Pisanu D, Walter J, Franke J (2025)
Publication Type: Conference contribution
Publication year: 2025
Publisher: Institute of Electrical and Electronics Engineers Inc.
Conference Proceedings Title: 3rd International Conference on Integrated Systems in Medical Technologies, ISMT 2025 - Conference Proceedings
Event location: Erlangen, DEU
ISBN: 9798331557157
DOI: 10.1109/ISMT68188.2025.11488082
With healthcare robots increasingly relying on adaptive machine learning to personalize assistance, privacy compliance is becoming a critical design requirement. Under regulations such as the General Data Protection Regulation, patients have the 'right to be forgotten', making machine unlearning, the removal of specific training data from a model without retraining from scratch, a key yet underexplored capability in robotics. Despite its growing relevance, no previous work has investigated how unlearning impacts the safety, utility, and adaptability of robotic systems that interact with sensitive patient data. This paper makes two main contributions. First, we present the first simulation-based evaluation framework for machine unlearning in medical robotics, implemented in a custom OpenAI Gym environment that models a virtual rehabilitation assistant engaging with patient-like activity data. Second, we perform a comprehensive benchmarking of state-of-the-art unlearning techniques in this setting, assessing both compliance-driven data removal and preservation of robot functional performance. Results show clear utility-privacy trade-offs, with SISA method achieving the most favorable balance for compliance-aware robotic deployment. Our findings provide the first evidence that machine unlearning can be embedded into adaptive robotic systems, enabling privacy-by-design healthcare robots that respect patient autonomy while maintaining therapeutic efficacy.
APA:
Pisanu, D., Walter, J., & Franke, J. (2025). Balancing Utility and Privacy: Machine Unlearning for Medical Robotics. In 3rd International Conference on Integrated Systems in Medical Technologies, ISMT 2025 - Conference Proceedings. Erlangen, DEU: Institute of Electrical and Electronics Engineers Inc..
MLA:
Pisanu, Daniela, Jonas Walter, and Jörg Franke. "Balancing Utility and Privacy: Machine Unlearning for Medical Robotics." Proceedings of the 3rd International Conference on Integrated Systems in Medical Technologies, ISMT 2025, Erlangen, DEU Institute of Electrical and Electronics Engineers Inc., 2025.
BibTeX: Download