Sign-flipped Sylvester Hadamard matrices make orthogonal RNN recurrent weights binary or sparse ternary while preserving long-memory performance.
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
-
HadamRNN: Binary and Sparse Ternary Orthogonal RNNs
Sign-flipped Sylvester Hadamard matrices make orthogonal RNN recurrent weights binary or sparse ternary while preserving long-memory performance.