Dataset Distillation by Matching Training Trajectories
The DOI matches its Crossref record; 40 of the 40 references it cites were found in a registry or an index.
Abstract
Dataset distillation is the task of synthesizing a small dataset such that a model trained on the synthetic set will match the test accuracy of the model trained on the full dataset. In this paper, we propose a new formulation that optimizes our distilled data to guide networks to a similar state as those trained on real data across many training steps. Given a network, we train it for several iterations on our distilled data and optimize the distilled data with respect to the distance between the synthetically trained parameters and the parameters trained on real data. To efficiently obtain the initial and target network parameters for large-scale datasets, we pre-compute and store training trajectories of expert networks trained on the real dataset. Our method handily outperforms existing methods and also allows us to distill higher-resolution visual data.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 1×
“In Table 3, we show our baseline ConvNet performance and evaluate on ResNet [14], VGG [35], and AlexNet [19].”
MethodVery Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148 · cited 1×
“In Table 3, we show our baseline ConvNet performance and evaluate on ResNet [14], VGG [35], and AlexNet [19].”
MethodImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155 · cited 3×
“Our method is solidly the top per-Plane former on all the transfer models except for AlexNet where we lie within one standard deviation of DSA .”
MethodImageNet: A large-scale hierarchical image database
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009) · 2009 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2009.5206848 · Cited by 65,333 · cited 4×
“Our method is the first capable of distilling higher-resolution (128×128) images, allowing us to explore the ImageNet [6] dataset.”
MethodInfluentialGradient-based learning applied to document recognition
Lecun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998) · Proceedings of the IEEE · doi:10.1109/5.726791 · Cited by 58,805
Learning Multiple Layers of Features from Tiny Images
Krizhevsky, A. (2024) · doi:10.57702/zp44cu3g · Cited by 23,395
Distilling the Knowledge in a Neural Network
Hinton, G., Vinyals, O., & Dean, J. (2015) · arXiv · doi:10.48550/arxiv.1503.02531 · Cited by 14,204 · cited 1×
“In the seminal 2015 paper, Hinton et al. [15] proposed model distillation , which aims to distill the knowledge of a complex model into a simpler one.”
BackgroundModel-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
Finn, C., Abbeel, P., & Levine, S. (2017) · arXiv · doi:10.48550/arxiv.1703.03400 · Cited by 8,512 · cited 1×
“Dataset distillation was first introduced by Wang et al. [44], who proposed expressing the model weights as a function of distilled images and optimized them using gradient-based hyperparameter optimization [23], which is also widely used in meta-learning research [8, 27].”
MethodInstance Normalization: The Missing Ingredient for Fast Stylization
Ulyanov, D., Vedaldi, A., & Lempitsky, V. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1607.08022 · Cited by 2,775 · cited 1×
“The architecture consists of several convolutional blocks, each containing a 3× 3 convolution layer with 128 filters, Instance normalization [42], RELU, and 2×2 average pooling with stride 2.”
MethodGenerative adversarial imitation learning
Ho, J., & Ermon, S. (2016) · arXiv · doi:10.48550/arxiv.1606.03476 · Cited by 1,942
A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning
Ross, S., Gordon, G. J., & Bagnell, J. A. (2018) · Carnegie Mellon University · doi:10.1184/r1/6550949 · Cited by 1,941
DeepMimic: example-guided deep reinforcement learning of physics-based character skills
Peng, X. B., Abbeel, P., Levine, S., & van de Panne, M. (2018) · ACM Transactions on Graphics · doi:10.1145/3197517.3201311 · Cited by 1,784 · cited 1×
“Imitation learning attempts to learn a good policy by observing a collection of expert demonstrations [29, 30, 31].”
BackgroundOn First-Order Meta-Learning Algorithms
Nichol, A., Achiam, J., & Schulman, J. (2018) · arXiv · doi:10.48550/arxiv.1803.02999 · Cited by 1,438 · cited 1×
“Dataset distillation was first introduced by Wang et al. [44], who proposed expressing the model weights as a function of distilled images and optimized them using gradient-based hyperparameter optimization [23], which is also widely used in meta-learning research [8, 27].”
