Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

Fine-Grained and Efficient Self-Unlearning with Layered Iteration

Authors: Hongyi Lyu, Xuyun Zhang, Hongsheng Hu, Shuo Wang, Chaoxiang He, Lianyong Qi

IJCAI 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experiments on three benchmark datasets demonstrate that SULI achieves superior performance in effectiveness, efficiency, and privacy compared to the state-of-the-art baselines in both class-wise and instance-wise unlearning scenarios.
Researcher Affiliation Academia Hongyi Lyu1 , Xuyun Zhang1 , Hongsheng Hu2 , Shuo Wang3 , Chaoxiang He3 , Lianyong Qi4 1Macquarie University 2University of Newcastle 3Shanghai Jiao Tong University 4China University of Petroleum (East China)
Pseudocode Yes Algorithm 1 Self-Unlearning with Layered Iteration (SULI)
Open Source Code Yes The source code is released at https://github.com/Hongyi-Lyu-MQ/SULI.
Open Datasets Yes Datasets. We follow the previous works [Chen et al., 2023; Cha et al., 2024] and use three datasets: CIFAR-10 [Krizhevsky, 2009], VGGFace2 [Cao et al., 2018], and UTKFace [Zhang et al., 2017].
Dataset Splits No The paper defines Dtrain, Df (forgetting dataset), and Dr (retaining set) for the unlearning task but does not explicitly provide the train/test/validation splits for the datasets (CIFAR-10, VGGFace2, UTKFace) used for initial model training in the main text. It mentions details are in Appendix B and C, which are not provided.
Hardware Specification Yes Our experimental environment includes an NVIDIA RTX 4070 GPU, Python 3.11, and Py Torch 2.1.1.
Software Dependencies Yes Our experimental environment includes an NVIDIA RTX 4070 GPU, Python 3.11, and Py Torch 2.1.1.
Experiment Setup Yes We utilize the ADAM optimizer [Kingma and Ba, 2014] with carefully selected learning rates optimized for both class-wise and instance-wise unlearning tasks. ... We perform a grid search (the results are shown in appendix D) to optimize the hyperparameter t within the range [1, 25], selecting t = 2 for all experiments as it balances model utility and unlearning effectiveness. Our experiments cover two primary unlearning scenarios: class-wise unlearning, where early stops when the model s accuracy on Df approaches zero, and instance-wise unlearning, where unlearning ceases when the model s accuracy on Df matches that on a 1% reference dataset.