p10 / Conference
SFTC: Machine Unlearning via Selective Fine-tuning and Targeted Confusion
European Interdisciplinary Cybersecurity Conference (EICC 2024), 29-36 · 2024
In plain language
SFTC addresses requests to remove the influence of selected training examples from an existing image classifier.
Research context
It studies a practical alternative to fully retraining a model whenever particular training samples must be forgotten.
Approach
Selective fine-tuning on retained data is combined with targeted confusion guided by a biased random generator. Evaluation covers three image datasets.
Findings
The accompanying institutional report describes effective forgetting while preserving the model’s original accuracy. The paper received the EICC 2024 Best Paper Award.
Summary based on abstract. Methods follow the Scholar abstract. Findings and Best Paper Award are corroborated by the Archimedes institutional announcement.