On the Applicability of Safety Nets: A Safety-By-Design Solution for Certifying Neural Networks

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Johann Maximilian Christensen, Thomas Stefani, Elena Hoemann, Frank Köster, Sven Hallerbach
• Venue: arXiv: cs.AI
• Published: 2026-08-20
• Collected: 2026-08-21
• URL: http://arxiv.org/abs/2608.20053v1
• PDF: https://arxiv.org/pdf/2608.20053v1
• DOI: 10.71945/icas2026_0210
• Registry group: academic_apis_and_preprints
• Evidence role: canonical_academic_index
• Language: en
• Country / region: UNKNOWN

Plain-Language Brief

A research study presents the first comprehensive analysis of Safety Nets, a combination of neural network compression and lookup tables, as a Safety-by-Design approach for certifying AI systems in aviation. The study identifies optimal design parameters for Safety Nets that balance storage requirements with certification compliance, demonstrating significant size reduction while guaranteeing correct outputs.

What the Document Says

• The integration of AI in safety-critical aviation systems poses challenges.
• Aviation relies on numerous safety-critical systems and is considered safe.
• EASA requires a Safety-by-Design approach for future AI-based systems, which can be achieved using Safety Nets.
• Safety Nets combine neural network compression with lookup tables to ensure correct runtime behavior across the operational design domain.
• No comprehensive study of Safety Net performance characteristics and system design trade-offs had been conducted before this work.
• This study identifies optimal design parameters for Safety Nets, including the number and size of hidden layers and nodes, and one-hot encoding.
• The identified architecture accurately represents at least 97% of the data and uses compact lookup tables to handle errors.
• Safety Nets reduce system size by almost three orders of magnitude, fitting within the memory budget of current avionics hardware.
• The study provides an open-source implementation of Safety Nets for HCAS and VCAS with replicable results.
• Safety Nets guarantee 100% correct outputs across the entire discretized input space as required by EASA guidelines.

Key Concepts Mentioned

• Artificial Intelligence
• Aviation
• Safety-critical systems
• EASA
• Safety-by-Design
• Safety Nets
• Neural network compression
• Lookup tables
• One-hot encoding
• Certification
• Memory budget
• Current avionics hardware

Claims Made by the Document

• Safety Nets are a viable Safety-by-Design solution for safety-critical applications.
• Optimal design parameters for Safety Nets have been identified.
• Safety Nets can reduce system size by almost three orders of magnitude.

Evidence Used

• The study presents a systematic analysis and comparison of neural networks with diverse architectures to identify optimal design parameters.
• Results demonstrate that the identified architecture reduces system size significantly and guarantees correct outputs.
• The work provides an open-source implementation of Safety Nets.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

The integration of Artificial Intelligence (AI) in safety-critical aviation systems presents significant challenges for certification and deployment. Aviation, often regarded as the safest form of transportation, relies on numerous safety-critical systems. For future safety-critical AI-based systems, EASA requires a Safety-by-Design approach, which can be achieved by using Safety Nets that combine neural network compression with lookup tables to ensure 100 % correct runtime behavior across the discretized operational design domain. Although Safety Nets have been studied, no comprehensive study of their performance characteristics and system design trade-offs has been conducted. This work presents the first systematic analysis of the trade-off between neural network and lookup table size in Safety Nets. By systematically comparing neural networks with diverse architectures, this study identifies optimal design parameters that minimize overall storage and memory requirements while maintaining certification compliance. Results demonstrate that architectures with 3 to 5 hidden layers, each with approximately 50 to 100 nodes, combined with one-hot encoding, achieve the best balance. In these configurations, neural networks accurately represent at least 97 % of the data, while compact lookup tables handle the remaining errors. The resulting Safety Nets reduce the system size by almost three orders of magnitude, fitting within the memory budget of current avionics hardware while guaranteeing 100 % correct outputs across the entire discretized input space, as required by EASA guidelines. This work provides the first-ever open-source implementation of Safety Nets for HCAS and VCAS with replicable results, demonstrating a practical pathway toward certifiable AI-based systems in aviation and establishing Safety Nets as a viable Safety-by-Design solution for safety-critical applications.