Tag
该论文提出一个多目标对齐的小语言模型框架，用于生成物质使用障碍（SUD）咨询中的患者对话，结合知识蒸馏、人类标注偏好优化和注意力引导的奖励塑形，以提升认知一致性与临床真实性。
The paper proposes using Generative Flow Networks (GFlowNets) to generate diverse synthetic expert conversations for post-training LLMs, avoiding the mode collapse seen in prompting and reward-based synthesis, with stronger downstream outcome-prediction results across tutoring and emotional-support dialogues.
Introduces ODRA, a framework for synthesizing Cognitive Behavioral Therapy dialogues using structured Chain-of-Thought and a resistance orchestrator to model patient behavior, achieving higher clinical fidelity and downstream robustness.
The paper proposes EmoTrace, a multi-turn dialogue generation framework for psychological support that models seekers' emotional trajectories to improve empathy and emotional richness in counselor responses, outperforming existing methods.
Dia is a 1.6B-parameter open-source text-to-speech model that generates English dialogue from transcripts, supporting two-speaker generation, audio conditioning, and nonverbal cues.