MethodTraining Generative Adversarial Networks with Limited Data
Karras, T., Aittala, M., Hellsten, J., Laine, S., Lehtinen, J., & Aila, T. (2020) · arXiv · doi:10.48550/arxiv.2006.06676 · Cited by 1,170 · cited 1×
“…perform N gradient descent updates on the student parameters with respect to the classification loss of the synthetic data: where A is the differentiable augmentation technique [17, 40, 48, 49] used in previous work [45], and α is the (trainable) learning rate used to update the student network.”
MethodDynamic Few-Shot Visual Learning Without Forgetting
Gidaris, S., & Komodakis, N. (2018) · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2018.00459 · Cited by 1,128
Core Vector Machines: Fast SVM Training on Very Large Data Sets
Tsang, I. W.-H., Kwok, J. T.-Y., & Cheung, P.-M. (2005) · doi:10.5555/1046920.1058114 · Cited by 928
Active Learning for Convolutional Neural Networks: A Core-Set Approach
Sener, O., & Savarese, S. (2017) · arXiv · doi:10.48550/arxiv.1708.00489 · Cited by 922
Gradient-based Hyperparameter Optimization through Reversible Learning
Maclaurin, D., Duvenaud, D., & Adams, R. P. (2015) · arXiv · doi:10.48550/arxiv.1502.03492 · Cited by 509 · cited 2×
“Dataset distillation was first introduced by Wang et al. [44], who proposed expressing the model weights as a function of distilled images and optimized them using gradient-based hyperparameter optimization [23], which is also widely used in meta-learning research [8, 27].”
MethodDataset Condensation with Gradient MatchingPreprint
Zhao, B., Mopuri, K. R., & Bilen, H. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.05929 · Cited by 440 · cited 9×
“To reduce the optimization difficulty, other methods [45, 47] focus on short-range behavior, enforcing a single training step on distilled data to match that on real data.”
BackgroundMethodInfluentialDataset Condensation with Differentiable Siamese AugmentationPreprint
Zhao, B., & Bilen, H. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.08259 · Cited by 404
AMP: adversarial motion priors for stylized physics-based character control
Peng, X. B., Ma, Z., Abbeel, P., Levine, S., & Kanazawa, A. (2021) · ACM Transactions on Graphics · doi:10.1145/3450626.3459670 · Cited by 396
A review of instance selection methods
Olvera-López, J. A., Carrasco-Ochoa, J. A., Martínez-Trinidad, J. F., & Kittler, J. (2010) · Artificial Intelligence Review · doi:10.1007/s10462-010-9165-y · Cited by 394 · cited 1×
“Similar to dataset distillation, coreset [1, 4, 13, 34, 41] and instance selection [28] aim to select a subset of the entire training dataset, where training on this small subset achieves good performance.”
MethodAn Algorithmic Perspective on Imitation Learning
Osa, T., Pajarinen, J., Neumann, G., Bagnell, J. A., Abbeel, P., & Peters, J. (2018) · now publishers, Inc. eBooks · doi:10.1561/9781680834116 · Cited by 368
Kornia: an Open Source Differentiable Computer Vision Library for PyTorch
Riba, E., Mishkin, D., Ponsa, D., Rublee, E., & Bradski, G. (2020) · 2020 IEEE Winter Conference on Applications of Computer Vision (WACV) · doi:10.1109/wacv45572.2020.9093363 · Cited by 333 · cited 2×
“For KIP, instead of the Kornia [32] ZCA implementation, we use the authors’ custom ZCA implementation for evaluation of their method.”
MethodD4RL: Datasets for Deep Data-Driven Reinforcement Learning
Fu, J., Kumar, A., Nachum, O., Tucker, G., & Levine, S. (2020) · arXiv · doi:10.48550/arxiv.2004.07219 · Cited by 310 · cited 1×
“Such methods (behavior cloning in particular) have been shown to work well in offline settings [9, 12].”
