{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/model-compression-as-constrained-optimization","title":"Model compression as constrained optimization, with application to neural nets. Part II: quantization","arxiv_id":"1707.04319","date":"2017-07-13","proceeding":null,"authors":["Miguel Á. Carreira-Perpiñán","Yerlan Idelbayev"],"abstract":"We consider the problem of deep neural net compression by quantization: given\na large, reference net, we want to quantize its real-valued weights using a\ncodebook with $K$ entries so that the training loss of the quantized net is\nminimal. The codebook can be optimally learned jointly with the net, or fixed,\nas for binarization or ternarization approaches. Previous work has quantized\nthe weights of the reference net, or incorporated rounding operations in the\nbackpropagation algorithm, but this has no guarantee of converging to a\nloss-optimal, quantized net. We describe a new approach based on the recently\nproposed framework of model compression as constrained optimization\n\\citep{Carreir17a}. This results in a simple iterative \"learning-compression\"\nalgorithm, which alternates a step that learns a net of continuous weights with\na step that quantizes (or binarizes/ternarizes) the weights, and is guaranteed\nto converge to local optimum of the loss for quantized nets. We develop\nalgorithms for an adaptive codebook or a (partially) fixed codebook. The latter\nincludes binarization, ternarization, powers-of-two and other important\nparticular cases. We show experimentally that we can achieve much higher\ncompression rates than previous quantization work (even using just 1 bit per\nweight) with negligible loss degradation.","url_abs":"http://arxiv.org/abs/1707.04319v1","url_pdf":"http://arxiv.org/pdf/1707.04319v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"model-compression-as-constrained-optimization","repo_url":"https://github.com/UCMerced-ML/LC-model-compression","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"binarization","task_name":"Binarization"},{"task_slug":"model-compression","task_name":"Model Compression"},{"task_slug":"quantization","task_name":"Quantization"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1707.04319","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}