Algorithm Debt in Scientific Software

Goal: Algorithm Debt has been found in performance-critical and algorithm-intensive projects (e.g., Deep Learning frameworks) (Liu et al., 2020), used to support other works. Thus, Algorithm Debt may have a transitive effect–i.e., if a piece of scientific software is not robust and thus prone to errors, then the research that leverages that software may be negatively impacted or even suffer from validity treats (Vidoni, 2021). This is concerning, considering the high rates of research retracted due to incorrect results, even after reaching social media (Marton et al., 2021).

With Scientific Software used in different computation disciplines as a research tool (Sculley et al., 2015), there is a need to deliver high-quality and defect-free software. Understanding Algorithm Debt is essential to define clear pathways to mitigate its occurrence, simplify its management, and facilitate researchers’ informed decisions about the software they choose.

Therefore, it can be stated that Algorithm Debt is a nascent research area at the forefront of Software Engineering for scientific software. This research aims to characterise Algorithm debt, provide tools for its detection, and understand the human-centric challenges to provide practical, translatable results that will benefit the broader research community.

This project is Iko-ojo Simon’s PhD Topic.

Participants: - Dr. Melina Vidoni, Australian National University, Australia. - Dr. Fatemeh H. Fard, University of British Columbia, Canada.

Students: - Iko-ojo Simon, PhD Student, Australian National University, Australia. - Richeng Zhang, MsC Student, Australian National University, Australia.

Dr Melina Vidoni
Software Manager @ eWater

Software Manager @ eWater. Lecturer @ ANU (on leave).