Mem-alpha uses RL to train agents on memory construction via QA-based rewards, yielding better performance than baselines and generalization from 30k to over 400k tokens.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Mem-{\alpha}: Learning Memory Construction via Reinforcement Learning
Mem-alpha uses RL to train agents on memory construction via QA-based rewards, yielding better performance than baselines and generalization from 30k to over 400k tokens.