MobileMem: Learning from a Year of Mobile Experiences
MobileMem tests whether mobile agents can use year-scale, personal app histories as memory, not just retrieve isolated facts.
The paper introduces a benchmark and framework for on-device long-term memory built from synthesized long-horizon mobile experience trajectories. It covers text and multimodal tasks, including multi-hop and temporal reasoning, knowledge updating, and implicit preference inference. The authors frame the goal as agents that can remember past user experiences, interpret the present, and adapt over time. HF Daily Papers' note
The paper introduces a benchmark and framework for on-device long-term memory built from synthesized long-horizon mobile experience trajectories. It covers text and multimodal tasks, including multi-hop and temporal reasoning, knowledge updating, and implicit preference inference. The authors frame the goal as agents that can remember past user experiences, interpret the present, and adapt over time. HF Daily Papers' note
score 5