AI Paper Atlas

Research map / Large language models / Alignment and preferences

LoopWeaver: Weaving Feedback Loops into Hierarchical Generation under Constraints

Yifan Zhu, Guanting Chen, Bing Wei, Haoran Luo

NeurIPS 2026 · Large language models › Alignment and preferences · approach: Robust and calibrated DPO · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes LoopWeaver, a feedback-loop approach to hierarchical LLM generation for long text under constraints, jointly optimizing global structure, local coherence, and constraint satisfaction.

Problem

Constrained long-text generation with LLMs requires jointly balancing global structure, local coherence, and constraint satisfaction, but existing static or loosely coupled pipelines struggle with cross-stage collaborative optimization.

Approach

Couples hierarchical generation stages through feedback loops so that stages collaboratively optimize rather than running as a static pipeline (abstract is truncated, so the specific mechanism is not stated).

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply LoopWeaver planning + filtering + two-stage DPOQwen2.5-7B-Instruct · LongGenBench-style constraint accuracyavg accuracy: 30.74 -> 48.41Improved
apply LoopWeaverQwen2.5-7B-Instruct · LongGenBench-style text qualityavg score: 71.34 -> 78.77Improved
apply LoopWeaverQwen2.5-0.5B-Instruct · LongGenBench periodic constraintsPeri. accuracy: 15.69 -> 15.80 (CogWriter 17.70)No clear effect
apply LoopWeaverQwen2.5-0.5B-Instruct · LongGenBench affective consistencyAffect. score: 39.86 -> 37.39Hurt
apply CogWriterQwen2.5-0.5B-Instruct · LongGenBench text qualityavg score: 27.49 -> 25.16Hurt
apply LongWriterQwen2.5-7B-Instruct · LongGenBench constraint accuracyavg accuracy: 30.74 -> 29.21No clear effect
add plan-level DPO only (ablation)Qwen2.5-7B-Instruct · LongGenBench constraint accuracyavg accuracy: 30.74 -> 44.20Improved
add generation-level DPO only (ablation)Qwen2.5-7B-Instruct · LongGenBench constraint accuracyavg accuracy: 30.74 -> 44.78Improved
combine both DPO stages and workflow (ablation)Qwen2.5-7B-Instruct · LongGenBench constraint accuracyavg accuracy: 44.78 -> 48.41Improved
combine both stages vs best single stage (ablation)Qwen2.5-7B-Instruct · LongGenBench text qualityavg score: 77.23 -> 78.77Improved
vary reward threshold tau 0.7 -> 0.6 or 0.8 (ablation)reward construction · Fixed-candidate auditselection retained: 97.6% / 99.2% retainedNo clear effect
feasibility filter on plansLlama-3.1-8B · Filter audit 800 candidatesproxy precision/recall: 61.0 / 72.8 percent; false rejection 27.2%Mixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy