Adapt Language Agent to Different Task via Automatic Mechanism Activation
Abstract: Language Agent (LA) could be endowed with different mechanisms for autonomous task accomplishment. Current LAs typically rely on fixed mechanism or a set of mechanisms activated in a predefined order, limiting their adaptability to varied potential task solution structures. To this end, this paper introduces Unify agent mechanisms by Actions (UniAct), a unified agent that integrates different mechanisms. Additionally, we propose Automatic Language Agent Mechanism Activation Learning with Self-Exploration (ALAMA), which focuses on optimizing mechanism activation adaptability without reliance on expert models. By leveraging self-generated UniAct trajectories with different rewards, ALAMA enables the agent to adaptively activate mechanisms that may result in high downstream task rewards based on the potential characteristics of the task. Experimental results demonstrate significant improvements in downstream agent tasks, affirming the effectiveness of our approach in facilitating more dynamic and context-sensitive mechanism activation.
Paper Type: Long
Research Area: Question Answering
Research Area Keywords: math QA,open-domain QA
Contribution Types: NLP engineering experiment, Publicly available software and/or pre-trained models
Languages Studied: English
Submission Number: 2451
Loading