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..
Boosting Adversarial Transferability with Spatial Adversarial Alignment
Authors: Zhaoyu Chen, HaiJing Guo, Kaixun Jiang, Jiyuan Fu, Xinyu Zhou, Dingkang Yang, Hao Tang, Bo Li, Wenqiang Zhang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on various architectures on Image Net show that aligned surrogate models based on SAA can provide higher transferable adversarial examples, especially in cross-architecture attacks. Experiments on 6 CNNs and 4 Vi Ts show that SAA has state-of-the-art adversarial transferability, especially in cross-model transferability. |
| Researcher Affiliation | Collaboration | 1College of Intelligent Robotics and Advanced Manufacturing, Fudan University 2College of Computer Science and Artificial Intelligence, Fudan University 3China Southern Power Grid Artificial Intelligence Technology Co., Ltd. 4School of Computer Science, Peking University 5vivo Mobile Communication Co., Ltd. EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1 Loss LSAA |
| Open Source Code | Yes | We provide the implementation of our Spatial Adversarial Alignment (SAA) in the Core Codes of Supplementary Material. |
| Open Datasets | Yes | Our experiments utilize the Image Net-compatible dataset [27], a widely adopted subset containing 1,000 images from the Image Net validation set [11]. |
| Dataset Splits | Yes | Our experiments utilize the Image Net-compatible dataset [27], a widely adopted subset containing 1,000 images from the Image Net validation set [11]. |
| Hardware Specification | Yes | SAA needs about 10 hours training time on Image Net with batch size of 64 under Nvidia RTX 3090. |
| Software Dependencies | No | The paper does not explicitly list software dependencies with version numbers. It mentions 'timm' for model weights, but without a version. |
| Experiment Setup | Yes | For MI, we set the perturbation magnitude ฯต = 16 [13, 18], perform 10 iterations, with a step size of 16 / 10 = 1.6, and use a momentum ยต = 1. During the Spatial Adversarial Alignment, all surrogate models are fine-tuned for 1 epoch using stochastic gradient descent (SGD) with a momentum of 0.9, and no learning rate adjustments are applied. |