Computer System and Method for Quantizing Artificial Neural Network Model

NOTA, INC. filed a patent application for a method of quantizing artificial neural network models. The method involves identifying outliers from activation elements output by a layer, determining and regularizing weights based on their relevance to the identified outlier, and then quantizing the neural network model. The patent is classified under CPC G06N 3/082 (Computer systems based on biological models, specifically learning methods). Application number 18945122 was published April 2, 2026.

This is a patent application publication only—it does not create compliance obligations. Technology companies developing neural network quantization methods should review the published claims to assess potential intellectual property considerations if implementing similar techniques. There is no regulatory action required and no penalty or enforcement implication from this publication.