Bright Box

A diagnostic and monitoring tool for insightful investigation of the reasons behind machine learning 
models errors.

There are practical areas of application of AI / ML methods, where the possibility of understanding and explaining the nature of their operation is legally required (or the regulation is planned) formally.

BrightBox will help you with:

01.

Diagnosing machine learning models – both on global and local levels

02.

Continuous auditing and monitoring of machine learning models operations

03.

Errors and uncertainty estimation & analysis

04.

Prescriptive and what-if analysis for machine learning model uncertainty and decisions

Tech behind the business value

BrightBox is a diagnostic technology that can be used to analyze prediction models and identify model- and data-related issues without direct access to the model.

You just need the preprocessed reference data that you used to train the model, and other preprocessed data (for example, the data used in the model evaluation process or the data from the model’s production environment) and the model predictions based on which you want to diagnose the model.

BrightBox is also a software toolbox for ML models uncertainty estimation and analysis, comprehensive diagnostics of ML models errors and identifying root causes of errors in modeling.

How does it work?

BrightBox technology allows for the diagnosis of ML models – investigation of error types and their possible causes for singular data points and then providing a framework for analysis and generalization of the local results into the global diagnostic of the model- and data-related issues. In this way we aim to provide ML engineers with insight into the actual reasons behind errors and enable better-informed decisions regarding the model and data updating process.

BrightBox is intended to be used by Data Science teams communicating with Business Owners, as a means to improve Machine Learning models on one hand, and bridge the gap in business understanding on the other.

Features:

Diagnostics (global and local) of ML models

ML models errors monitoring.

ML model Uncertainty Estimation & Analytics

Prescriptive analytics of ML model uncertainty

Prescriptive analytics of ML model approximator (surrogate model) decisions

What-if analysis for ML model uncertainty and ML model approximator (surrogate model) decisions

Diagnostics (global and local) of ML models

What-if analysis for ML model uncertainty and ML model approximator (surrogate model) decisions

ML model Uncertainty Estimation & Analytics

Features:

Prescriptive analytics of ML model uncertainty

Prescriptive analytics of ML model approximator (surrogate model) decisions

ML models errors monitoring

Schedule a meeting

with one of our explainable AI experts