Deliver EU AI-Act compliant AI systems following ISO/IEC 24029 suite of standards with Saimple
Saimple is the only software on the market to date that allows you to validate robustness of your neural network models in conformance with 24029 ISO/IEC suite of standards and European law on artificial intelligence (EU AI-Act) regarding the robustness requirements.
New regulations are brewing in Europe (the AI-Act), certification framework is being developed in the ISO/IEC (ISO/IEC 42001), and many standards are being published on several topics regarding artificial intelligence: robustness, bias, performance, ethics, etc. Certification framework are being developed in the automotive (SOTIF) and the aeronautic (WG114) sectors for example. All of these changes will drastically redistribute the cards around the industries and allow newcomers, new usages and new products to take place.
This evolution concerns all the sectors where software safety is common practice, such as the automotive, aeronautic, railway, healthcare or the Defense sectors. But it concerns also now new industries less used to these concepts, such as the banking or the HR industries.
Due to the way AI systems are designed and their changing behavior, they cannot be validated following traditional software quality framework (e.g., SQuaRE) or code certification (e.g., DO-178C).
You are in good hands
Saimple is built by Numalis, a leading software design company which is the editor of the ISO/IEC 24029 series of standards that are going to be the foundation of the robustness assessment of AI system worldwide. Saimple uses formal validation to prove AI robustness, which is the strongest method retained by the standardization.
Assessing robustness implies to verify several requirements along the life cycle of an AI system, with special care taken during the design, validation, deployment and monitoring stages. At each stage different activities can be done in order to assess the robustness, improve the system and document the whole process to comply with the regulation.
In particular Saimple helps you realize several of the ISO/IEC 24029-2 requirements, such as:
- Justifying the behavior of the system using the recognized features
- Defining and testing the robustness against custom-made noise
- Measuring the separation between classes
- Covering part of the input domain
- Tracing every change in the robustness at each version of your ML model
Under active development and coming soon :
- Visualizing the robustness across the input domain
- Measuring robustness drift in operation
ISO/IEC standards and EU regulation compliance gives you several benefits:
- Builds trustworthy AI and reinforces brand reputation / excellence
- Proves model reliability and eases AI adoption
- Helps mitigate risks
- Opens new markets
- Gives competitive advantage compared to other solutions
- Makes your services / devices / tools marketable
International Standards on AI we’ve published
This document provides background about existing methods to assess the robustness of neural networks.
This document provides methodology for the use of formal methods to assess robustness properties of neural networks. The document focuses on how to select, apply and manage formal methods to prove robustness properties.