The best sparse neural architecture for deep RL agents depends on whether hidden-layer weights are fixed or learned, and spatial sparsity is not always best even in spatially-structured games.
friendly bullet
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
-
On the Interplay Between Sparsity and Training in Deep Reinforcement Learning
The best sparse neural architecture for deep RL agents depends on whether hidden-layer weights are fixed or learned, and spatial sparsity is not always best even in spatially-structured games.