Channel importance splits into task relevance and local replaceability; local-axis metrics predict safe removal under pruning better than target-axis metrics across multiple CNNs and datasets.
IEEE Trans
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
verdicts
UNVERDICTED 5representative citing papers
Meta-analysis of 28 FFS studies shows experimental design choices explain 33% of variance in new method performance against baselines.
EBBS augments the MIO best-subsets objective with an aggregated expert prior expressed as a log-odds penalty so that selected features align with domain consensus while reducing to ordinary best subsets when experts provide no input.
Kernel ridge regression combined with mRMR feature selection improves prediction of full benchmark scores from question subsets over existing efficient benchmarking techniques.
A replicator-type dynamic on the standard simplex for feature weights from a normalized data matrix converges globally to a unique interior equilibrium.
citing papers explorer
-
Task Relevance Is Not Local Replaceability: A Two-Axis View of Channel Information
Channel importance splits into task relevance and local replaceability; local-axis metrics predict safe removal under pruning better than target-axis metrics across multiple CNNs and datasets.
-
Bias in Filter Feature Selection Evaluation: A Meta-Analysis of Datasets, Baselines, and Experimental Design Choices
Meta-analysis of 28 FFS studies shows experimental design choices explain 33% of variance in new method performance against baselines.
-
A Mathematical Optimization Approach for Expert-Informed Bayesian Best Subset Selection
EBBS augments the MIO best-subsets objective with an aggregated expert prior expressed as a log-odds penalty so that selected features align with domain consensus while reducing to ordinary best subsets when experts provide no input.
-
Efficient Benchmarking Is Just Feature Selection and Multiple Regression
Kernel ridge regression combined with mRMR feature selection improves prediction of full benchmark scores from question subsets over existing efficient benchmarking techniques.
-
Feature weighting for data analysis via evolutionary simulation
A replicator-type dynamic on the standard simplex for feature weights from a normalized data matrix converges globally to a unique interior equilibrium.