WRSWorld Research StateContribute
WORLD RESEARCH STATE

ThiNet: next-layer reconstruction for structured channel pruning

Contribution grade

Structured neural-network compression · Human-designed research method or resource

Reference score105,000Provisional editorial estimate of this exact contribution; its grade is assessed separately.

What changed

Selected channels using sampled next-layer responses and a greedy reconstruction objective, then pruned corresponding filters and fine-tuned the narrowed network.

Scientific scope

Selected channels using sampled next-layer responses and a greedy reconstruction objective, then pruned corresponding filters and fine-tuned the narrowed network. Retaining network structure means topology, not unchanged channel counts. Reconstruction selection is NP-hard and the implemented greedy solution is suboptimal; sampled reconstruction is not exact function preservation or a guarantee of classification accuracy. Training data, fine-tuning and deployment kernels remain part of the cost.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Retaining network structure means topology, not unchanged channel counts. Reconstruction selection is NP-hard and the implemented greedy solution is suboptimal; sampled reconstruction is not exact function preservation or a guarantee of classification accuracy. Training data, fine-tuning and deployment kernels remain part of the cost.

Contributors

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction