Monitaur is a Machine Learning Assurance platform addressing the needs of companies using models to make decisions in regulated industries. Monitaur delivers the transparency and confidence necessary to manage compliance and unlock innovation. It enables recording, auditing, monitoring, and assurance of machine learning models. Having the assurance that your models are doing what they are designed to do allows a company to unlock innovation and increase top-line growth by deploying models that have previously stayed sandboxed and/or decrease compliance cost.
Monitaur is platform and environment agnostic, with planned support for all classical machine learning and deep learning implementations in Python, R, and Java. With full model versioning and transaction reproducibility, Monitaur allows counterfactuals for auditability of machine learning models. Monitaur can be deployed on-prem, on the Monitaur Cloud, or on a customer’s Cloud.
The Monitaur client library is designed to record your machine learning model's meta information along with hashed versions of the serialized model and production files. Each transaction is then recorded as it runs on your production infrastructure.
In addition, Monitaur versions your models based on changes to the production and trained model file hashes. When changes occur, the platform automatically generates alerts. You can also define your own alerting logic based on transaction-level details.
The Monitaur GRC (Governance, Risk, and Compliance) web application enables a nontechnical user to find individual machine learning transactions and verify their inputs and decisions for audit and compliance requests. With full model versioning and transaction reproducibility, Monitaur supports counterfactuals for model auditing. Counterfactuals offer the ability to reperform transactions with the same model version and inputs, or even with slight variations, for "what-if" analyses.
Monitaur allows users to set alerts, rules, and application controls to proactively manage risks, bias, and compliance. It establishes metrics for model and feature drift to detect model degradation. Also, it introduces bias controls for controlled variables such as age, sex, or race.
Monitaur provides automated audit reports exportable for a third party audit or for tracking over time. This supports continuous and independent validation of ML compliance. Monitaur also facilitates model white paper creation and workflow process development around proven machine learning governance frameworks.
Machine Learning models are being used to reduce readmissions and optimize healthcare quality. With Monitaur, an audit trail is created that is verifiable, with full model versioning and transaction reproducibility to provide assurance around model decisions.
A Machine Learning classification model, such as Gradient Boosted Machine, can be used to determine if a customer will default on their loan payments. Monitaur provides recording, auditing, monitoring, and assurance to provide peace of mind and assurance for companies in regulated industries using models to make decisions.