Searcharxiv⌕ Search

arXiv subjects

Maryam Shoaeinaeini

Publications and source records attributed to Maryam Shoaeinaeini.

3 recordsLinked to original sources

Persona-Guided LLM Agents for Task-Oriented Dialogue

Prior work has shown that large language models can express diverse personality traits in open-ended text generation, but it is unclear whether they can do so in goal-directed dialogue without harming task completion, and whether adapting to a user's personality improves the interaction. We study these questions in task-oriented dialogue (TOD) with a training-free framework in which a personality-conditioned user agent interacts with a system agent under three conditions: Neutral (no personality information), Try (adapt to cues observed in the dialogue), and Oracle (the user's trait label is given explicitly). Across GPT-4o, Qwen3-Next-80B, and Gemini 2.0 Flash, two Schema-Guided Dialogue domains, and ten Big Five poles (9,000 dialogues), each scored by two independent LLM judges, user agents express most traits while task performance remains high (89.94% completion, 85.27% success). The exceptions are introversive and closed users: these traits manifest mainly as absent behavior and are rarely detected. Both adaptive conditions improve pairwise user satisfaction and lower truthfulness, revealing a personalization-grounding trade-off, but only cue-responsive Try also improves constraint satisfaction and inform rate. How adaptation is implemented matters: Oracle's gains depend on how visibly the trait is expressed, and for disagreeable users the label leads Qwen3 and Gemini to mirror hostility. Try avoids this failure, its satisfaction gains do not depend on trait visibility, and it is preferred over Oracle overall. A 175-pair human study supports the reliability of the LLM judges: on clear-winner cases, annotators agree with the satisfaction judges more often than with each other (76.3% vs. 65.1% agreement). Our results suggest that responding to observable behavior is a more robust basis for social adaptation than conditioning on psychological labels.

cs.CL↗

Skill-Latent Rewards: LLM-Guided Reward Formulation for Hierarchical Reinforcement Learning

Long-horizon reinforcement learning under sparse and delayed feedback remains an open challenge. Existing return-decomposition methods redistribute episodic rewards to ease credit assignment, but they operate on raw state--action transitions and disregard the subgoal structure that governs meaningful progress in complex tasks. We propose Skill-LR, a hierarchical reinforcement learning framework in which a vision--language model identifies the agent's active subgoal at each timestep and a large language model generates a matching latent reward formulation through a two-stage prompting process. By conditioning reward signals on the currently active subgoal, Skill-LR produces stage-specific credit that evolves with the hierarchical plan, reducing temporal variance and improving sample efficiency. On multi-stage MiniGrid and BabyAI benchmarks, including DoorKey and Unlock-to-Unlock, Skill-LR converges faster and more reliably than sparse environmental rewards and strong credit-assignment baselines such as RUDDER and LaRe, demonstrating that subgoal-aligned latent rewards offer a principled path toward stable, interpretable credit assignment in long-horizon tasks.

cs.LG↗

Guiding Reinforcement Learning Using Uncertainty-Aware Large Language Models

Human guidance in reinforcement learning (RL) is often impractical for large-scale applications due to high costs and time constraints. Large Language Models (LLMs) offer a promising alternative to mitigate RL sample inefficiency and potentially replace human trainers. However, applying LLMs as RL trainers is challenging due to their overconfidence and less reliable solutions in sequential tasks. We address this limitation by introducing a calibrated guidance system that uses Monte Carlo Dropout to enhance LLM advice reliability by assessing prediction variances from multiple forward passes. Additionally, we develop a novel RL policy shaping method based on dynamic model average entropy to adjust the LLM's influence on RL policies according to guidance uncertainty. This approach ensures robust RL training by relying on reliable LLM guidance. To validate our contributions, we conduct extensive experiments in a Minigrid environment with three goals in varying environment sizes. The results showcase superior model performance compared to uncalibrated LLMs, unguided RL, and calibrated LLMs with different shaping policies. Moreover, we analyze various uncertainty estimation methods, demonstrating the effectiveness of average entropy in reflecting higher uncertainty in incorrect guidance. These findings highlight the persistent overconfidence in fine-tuned LLMs and underscore the importance of effective calibration in sequential decision-making problems.

cs.LG↗