Pruning and Deep Compression: retraining, weight sharing and entropy coding

What changed

Pruned small-magnitude trained connections and retrained survivors, then combined pruning with quantized weight sharing/codebook retraining and Huffman coding to reduce stored model size.

Scientific scope

Pruned small-magnitude trained connections and retrained survivors, then combined pruning with quantized weight sharing/codebook retraining and Huffman coding to reduce stored model size. Bundles the2015 pruning precursor and compression pipeline with distinct four/three-author bylines. Inherits earlier pruning/quantization/entropy-coding techniques; no invention of all pruning. Storage savings and retained accuracy are model/evaluation-specific, not guaranteed latency or energy reduction on dense hardware. First preprint title differs from laterDeep Compression title.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Bundles the2015 pruning precursor and compression pipeline with distinct four/three-author bylines. Inherits earlier pruning/quantization/entropy-coding techniques; no invention of all pruning. Storage savings and retained accuracy are model/evaluation-specific, not guaranteed latency or energy reduction on dense hardware. First preprint title differs from laterDeep Compression title.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Learning both Weights and Connections for Efficient Neural Networks
• A Deep Neural Network Compression Pipeline: Pruning, Quantization, Huffman Encoding

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