Advances in Machine Unlearning and Fairness

The field of machine learning is increasingly focused on developing methods for machine unlearning, which involves removing specific data from trained models. Recent research has made significant progress in this area, with a number of innovative approaches being proposed. One key direction is the development of methods for efficient and effective machine unlearning, such as model splitting and core sample selection. Another important area of research is the evaluation of machine unlearning methods, with a number of frameworks and metrics being proposed to assess their effectiveness. Additionally, there is a growing recognition of the need to consider fairness and privacy in machine unlearning, with methods being developed to ensure that unlearning does not introduce bias or compromise user privacy. Notably, papers such as 'Unilogit: Robust Machine Unlearning for LLMs Using Uniform-Target Self-Distillation' and 'Enabling Group Fairness in Graph Unlearning via Bi-level Debiasing' have made significant contributions to the field, proposing novel methods for robust machine unlearning and fair graph unlearning.

Sources

Anticipating Gaming to Incentivize Improvement: Guiding Agents in (Fair) Strategic Classification

Unilogit: Robust Machine Unlearning for LLMs Using Uniform-Target Self-Distillation

PRUNE: A Patching Based Repair Framework for Certiffable Unlearning of Neural Networks

Efficient Machine Unlearning by Model Splitting and Core Sample Selection

From Search To Sampling: Generative Models For Robust Algorithmic Recourse

Generalization Bounds and Stopping Rules for Learning with Self-Selected Data

Mirror Mirror on the Wall, Have I Forgotten it All? A New Framework for Evaluating Machine Unlearning

Online Learning and Unlearning

MUBox: A Critical Evaluation Framework of Deep Machine Unlearning

Model-free Online Learning for the Kalman Filter: Forgetting Factor and Logarithmic Regret

Layered Unlearning for Adversarial Relearning

Enabling Group Fairness in Graph Unlearning via Bi-level Debiasing

Built with on top of