Index modelling with deep learning characteristic

Abstract

An artificially designated relationship can assist the realization of mapping, which can compress the output structure of a machine learning model. For example, an index model of a maximal entropy model can utilize machine learning embedding and mapping to generate classified output. In this way, the feature finding function of a machine learning model (for example, deep networks) and relationship, which is developed based on the people's understanding of structural properties of a problem to be solved, are cooperatively combined to compress the output structure of the model without generatingobvious precision loss. These compressed models improve the adaptability of the models in facilities or other resource limited scenes.

Description

[0001] priority claim [0002] This application claims priority to U.S. Patent Application Serial No. 62 / 752,128, filed October 29, 2018, entitled "Exponential Modeling with Deep Learning Features," the entire disclosure of which is incorporated herein by reference. technical field [0003] This disclosure relates generally to machine learning. More specifically, the present disclosure relates to machine learning models, including exponential models (eg, maximum entropy models) that exploit the mapping between output classes and embedding parameters to provide compression gains. Background technique [0004] Various forms of machine learning models have revolutionized many areas of machine intelligence. As an early example, at some point in the past, maximum entropy models provided state-of-the-art performance in natural language processing and other technical domains. Maximum entropy models follow the principle that the model should provide as few biased estimates as poss...

AI Technical Summary

Problems solved by technology

However, to date, efficient solutions to achieve compression gains have not been identified for output layers of models that output to a very large number of different possible classes

Images