SoftBinary Coding uses a stochastic binary latent space and a novel rate-optimal binary channel simulation to address train-test mismatch and smoothness bias in neural compression, with experimental gains over NTC and SOTA vector quantization results.
Canadian Journal of Mathematics , volume=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.IT 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
SoftBinary Coding: A New Information-Theoretic Neural Compression Paradigm
SoftBinary Coding uses a stochastic binary latent space and a novel rate-optimal binary channel simulation to address train-test mismatch and smoothness bias in neural compression, with experimental gains over NTC and SOTA vector quantization results.