RL-Guided Temporal Localization for Dual-Channel Retrieval in Long-Horizon Agent Memory

Key points

• Even time-aware approaches inject temporal cues only during memory organization or answer generation, overlooking temporal constraints at the retrieval stage, where candidate selection occurs - the root cause of the displacement.1
• Inspired by the temporal contiguity effect in human episodic memory, we propose TIDER (Temporal Interval-driven Dual-channel Evidence Retrieval), a time-aware agentic retrieval framework.1
• Central to TIDER is a reinforcement learning (RL)-trained temporal interval locator that infers the temporal windows in which the evidence lies.1
• RL optimizes window localization as a discrete, non-differentiable retrieval decision, using composite rewards to convert sparse answer supervision into fine-grained retrieval feedback.1

Sources (1)

• [1]RL-Guided Temporal Localization for Dual-Channel Retrieval in Long-Horizon Agent Memory
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 16, 03:16 PM

  Time-aware retrieval is crucial for long-horizon LLM agents to effectively leverage ever-growing long-term memory.

  Even time-aware approaches inject temporal cues only during memory organization or answer generation, overlooking temporal constraints at the retrieval stage, where candidate selection occurs - the root cause of the displacement.

Extractive summary: sentences quoted from the sources.