Label-Retrieval-Augmented Diffusion Models for Learning from Noisy Labels

Authors: Jian Chen, Ruiyi Zhang, Tong Yu, Rohan Sharma, Zhiqiang Xu, Tong Sun, Changyou Chen

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments are conducted for evaluation. Our model achieves new state-of-the-art (SOTA) results on all standard real-world benchmark datasets.
Researcher Affiliation Collaboration Jian Chen1 Ruiyi Zhang2 Tong Yu2 Rohan Sharma1 Zhiqiang Xu3 Tong Sun2 Changyou Chen1 1University at Buffalo 2Adobe Research 3MBZUAI
Pseudocode Yes Algorithm 1 Training Input: training set {X, Y}, image encoder fp, fq.
Open Source Code Yes Code is available at https://github.com/puar-playground/LRA-diffusion
Open Datasets Yes We conduct simulation experiments on the CIFAR-10 and CIFAR-100 datasets [51] to evaluate our method s performance under various noise types.
Dataset Splits Yes The dataset includes a clean training set, validation set, and test set with manually refined labels, consisting of approximately 47.6k, 14.3k, and 10k pictures, respectively.
Hardware Specification Yes All experiments were done on four NVIDIA Titan V GPUs.
Software Dependencies No No specific software versions (e.g., library names with version numbers) are mentioned in the paper.
Experiment Setup Yes We train LRA-diffusion models for 200 epochs with Adam optimizer. The batch size is 256. We used a learning rate schedule that included a warmup phase followed by a half-cycle cosine decay. The initial learning rate is set to 0.001.