The field of machine learning is moving towards developing more privacy-preserving techniques, with a focus on designing models that can forget or remove specific data upon request. This direction is driven by growing societal and regulatory demands, particularly the need to comply with privacy frameworks such as the GDPR. Researchers are exploring innovative approaches to achieve this goal, including prompt-based learning frameworks, concept unlearning, and federated unlearning. These methods aim to provide a balance between model performance and privacy, enabling the development of more ethical and responsive AI models. Noteworthy papers in this area include 'Pre-Forgettable Models: Prompt Learning as a Native Mechanism for Unlearning' and 'Beyond Sharp Minima: Robust LLM Unlearning via Feedback-Guided Multi-Point Optimization', which propose novel frameworks for unlearning and forgetting in large language models.
Advances in Privacy-Preserving Machine Learning
Sources
Pre-Forgettable Models: Prompt Learning as a Native Mechanism for Unlearning
RaceGAN: A Framework for Preserving Individuality while Converting Racial Information for Image-to-Image Translation
Adversarial generalization of unfolding (model-based) networks
Concept Unlearning in Large Language Models via Self-Constructed Knowledge Triplets
Sparse-Autoencoder-Guided Internal Representation Unlearning for Large Language Models
ToFU: Transforming How Federated Learning Systems Forget User Data
A Weighted Gradient Tracking Privacy-Preserving Method for Distributed Optimization
MER-Inspector: Assessing model extraction risks from an attack-agnostic perspective
FlowCrypt: Flow-Based Lightweight Encryption with Near-Lossless Recovery for Cloud Photo Privacy
R-CONV++: Uncovering Privacy Vulnerabilities through Analytical Gradient Inversion Attacks
Towards Privacy-Aware Bayesian Networks: A Credal Approach
CURE: Centroid-guided Unsupervised Representation Erasure for Facial Recognition Systems
Generative Adversarial Networks Applied for Privacy Preservation in Biometric-Based Authentication and Identification
Beyond Sharp Minima: Robust LLM Unlearning via Feedback-Guided Multi-Point Optimization
PerFace: Metric Learning in Perceptual Facial Similarity for Enhanced Face Anonymization