SoftSAE replaces fixed-K sparsity in autoencoders with a learned, input-dependent number of active features via a soft top-k operator.
Labeling neural representations with inverse recognition.Advances in Neural Information Processing Systems, 36:24804–24828
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
SoftSAE: Dynamic Top-K Selection for Adaptive Sparse Autoencoders
SoftSAE replaces fixed-K sparsity in autoencoders with a learned, input-dependent number of active features via a soft top-k operator.