Enabling trust in data-driven systems - Machine learning assurance
Machine learning assurance provides a risk-based approach to establish trust in systems that use machine learning (ML).
Machine learning assurance
Purchasing and implementing third-party machine learning systems
Interest in systems incorporating machine learning, artificial intelligence (AI) and other data-driven techniques is spiking, as organizations seek ways to do things better, faster and to push the boundaries of what is possible. As an enterprise considering using data-driven applications, you need evidence that the system is fit for purpose to limit your organization’s exposure to liability, that your decisions are based on unbiased information and that the system will meet your business needs. How can you select a supplier of ML systems with confidence?
Building and selling machine learning systems
While opportunities for companies building machine learning solutions are plenty, there are many challenges. Making the right decisions on algorithms, techniques, data and evaluation criteria is crucial.
The complexity of the data and of the training algorithms, coupled with a lack of specific standards and regulations, can make establishing trust in such solutions a difficult task. As a result, potential investors and customers often take a conservative stance or refrain from adopting such solutions until these have matured. How can you bring to market proven ML solutions?
DNV GL's recommended practice for machine learning
With the Recommended Practice DNVGL-RP-0510 Framework, stakeholders obtain a better understanding of the risks of machine learning projects. Through an assessment of the design, development, testing and deployment of your machine learning model, this Recommended Practice (RP) allows you to identify and manage risks, increasing the likelihood of a successful outcome. The RP provides sound guidance to assess:
- the development process
- the model
- the risk of using the model
DNV GL is a recognized subject matter expert to identify, assess and mitigate risks. Our framework helps you to scrutinize decisions made by the developer, even if you lack the domain knowledge. With organized claims and documentation, you can assess the implications of using the system, what returns to expect, and when.
DNV GL’s assurance and risk assessment service for data-driven models
The machine learning assurance and risk assessment service covers the complete pipeline, from data collection and ingestion to data preparation, modelling, prediction and deployment. For each stage in the pipeline, aspects of high risk are identified along with possible mitigating actions. We help you with a framework that eases communication about a machine learning project’s risk to all stakeholders – it provides easy-to-understand information about complex data science topics. Using a workshop and questionnaire-driven format, the service delivers:
- A detailed register of risk items and mitigating actions
- Feedback to your machine learning project team or data science vendor from DNV GL’s domain experts and data scientists
- Trust and assurance that the machine learning model will deliver the expected results