Distributional Energy-Based Models for Uncertainty-Aware Structured LLM Reasoning
The paper presents a novel approach to improve the reasoning capabilities of Large Language Models (LLMs) when generating structured outputs. It introduces a decomposed energy function that combines a quality scorer with analytical constraint penalties to verify outputs. The proposed method outperforms existing models on multiple benchmarks, demonstrating significant reductions in constraint violations and improved reasoning accuracy.
- ▪The study proposes a decomposed energy function for verifying structured outputs from LLMs.
- ▪A quality scorer, consisting of low-rank adapters, is used to rank candidates and assess uncertainty.
- ▪The method achieved a 53% reduction in constraint violations compared to existing models on the TravelPlanner benchmark.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Machine Learning arXiv:2605.18871 (cs) [Submitted on 15 May 2026] Title:Distributional Energy-Based Models for Uncertainty-Aware Structured LLM Reasoning Authors:Shireen Kudukkil Manchingal, Abhey Kalia, Fernanda Gonçalves, Shebin Rawther View a PDF of the paper titled Distributional Energy-Based Models for Uncertainty-Aware Structured LLM Reasoning, by Shireen Kudukkil Manchingal and 3 other authors View PDF HTML (experimental) Abstract:When Large Language Models produce structured outputs such as travel plans, code solutions, or multi-step proofs, individual reasoning steps may appear correct while the output as a whole violates budgets, fails test cases, or contradicts earlier deductions.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
