MemoWM: How World Models Change What Agents Need to Remember
We formulate the problem of memory allocation conditioned on a world model and introduce MemoWM, a framework that uses shared predictions to compress retained information and reconstruct omitted content.
Key points
- Long-term agents face growing storage demands as they accumulate experience.
- World models capture reusable regularities that can reduce the information stored for each experience.
- Across five long-term agent-memory benchmarks, MemoWM achieves 42.42% average answer accuracy, exceeding the strongest baseline by 2.62 percentage points, while reducing average experience-specific storage by 53.9% relative to MIRIX, the most storage-efficient baseline.
- Accounting for model parameters reveals a trade-off between shared model capacity and recurring storage costs, with the capacity that minimizes total storage increasing as more interactions are retained.
Sources (1)
- [1]MemoWM: How World Models Change What Agents Need to RememberarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 06:33 PM
We formulate the problem of memory allocation conditioned on a world model and introduce MemoWM, a framework that uses shared predictions to compress retained information and reconstruct omitted content.
Long-term agents face growing storage demands as they accumulate experience.
Extractive summary: sentences quoted from the sources.