Machine learning (ML) engineering increasingly incorporates principles from software and requirements engineering to improve development rigor; however, key non-functional requirements (NFRs) such as interpretability and explainability remain difficult to specify and verify using traditional requirements practices. Although prior work defines these qualities conceptually, their lack of measurable criteria prevents systematic verification. This paper presents a novel provenance-driven approach that decomposes ML interpretability and explainability NFRs into verifiable functional requirements (FRs) by leveraging model and data provenance to make model behavior transparent. The approach identifies the specific provenance artifacts required to validate each FR and demonstrates how their verification collectively establishes compliance with interpretability and explainability NFRs. The results show that ML provenance can operationalize otherwise abstract NFRs, transforming interpretability and explainability into quantifiable, testable properties and enabling more rigorous, requirements-based ML engineering.
Collision avoidance systems are commonly used to avoid fragmentation events occuring in Low-Earth Orbit (LEO) and Geosynchronous Equatorial Orbit (GEO). However, these events have been growing in frequency as orbital congestion worsens with the launch of megaconstellations. Consequently, conjunction alerts and collision risks are becoming increasingly common. Current practices, which are commonly manual or rule-based, have difficulty scaling to these worsening dynamic environments. To address this intensifying situation, we propose a reinforcement-learning policy for autonomous collision avoidance, trained via Proximal Policy Optimization (PPO) along with an open-source, high-fidelity astrodynamics simulator for training and evaluation. In 1,000 deterministic GEO episodes, our agent achieves a 97.5% collision avoidance success rate, outperforming traditional controllers such as a rule-based baseline (20.7% success) and an impulsive Δv planner baseline (27.5% success). To achieve these results, we designed a simulator to train and evaluate our agent, using real-world and simulated debris. We simulate Newtonian two-body dynamics using Sun/Moon third-body perturbations, fuel-dependent thrust, and configurable debris fields. The agent is trained with curriculum learning and shaped rewards oriented toward encouraging survival, adequate projected miss distance, and Δv conservation. Finally, our evaluation consisted of a fully deterministic pipeline, including shared seeds, per-episode logs, and telemetry exports a.
9 May 2026 in ICCK Journal of Software Engineering
Non-Functional Requirements (NFRs) are quality-focused attributes of a system that impact functional components. There are 24 common NFR classes with some of the most used being performance, scalability, availability, reliability, and security. The implementations of these classes are ambiguous and describe the attributes on the behavior of software. Due to their nature, NFRs are typically realized through the specification and implementation of functional requirements. For traditional software systems, the NFR definitions are well known. However, in Machine Learning (ML), which has numerous applications across a multitude of domains, a current challenge is that traditional software non-functional class definitions do not consider stochasticity and black-box nature of ML, and thus, are not appropriately reflecting the ML context. This paper aims to address that challenge by mapping traditional software NFR definitions into ML-NFR definitions by redefining them with ML characteristics in consideration. This research separates the 24 common NFRs into system NFRs and ML NFRs to delineate which of the original 24 carries over to the ML context. Because of the proliferation of ML, requirement engineers need to be cognizant of the nuances of ML behavior as it pertains to NFRs. This paper presents a mapping of NFRs into the ML context including new definitions, as needed, for NFRs in ML systems.