Tag
RL-ADA is a co-evolutionary framework that replaces human labels with world feedback for training adversarially robust enterprise dialogue agents, using reinforcement learning to improve performance without annotation data.
This paper investigates failures in a 2B model for dialogue games and introduces a diagnosis-guided post-training recipe using SFT, DPO, and LoRA to boost performance while maintaining general capabilities.
CoreMem proposes a resource-efficient edge-cloud memory architecture for dialogue agents, using Riemannian retrieval with a Fisher-Rao metric and Fisher-guided discrete token distillation to achieve strong accuracy improvements within an 8 GB VRAM budget.
G-Long proposes a graph-enhanced memory management framework for long-term dialogue agents, using a fine-tuned small language model for structured triplet extraction and associative retrieval, achieving state-of-the-art performance in response generation and memory retrieval with reduced computational overhead.
This paper theoretically identifies and mitigates context distribution shift in multi-turn dialogue RL, proposing Calibrated Interactive RL that couples interactive RL with simulator alignment to reduce the sim-to-real gap and achieve state-of-the-art performance.
SAVOIR framework applies cooperative game theory and Shapley values to train language agents with improved social intelligence, achieving SOTA on SOTOPIA benchmark and matching GPT-4o performance.