Interview with the 2026 ACM TOSEM Outstanding Paper (TOP) Award Recipient
Interview conducted by Ruijie Meng and Damian Tamburri,
ACM TOSEM Social Media Editors
The ACM TOSEM Outstanding Paper Award, or TOP Award in brief, is an annual award from ACM Transactions on Software Engineering and Methodology (TOSEM). The award recognizes a paper published five years ago that has gone on to inspire a particularly promising research direction. By emphasizing a medium-term horizon, the TOP Award helps the community identify contributions whose impact is clearer in hindsight. For the 2026 award, journal first papers published in 2020–21 were considered. The selection committee was chaired by IpeK Ozkaya and Dongmei Zhang.
We congratulate the authors — Wei Ma, Mike Papadakis, Anestis Tsakmalis, Maxime Cordy, and Yves Le Traon — on their paper, “Test Selection for Deep Learning Systems” (TOSEM, Vol. 30, №2, Article 13, December 2020), which was selected for the 2026 TOP Award. We are pleased to interview the author representative, Prof. Michail Papadakis, to share the story behind this work.
First, congrats for receiving the award! What motivated you to conduct this research on test selection for deep learning systems?
We originally come from a software engineering background, with a strong focus on validation and verification. We also collaborate closely with industry that already at the time was facing several issues with the testing of AI models. When we started this work, AI was not yet as pervasive as it is today, but we were already seeing its rapid rise and early industrial adoption. This naturally raised the question of how to properly test and validate machine learning systems. This motivated us to build a research agenda bridging software engineering and AI/ML, and this paper represents one of the first strong outcomes of that long-term investment.
In your view, what is the impact of this research contribution on the field?
This work was among the first to explicitly formulate the problem of test selection for machine learning systems and to propose uncertainty-based baselines to address it. By framing the problem clearly, we helped open a new research direction and contributed (together with other contemporaneous works) to the progressive emergence of a research taxonomy for ML testing that has structured subsequent research in the area.
Can your research results be applied in practice, such as in industrial settings?
Yes, and we actually did already. We collaborated with a major banking industry partner in Luxembourg to explore practical applications. More broadly, test selection is a general problem that applies to virtually any ML model or system. The metrics proposed in the paper are also practical to use, since they rely on simple assumptions, such as access to model logits. We are currently integrating these ideas into a larger initiative aimed at building a generic, industry-ready AI system assessment tool.
What were the main challenges you encountered during this work? Did you expect that the paper might receive an Outstanding Paper Award?
One challenge was that, at the time, ML testing was still an emerging area, with limited established methodologies and terminology. This meant we had to define the problem and evaluation setup largely from scratch while ensuring relevance to both software engineering and machine learning communities.
Regarding the award, it was honestly unexpected. We were mainly focused on addressing what we believed was an important research gap, so receiving this recognition several years later is both surprising and very rewarding.
What advice would you give about how to conduct impactful research?
Our experience suggests that impactful research often comes from identifying emerging technological shifts early and approaching them with strong methodological foundations grounded in problems that are either open or that emerge from these technological shifts.
Disclaimer: The posts in the SIGSOFT Blog are written by individual contributors and any views or opinions represented in their posts are personal, belong solely to the blog authors, and do not necessarily represent those of ACM SIGSOFT or ACM.
