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..
Label-Noise Robust Diffusion Models
Authors: Byeonghu Na, Yeongmin Kim, HeeSun Bae, Jung Hyun Lee, Se Jung Kwon, Wanmo Kang, Il-chul Moon
ICLR 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 4 EXPERIMENTS |
| Researcher Affiliation | Collaboration | Byeonghu Na1, Yeongmin Kim1, Hee Sun Bae1, Jung Hyun Lee2, Se Jung Kwon2, Wanmo Kang1 & Il-Chul Moon1,3 1KAIST, 2NAVER Cloud, 3summary.ai |
| Pseudocode | Yes | Algorithm 1: Training algorithm with TDSM |
| Open Source Code | Yes | Our code is available at: https://github.com/byeonghu-na/tdsm. |
| Open Datasets | Yes | We evaluate our method on three benchmark datasets commonly used for both image generation and label noise learning: MNIST (Le Cun et al., 2010), CIFAR-10, and CIFAR-100 (Krizhevsky, 2009). |
| Dataset Splits | No | The paper mentions 'training dataset' and 'test dataset' but does not explicitly provide percentages or absolute sample counts for training, validation, and test splits within the main text. |
| Hardware Specification | Yes | We utilized 8 NVIDIA Tesla P40 GPUs and employed CUDA 11.4 and Py Torch 1.12 versions in our experiments. |
| Software Dependencies | Yes | We utilized 8 NVIDIA Tesla P40 GPUs and employed CUDA 11.4 and Py Torch 1.12 versions in our experiments. |
| Experiment Setup | Yes | The score network was trained with a batch size of 512, and the training iterations were set to 400,000 for MNIST and CIFAR-10, 200,000 for CIFAR-100. For Clothing-1M, the score network is trained with a batch size 256 for 200,000 training iterations. |