Browse Prior Art Database

System and Method for Visualization, Run-time Validation, and Debugging of Deep Learning Models

IP.com Disclosure Number: IPCOM000250620D
Publication Date: 2017-Aug-10
Document File: 4 page(s) / 466K

Publishing Venue

The IP.com Prior Art Database

Abstract

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 53% of the total text.

The system proposes a set of cognitive tools set for visualization, validation, run-time verification and debugging of deep learning models.

Disclosed is a system that aims for visualization, validation, run-time verification, and debugging of deep learning models. Deep learning helps in learning complex representations from noisy data to enable better data understanding. Constructing and training any deep learning model, is a cognitively challenging task for humans requiring technical skills. Major challenges in training deep learning are handling large volumes of data, requirement of huge compute capacity and memory. Additionally, training deep learning models is time intensive task and hence results are delayed. Since, training happens like a black box, what is being learnt is never discovered and incase of modifications to the model, the entire training starts from scratch.

Current system for debugging of deep learning models is very limited. Metrics required to monitor training procedure needs to be manually selected. Further, tools don’t aim to perform real time visualization of watch variables, debugging using break points. These tools don’t enable auto hyperparameter suggestion and resource prediction for deep learning models.

The system proposes a set of cognitive tools for run-time interpretation and reliable training of deep learning models, through deep learning model validation, visualization, verification and debugging. The overview of the proposed system is shown in the Figure 1

Figure 1: System Overview

The system aims to:

1. Automatic resource prediction and validation of deep learning model to suit the user requirements: This creation automatically predicts the amount of resources (time, space, computation)

and validates the provided deep learning model with respect to the user requirements. The overview of the method is shown in Figure 2.

Figure 2 Automatic resource predic...