analiza aiqbench
The Pre-Reasoning Perception Framework (PRPF) introduces a two-stage architecture that separates intervention gating from assistance generation to address inefficiencies in current unified MLLM pipelines. By activating the heavy reasoning model only when necessary, this approach significantly reduces false trigger rates and redundant inference while improving success rates on mobile benchmarks. This method offers a practical pathway for optimizing proactive mobile agents without compromising their ability to assist users effectively.
Perceive Before Reasoning: A Pre-Reasoning Perception Framework for Efficient and Reliable Proactive Mobile Agents
arXiv:2606.03236v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) have substantially advanced mobile agents, yet proactive mobile assistance remains challenging because agents must decide \emph{when} to intervene before determining \emph{how} to assist. Existing systems often implement these two decisions within a unified MLLM-based pipeline, leading to goal misalignment between conservative intervention filtering and comprehensive assistance generation, as well as redundant inference when the agent should remain silent. To address these limitations, we propose the \textbf{Pre-Reasoning Perception Framework (PRPF)}, a two-stage framework built on perceiving before reasoning. PRPF introduces a lightweight Multimodal Proactive Perceptor (MPP) for intervention gating and context compression, and activates the Proactive Agent Reasoner (PAR) only when intervention is warranted. Experiments on the ProactiveMobile benchmark show that PRPF substantially reduces false trigger rates (FTR) while improving success rates (SR) and inference efficiency over the ProactiveMobile baseline.
Abstract:Multimodal large language models (MLLMs) have substantially advanced mobile agents, yet proactive mobile assistance remains challenging because agents must decide \emph{when} to intervene before determining \emph{how} to assist. Existing systems often implement these two decisions within a unified MLLM-based pipeline, leading to goal misalignment between conservative intervention filtering and comprehensive assistance generation, as well as redundant inference when the agent should remain silent. To address these limitations, we propose the \textbf{Pre-Reasoning Perception Framework (PRPF)}, a two-stage framework built on perceiving before reasoning. PRPF introduces a lightweight Multimodal Proactive Perceptor (MPP) for intervention gating and context compression, and activates the Proactive Agent Reasoner (PAR) only when intervention is warranted. Experiments on the ProactiveMobile benchmark show that PRPF substantially reduces false trigger rates (FTR) while improving success rates (SR) and inference efficiency over the ProactiveMobile baseline.
Artificial Intelligence (cs.AI)
arXiv:2606.03236 [cs.AI]
(or arXiv:2606.03236v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2606.03236 arXiv-issued DOI via DataCite (pending registration)
Text integral sub licența deschisă a publicației · 152 cuvinte