# Self-Ensemble Protection: Training Checkpoints Are Good Data Protectors Source: [Self-Ensemble Protection: Training Checkpoints Are Good Data Protectors]() Collection index: [llms.txt]() Data updated: 2025-10-26T01:11:39.114Z Authors: Sizhe Chen, Geng Yuan, Xinwen Cheng, Yifan Gong, Minghai Qin, Yanzhi Wang, Xiaolin Huang Published: 2022-11-22T04:54:20.000Z ## Abstract As data becomes increasingly vital, a company would be very cautious about releasing data, because the competitors could use it to train high-performance models, thereby posing a tremendous threat to the company's commercial competence. To prevent training good models on the data, we could add imperceptible perturbations to it. Since such perturbations aim at hurting the entire training process, they should reflect the vulnerability of DNN training, rather than that of a single model. Based on this new idea, we seek perturbed examples that are always unrecognized (never correctly classified) in training. In this paper, we uncover them by model checkpoints' gradients, forming the proposed self-ensemble protection (SEP), which is very effective because (1) learning on examples ignored during normal training tends to yield DNNs ignoring normal examples; (2) checkpoints' cross-model gradients are close to orthogonal, meaning that they are as diverse as DNNs with different architectures. That is, our amazing performance of ensemble only requires the computation of training one model. By extensive experiments with 9 baselines on 3 datasets and 5 architectures, SEP is verified to be a new state-of-the-art, e.g., our small $\\ell\_\\infty=2/255$ perturbations reduce the accuracy of a CIFAR-10 ResNet18 from 94.56% to 14.68%, compared to 41.35% by the best-known method. Code is available at https://github.com/Sizhe-Chen/SEP. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Image Classification on CIFAR-10 (test) (Transformation Robustness)](): Image Classification; CIFAR-10 (test); Accuracy (Base): 28.43. - [Multi-class Classification on JV (test)](): Multi-class classification; JV (test); Accuracy: 50.18. - [Multi-class Classification on IF (test)](): Multi-class classification; IF (test); Accuracy: 63.72. - [Binary Classification on Employee (test)](): Binary Classification; Employee (test); Accuracy: 63.41. - [Binary Classification on KC1 (test)](): Binary Classification; KC1 (test); Accuracy: 59.65. - [Multi-class Classification on EOL (test)](): Multi-class classification; EOL (test); Accuracy: 31.38. - [Binary Classification on CH (test)](): Binary Classification; CH (test); Accuracy: 73.37. - [Classification on dry-bean (test)](): Classification; dry-bean (test); Accuracy: 42.26. - [Image Classification on CIFAR-10 (test)](): Image Classification; CIFAR-10 (test); Accuracy: 42.8. - [Image Classification on CIFAR-100 (test)](): Image Classification; CIFAR-100 (test); Accuracy: 28.4. ## Original sources and implementation [Original paper]() [PDF]()