What is Memory Decay?
Memory decay is a mechanism in agent memory systems that reduces the importance, relevance, or accessibility of memories over time, mimicking how human memory fades with the passage of time. This approach helps agents prioritize recent and frequently accessed information while gradually reducing the influence of old memories that may no longer be relevant to current tasks. Decay functions typically reduce memory scores exponentially or linearly based on time elapsed since creation or last access.
The implementation of memory decay involves assigning importance scores or weights to memories that decrease over time according to a decay function. When retrieving memories, the system considers both the semantic relevance and the time-adjusted importance, giving preference to recent memories when all else is equal. Some systems also implement recency-based retrieval where recently accessed or reinforced memories receive a boost, preventing important frequently-referenced information from decaying too quickly.
Memory decay serves several purposes in agent architectures: it helps manage limited memory resources by naturally deprioritizing old information, prevents outdated or obsolete memories from interfering with current decision-making, and creates more human-like behavior where recent experiences weigh more heavily. The decay rate is a tunable parameter that affects how quickly memories fade, with faster decay creating more reactive agents and slower decay supporting longer-term learning and consistency.