ISO/IEC 24029 neural-network robustness is not an aviation AMC

Why this page: the certifying ML in an aircraft system path already covers what official neural-network verifier pages claim and how that sits next to DO-178C / DO-330. ISO also publishes a numbered series on assessment of the robustness of neural networks. That series was not on the path. A vendor listing names it; the catalog is the authority. ISO sells the full text; this article does not reconstruct unpublished clauses.

Official ISO titles still say Artificial intelligence and AI. Super Intelligence (SI) is this site’s own-voice term for current U.S. executive-branch policy; it does not rewrite ISO titles.

What the catalog says Part 1 is

ISO lists ISO/IEC TR 24029-1:2021, Artificial Intelligence (AI) — Assessment of the robustness of neural networks — Part 1: Overview as a Technical Report, edition 1, 31 pages, from ISO/IEC JTC 1/SC 42. Status is Published. The catalog records publication date 2021-03 and, in the life-cycle table, International Standard published on 2021-03-10.

The public abstract is one sentence: This document provides background about existing methods to assess the robustness of neural networks.

That is an overview Technical Report. The catalog page does not, in that abstract, name an aviation authority, a design assurance level, or a certification scheme.

What the catalog says Part 2 is

ISO lists ISO/IEC 24029-2:2023, Artificial intelligence (AI) — Assessment of the robustness of neural networks — Part 2: Methodology for the use of formal methods as an International Standard, edition 1, 23 pages, from ISO/IEC JTC 1/SC 42. Status is Published. The catalog records publication date 2023-08 and, in the life-cycle table, International Standard published on 2023-08-01.

The public abstract: This document provides methodology for the use of formal methods to assess robustness properties of neural networks. It focuses on how to select, apply and manage formal methods to prove robustness properties.

Part 2 is therefore catalogued as methodology for using formal methods, not as an FAA advisory circular, an EASA acceptable means of compliance, or a DO-178C / DO-330 qualified tool.

Part 3 is still a draft

ISO lists ISO/IEC FDIS 24029-3, Artificial intelligence (AI) — Assessment of the robustness of neural networks — Part 3: Methodology for the use of statistical methods as a Final Draft International Standard. Status is Under development. The stage line is Final text received or FDIS registered for formal approval (50.00). The public abstract: This document provides methodology for the use of statistical methods to assess robustness properties of neural networks.

This site does not treat an FDIS catalog page as a published International Standard.

How vendor and research pages sit next to that catalog

Saimple says it leverages state-of-the-art techniques described in ISO/IEC 24029 standards for the assessment and validation of AI model robustness. That is a vendor sentence. It does not make 24029 an aviation AMC. See the Saimple listing.

The research verifiers on this path describe property checks, not ISO 24029 conformance. Marabou is an SMT-based tool that can answer queries over networks. nnenum is a high-performance neural network verification tool for ReLU networks. ERAN is an analyzer based on abstract interpretation for MNIST, CIFAR10, and ACAS Xu based networks. MathWorks AI Verification provides tools for assessing and verifying properties of deep neural networks, including verify the robustness properties of a network. None of those official pages, in the sentences we cite, say the tool is ISO/IEC 24029 certification. See what’s usable today.

This site has not found an official ISO, FAA or EASA page that names ISO/IEC 24029 as a means of compliance for airborne software, so it does not claim that.

Questions

Is ISO/IEC 24029 a replacement for DO-178C or DO-330?

No source we cite says that. Part 1 is catalogued as background on robustness-assessment methods. Part 2 is catalogued as methodology for using formal methods to prove robustness properties. DO-178C / DO-330 remain on their own explainer.

Does completing a 24029-style assessment certify an airborne ML function?

The catalog pages we cite do not say that. They do not name FAA, EASA or RTCA. Part 3 is still Under development.