potentially improving accuracy and flexibility in complex QA tasks.

0 engagement·1 source·arxiv
arXiv