Self-Consuming Generative Models with Adversarially Curated Data

Authors: Xiukun Wei, Xueru Zhang

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on both synthetic and real-world datasets demonstrate the effectiveness of the proposed algorithms.
Researcher Affiliation Academia Xiukun Wei 1 Xueru Zhang 1 1Department of Computer Science and Engineering, The Ohio State University, Columbus, Ohio, USA. Correspondence to: Xueru Zhang <EMAIL>.
Pseudocode Yes Algorithm 1 Gradient-based attack
Open Source Code No The paper does not contain any explicit statement about providing source code, a link to a code repository, or information about code in supplementary materials for the methodology described.
Open Datasets Yes Datasets. Similar to Ferbach et al. (2024), we conduct experiments on three datasets: 1. Synthetic Gaussian: A dataset following 8-mode Gaussian mixture model, the details are in Appendix C.2. 2. CIFAR-10 (Krizhevsky, 2009): It contains 60,000 images from 10 classes... 3. CIFAR-100 (Krizhevsky, 2009): It contains 100 classes, each with 600 images.
Dataset Splits Yes At each iteration, the generative model produces 50,000 data samples, of which 25,000 are selected (after curation by the reward model) for retraining the next-generation model.
Hardware Specification No The paper mentions conducting experiments but does not provide specific details about the hardware used, such as GPU models, CPU types, or memory.
Software Dependencies No The paper mentions using specific models like DDPM, VGG11, and ResNet56, but does not provide specific version numbers for any software dependencies or libraries used for implementation and training.
Experiment Setup Yes At each iteration, the generative model produces 50,000 data samples, of which 25,000 are selected (after curation by the reward model) for retraining the next-generation model. For the reward R and R may share the same architecture or differ. If they have the same architecture, we use pretrained VGG11 as the feature extractor and a linear layer containing 10 neurons for both R and R. If different architectures are used, R employs a pretrained VGG11 as the feature extractor, followed by three linear layers with 128, 64, and 10 neurons, respectively.