{"schema_version":"1.0","canonical_url":"https://patentable.app/patents/US-11494657","patent":{"patent_number":"US-11494657","title":"Quantizing neural networks using approximate quantization function","assignee":null,"inventors":[],"filing_date":"2019-10-08T00:00:00.000Z","publication_date":"2022-11-08T00:00:00.000Z","cpc_codes":["G06N","G06F","G06F","G06N","G06N","G06N","G06N","G06N","G06N","G06N","G06N","G06N","G06N"],"num_claims":20,"abstract":"Some embodiments of the invention provide a novel method for training a quantized machine-trained network. Some embodiments provide a method of scaling a feature map of a pre-trained floating-point neural network in order to match the range of output values provided by quantized activations in a quantized neural network. A quantization function is modified, in some embodiments, to be differentiable to fix the mismatch between the loss function computed in forward propagation and the loss gradient used in backward propagation. Variational information bottleneck, in some embodiments, is incorporated to train the network to be insensitive to multiplicative noise applied to each channel. In some embodiments, channels that finish training with large noise, for example, exceeding 100%, are pruned."},"analysis":{"summary":null,"layman_explanation":null,"technical_analysis":null,"business_analysis":null,"faqs":null,"topics":[],"tech_cluster":null},"seo":{"title":"Quantizing neural networks using approximate quantization function","description":"Some embodiments of the invention provide a novel method for training a quantized machine-trained network. Some embodiments provide a method of scaling a feature map of a pre-trained floating-point ne","keywords":[]},"attribution":{"source":"Patentable","source_url":"https://patentable.app","canonical_url":"https://patentable.app/patents/US-11494657","license":"CC-BY-4.0-like","license_terms":"AI-generated analysis on this page (summary, layman_explanation, technical_analysis, business_analysis, faqs) may be reused with attribution and a visible link back to the canonical URL above. Patent abstracts, claims, and bibliographic data are USPTO public domain.","required_link":"https://patentable.app/patents/US-11494657","citation_suggestion":"Patentable. \"Quantizing neural networks using approximate quantization function\" (US-11494657). https://patentable.app/patents/US-11494657","copyright_holder":"Nomic Interactive Technology LLC"},"links":{"html":"https://patentable.app/patents/US-11494657","json":"https://patentable.app/api/llm-context/US-11494657","site":"https://patentable.app","llms_txt":"https://patentable.app/llms.txt"},"generated_at":"2026-05-31T16:51:02.829Z"}