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..
FADRM: Fast and Accurate Data Residual Matching for Dataset Distillation
Authors: Jiacheng Cui, Xinyue Bi, Yaxin Luo, Xiaohan Zhao, Jiacheng Liu, Zhiqiang Shen
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Exhaustive empirical evaluations of our proposed FADRM on CIFAR-100 [15], Tiny-Image Net [16], Image Net-1K [8] and its subset demonstrate that it not only accelerates the dataset distillation process by 50% but also achieves superior accuracy that beats all previous state-of-the-art methods on both accuracy and generation speed. |
| Researcher Affiliation | Academia | 1VILA Lab, MBZUAI 2University of Ottawa |
| Pseudocode | Yes | Algorithm 1 FADRM: Residual Matching for Dataset Distillation |
| Open Source Code | Yes | Code: FADRM (Git Hub) |
| Open Datasets | Yes | We conduct experiments across datasets with varying resolutions, including CIFAR-100 (32 32) [15], Tiny-Image Net (64 64) [16], Image Net-1K (224 224) [8], and their subsets. |
| Dataset Splits | Yes | Each distilled dataset is used to train a student model from scratch, and the resulting model is evaluated on the original test set of the corresponding dataset (e.g., CIFAR-100, Tiny-Image Net, Image Net-1K). |
| Hardware Specification | Yes | Total training hours on a single RTX4090 vs. test set accuracy, comparing prior stateof-the-art methods with our proposed framework (+ denotes multi-model distillation). |
| Software Dependencies | No | For Image Net-1K, we employed the official Py Torch pretrained models, which have been extensively trained on the full Image Net-1K dataset. |
| Experiment Setup | Yes | Hyperparameters for Training the Original Dataset Optimizer SGD Learning Rate 0.1 Weight Decay 1e-4 Momentum 0.9 Batch Size 128 Loss Function Cross-Entropy Epochs 300 Augmentation Random Resized Crop, Horizontal Flip, Cut Mix Table 14: Hyperparameters for Training the Original Dataset. |