Agent memory compression improves action prediction accuracy
Stable Geometry with Divergent Task Evidence for Efficient Long-Horizon Agent Compression
Artificial Intelligence
Summary
When computer programs called agents try to remember lots of past experiences, their understanding and decision-making can slow down. The authors found that just looking at how similar these memories look doesn't guarantee the agent keeps the important information needed for making good choices. They created a method called GEM that selectively keeps memories rich in task-relevant clues while using geometric similarity only to fill in the gaps. This approach uses less memory but still helps the agent perform well on tasks.
What this means in practice
- •For ai system developers: Reduce memory and compute costs for agents performing extended tasks by selectively compressing history based on task relevance.
- •For robotics software teams: Enable robots to maintain efficient past experience records without losing critical action-relevant information for long-term tasks.
Authors
Mingxuan Wang, Fei Luo, Bo Wang, Guorun Yao, Yinglong Guo, Chao Ning, Hongyue Chen, Yanbiao Ma, Jungong Han
Abstract
Long horizon agents accumulate growing interaction histories that increase context and inference costs. We find that geometric redundancy alone is an insufficient criterion for safe compression. Although agent histories exhibit strong low dimensional structure, similar global geometry can preserve very different amounts of task evidence. At identical retained block counts, evidence aware selection raises next action Top 3 retention from 0.31 to 0.69, while centroid similarity remains 0.98. Controlled replacement further shows that action related information can be substantially altered while global geometric measures remain nearly unchanged. Motivated by this gap between geometry and evidence, we introduce Geometry Guided Evidence Preserving Memory (GEM), a training free compressor that protects task and execution evidence before using geometric residuals to complete coverage. GEM reduces mean combined token usage from 2.69M to 2.11M per task, a 21.4% reduction, while maintaining comparable task reward. Our results show that efficient agent history compression should optimize for preserved task evidence rather than geometric coverage alone.