Across nine RL environments with the PPO algorithm, Mamba-2 is the only model that both solves the long-memory MiniGrid task and matches the memory efficiency of a simple network, 8x lighter than Transformer-XL.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
RLBenchNet: The Right Network for the Right Reinforcement Learning Task
Across nine RL environments with the PPO algorithm, Mamba-2 is the only model that both solves the long-memory MiniGrid task and matches the memory efficiency of a simple network, 8x lighter than Transformer-XL.