Integration of external applications into deep neural networks

Abstract

During end-to-end training of a Deep Neural Network (DNN), a differentiable estimator subnetwork is operated to estimate a functionality of an external software application. Then, during inference bythe trained DNN, the differentiable estimator subnetwork is replaced with the functionality of the external software application, by enabling API communication between the DNN and the external software application.

Description

technical field [0001] The present invention relates to the field of deep neural networks (DNN). Background technique [0002] End-to-end learning of DNNs has gained ground over the past few years, achieving state-of-the-art performance in several domains including computer vision, text analysis, and speech recognition. In end-to-end learning, the learning process can train a classifier in a single end-to-end step, which removes the need for complex multi-stage machine learning pipelines. [0003] One of the challenges facing the industry is integrating DNNs with external applications. Enabling DNNs to use or interact with external applications (e.g., through application programming interfaces (APIs)) can yield superior solutions, since functionality or logic provided by external applications can be included in the DNN, or become part of the DNN. Accessing the API is done via the API's input and output parameters. For example, consider the simple question: "Is 7.2 greate...

AI Technical Summary

Problems solved by technology

[0004] There are inherent difficulties in training DNNs to interact with external applications through their APIs: most DNN training procedures rely on different variants of backpropagation of gradients

Images