.png){kind=icon}
.png){kind=icon}
Return
FuseMind: Fusing reflection and prediction elevates agent’s reasoning capabilities
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.131755.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Large language models (LLMs) are being increasingly utilized as goal-oriented agents to interact with various external environments, such as games, compilers, and APIs. Nevertheless, these language agents still face difficulties when it comes to long-term reasoning and planning. In this paper, we present FuseMind, a novel agent framework that integrates the Reflection Module and the Prediction Module to enhance the reasoning capabilities of large language model (LLM) agents. FuseMind consists of two key modules: (1) the Reflection Module, which enables agents to process task feedback through verbal reflection and store their insights in an episodic memory buffer, facilitating improved decision-making in subsequent interactions and efficient learning from past experiences; and (2) the Prediction Module, which encourages agents to reconcile discrepancies between forecasts and actual outcomes, thereby broadening their reasoning scope and promoting more strategically focused thought processes. Experimental results demonstrate that FuseMind achieves significant performance gains over strong baseline agents on a variety of challenging tasks. Ablation studies further reveal that FuseMind consistently improves LLM agent performance by enhancing both reasoning diversity and strategic orientation.
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W
View more organizations