NB! HDR talks take place in their cluster slots below or on Tuesdays at 10am in room HN 3.41.
Please note that AI/ML cluster holds its regular seminars here:
https://cs.anu.edu.au/ai-ml-friends/
Please note that Foundations cluster holds its regular seminars here:
https://comp.anu.edu.au/foundations/seminars/
We expect Data Science and Computational cluster oral presentations to take place in this common time slot.
Please email HDR Admin to book a slot for your Oral Presentation milestone.
Upcoming Talks
Date: Tuesday, 5 August 2025 Time: 10:00 am – 11:00 am Venue: Room 3.41, Hanna Neumann Building (#145)
Presentation Title: Characterising Algorithm Debt in Machine and Deep Learning Systems
Abstract
Machine and Deep Learning (ML/DL) systems (i.e., systems incorporating any ML/DL algorithms or models) have become integral to a wide range of applications and domains, from natural language processing to autonomous vehicles, driving innovation in industries such as healthcare, transportation, and industrial automation. However, the adoption of these technologies has introduced challenges in maintaining their long-term reliability and scalability due to Algorithm Debt (AD). AD is a form of technical debt that arises from suboptimal algorithmic choices during systems development, often leading to model degradation and poor scalability. Despite the impact of AD on ML/DL systems, it remains understudied. Many practitioners are unaware of its impact, and the complexity of ML/D algorithms further exacerbates system inefficiencies, hindering the long-term reliability of ML/DL systems. My thesis addresses AD in ML/DL systems through four main contributions: 1) a systematic literature review of 44 studies, defining AD and identifying eight AD smells to establish foundational insights for both researchers and practitioners; 2) a mixed-methods study with 21 interviews and 65 survey responses, uncovering 13 causes, eight effects, and 15 mitigation strategies for AD to guide practitioners in managing AD effectively; 3) an empirical analysis of ML/DL classifiers to enable automated, early identification of AD and support practitioners for its proactive management in development pipelines; and 4) a framework for the understanding of AD in ML/DL systems. By characterising AD in ML/DL systems, this thesis advances three key areas: enhancing the reliability and scalability of ML/DL systems, bridging academic research with industrial practices to mitigate AD, and fostering proactive awareness and management of AD among practitioners to ensure sustainable ML/DL software development.
Short Bio
Iko-Ojo Simon is a Ph.D. candidate at the Australian National University (ANU), School of Computing. He is supervised by Dr. Chirath Hettiarachchi (Primary Supervisor), Dr. Fatemeh Fard (University of British Columbia), Associate Professor Alex Potanin (ANU), and Professor Hanna Suominen (ANU, Panel Chair). Prior to his Ph.D., Iko-Ojo worked as an Assistant Lecturer at the University of Jos, Nigeria, where he taught courses in Software Engineering. At ANU, he has supervised undergraduate and master’s research projects. He holds a B.Sc. and M.Sc. in Computer Science from the University of Jos. Iko-Ojo’s research interests include Empirical Software Engineering, Technical Debt in ML/DL systems, and applications of AI in Healthcare.