CLSGen: A Dual-Head Fine-Tuning Framework for Joint Probabilistic Classification and Verbalized Explanation

[论文] CLSGen: A Dual-Head Fine-Tuning Framework for Joint Probabilistic Classification and Verbalized Explanation ## 论文概要 **研究领域**: cs.CL **作者**: WonJin Yoon, Kangyu Zhu, Ian Bulovic, Autumn Sehy, Yanjun Gao, Dmitriy Dligach, Majid Afshar, Timothy A. Miller **发布时间**: 2026-04-13 **arXiv**: [2604.11801](https://arxiv.org/abs/2604.11801) ## 中文摘要 随着大语言模型的最新进展，人们越来越有兴趣将这些模型应用于解决复杂问题。现代LLM能够处理长上下文并生成语言化解释，在解决实际应用方面具有巨大潜力。然而，将LLM部署用于实际决策的一个关键障碍是它们无法提供可靠的定量概率。本文提出CLSGen，一种用于二分类任务的新型LLM微调框架。该框架包含新的模型架构、训练方法和数据构建策略，能够在不牺牲模型固解释生成能力的情况下实现稳健的概率估计。实验表明CLSGen在分类指标上优于现有基线。 ## 原文摘要 With the recent progress of Large Language Models (LLMs), there is a growing interest in applying these models to solve complex and challenging problems. Modern LLMs, capable of processing long上下文和generating verbalized explanations, offer significant potential in addressing real-world applications. However, a critical hurdle in deploying LLMs for practical decision-making is their inability to provide reliable, quantitative probabilities. --- *自动采集于 2026-04-15* #论文 #arXiv #AI #小凯