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 Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Versatile Transferable Unlearnable Example Generator
Authors: Zhihao Li, Jiale Cai, Gezheng Xu, Hao Zheng, Qiuyue Li, Fan Zhou, Shichun Yang, Charles Ling, Boyu Wang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate the effectiveness and broad applicability of our approach. Code is available at https://github.com/zhli-cs/VTG. |
| Researcher Affiliation | Academia | 1 Western University 2 Central South University 3 Beihang University 4 Vector Institute EMAIL EMAIL EMAIL EMAIL |
| Pseudocode | Yes | The complete training procedure is delineated in Algorithm 1. |
| Open Source Code | Yes | Code is available at https://github.com/zhli-cs/VTG. |
| Open Datasets | Yes | We evaluate VTG on CIFAR-10 [13], CIFAR-100 [13], SVHN [14] and PACS [17]. |
| Dataset Splits | Yes | Specifically, we randomly extract 50% data from the original training dataset as the source training set and utilize the remaining 50% data as the target training set. |
| Hardware Specification | Yes | Our model is developed with the PyTorch framework and trained on a single RTX A5000 GPU. ... All experiments are conducted on a consistent hardware setup comprising an Intel(R) Xeon(R) Silver 4210R CPU and an RTX A5000 GPU. |
| Software Dependencies | No | Our model is developed with the PyTorch framework and trained on a single RTX A5000 GPU. ... We employ the Adam optimizer [50] with an initial learning rate of 0.001. |
| Experiment Setup | Yes | We employ the Adam optimizer [50] with an initial learning rate of 0.001. We train our model for 30 epochs on CIFAR-10 and SVHN, and 50 epochs on CIFAR-100 and PACS. We set the first loop training step T to 10 in all experiments. The input image resolution is standardized to 224 224 for PACS, while 32 32 for the remaining datasets. Perturbations are crafted in a class-wise manner, where we first generate perturbations for each sample, then we average the perturbations for each class. To ensure imperceptibility, we set the perturbation bound ϵ to 8/255. |