Remembering Has a Price: Learning Memory Worth under Resource Constraints

Key points

• Lifelong agents continually accumulate and reuse experience to improve after deployment, yet in practice they often operate under tight generation, interaction, and memory budgets.1
• Under such stringent joint constraints, we find that existing lifelong-memory methods can become ineffective, with several sophisticated approaches offering little advantage over simple retrieval or even no-memory agents.1
• When only a small amount of past experience can be retained or reused and opportunities for reasoning and recovery are limited, retrieving a relevant but ineffective memory can consume scarce context and hinder downstream execution.1
• These results suggest that effective lifelong adaptation under resource constraints depends not on remembering more, but on learning what is worth remembering.1

Sources (2)

• [1]Remembering Has a Price: Learning Memory Worth under Resource Constraints
  Hugging Face Daily Papers · Oct 19, 12:00 AM

  To address this challenge, we propose TokenWorth, a lightweight online memory framework that learns query-dependent memory worth from sparse execution feedback and applies it consistently to memory retrieval and bounded memory management.

  Lifelong agents continually accumulate and reuse experience to improve after deployment, yet in practice they often operate under tight generation, interaction, and memory budgets.

• [2]Remembering Has a Price: Learning Memory Worth under Resource Constraints
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 02:09 PM · same content

Extractive summary: sentences quoted from the sources.