A new sparsity loss that binarizes class-specific CNN filter activations during training improves neuro-symbolic rule extraction accuracy by 9% and reduces rule-set size by 53% versus the prior baseline.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Improving Interpretability and Accuracy in Neuro-Symbolic Rule Extraction Using Class-Specific Sparse Filters
A new sparsity loss that binarizes class-specific CNN filter activations during training improves neuro-symbolic rule extraction accuracy by 9% and reduces rule-set size by 53% versus the prior baseline.