A structure-preserving finite-time H2-optimal model reduction algorithm compresses deep diagonal state space models to 1/32 of their SSM parameters with comparable accuracy.
Long Range Lan- guage Modeling via Gated State Spaces,
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
-
Compression Method for Deep Diagonal State Space Model Based on $H^2$ Optimal Reduction
A structure-preserving finite-time H2-optimal model reduction algorithm compresses deep diagonal state space models to 1/32 of their SSM parameters with comparable accuracy.