MethodCoresets via Bilevel Optimization for Continual Learning and StreamingPreprint
Borsos, Z., Mutný, M., & Krause, A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.03875 · Cited by 310
Super-Samples from Kernel HerdingPreprint
Chen, Y., Welling, M., & Smola, A. J. (2012) · arXiv (Cornell University) · doi:10.48550/arxiv.1203.3472 · Cited by 290 · cited 2×
“We also compare our methods with instance selection algorithms including random selection ( random ), herding methods [4] ( herding ), and example forgetting [39] ( forgetting ).”
MethodDataset Condensation with Distribution Matching
Zhao, B., & Bilen, H. (2023) · 2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) · doi:10.1109/wacv56688.2023.00645 · Cited by 250 · cited 10×
“To generate the distilled images for our method, we employ the distillation process detailed in the previous section and Algorithm 1, using the same suite of differentiable aug-mentations as done in previous work [45, 46].”
BackgroundMethodInfluentialSoft-Label Dataset Distillation and Text Dataset Distillation
Sucholutsky, I., & Schonlau, M. (2021) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn52387.2021.9533769 · Cited by 189 · cited 3×
“Dataset distillation has become a lively research topic in machine learning [2, 25, 26, 38, 45, 46, 47] with various applications, such as continual learning, neural architecture search, and privacy-preserving ML.”
BackgroundMethodImage Augmentations for GAN TrainingPreprint
Zhao, Z., Zhang, Z., Chen, T., Singh, S., & Zhang, H. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.02595 · Cited by 175 · cited 1×
“…perform N gradient descent updates on the student parameters with respect to the classification loss of the synthetic data: where A is the differentiable augmentation technique [17, 40, 48, 49] used in previous work [45], and α is the (trainable) learning rate used to update the student network.”
MethodCAFE: Learning to Condense Dataset by Aligning Features
Wang, K., Zhao, B., Peng, X., Zhu, Z., Yang, S., Wang, S., Huang, G., Bilen, H., Wang, X., & You, Y. (2022) · 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52688.2022.01188 · Cited by 170
Smaller Coresets for k-Median and k-Means Clustering
Har-Peled, S., & Kushal, A. (2006) · Discrete & Computational Geometry · doi:10.1007/s00454-006-1271-x · Cited by 137
Guha, N., Talwalkar, A., & Smith, V. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1902.11175 · Cited by 126
Such, F. P., Rawal, A., Lehman, J., Stanley, K. O., & Clune, J. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1912.07768 · Cited by 123 · cited 1×
“A related line of research learns a generative model to synthesize training data [24, 36].”
BackgroundFederated Learning via Synthetic DataPreprint
Goetz, J., & Tewari, A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2008.04489 · Cited by 109 · cited 1×
“Dataset distillation has enabled various applications including continual learning [44, 45, 47], efficient neural architecture search [45, 47], federated learning [11, 37, 50], and privacy-preserving ML [22, 37] for images, text, and medical imaging data.”
BackgroundPractical Coreset Constructions for Machine LearningPreprint
Bachem, O., Lucic, M., & Krause, A. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1703.06476 · Cited by 102
Soft-Label Anonymous Gastric X-Ray Image Distillation
Li, G., Togo, R., Ogawa, T., & Haseyama, M. (2020) · Proceedings - International Conference on Image Processing · doi:10.1109/icip40778.2020.9191357 · Cited by 85 · cited 1×
“Dataset distillation has enabled various applications including continual learning [44, 45, 47], efficient neural architecture search [45, 47], federated learning [11, 37, 50], and privacy-preserving ML [22, 37] for images, text, and medical imaging data.”
BackgroundReducing catastrophic forgetting with learning on synthetic data
Masarczyk, W., & Tautkutė, I. (2020) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw50498.2020.00134 · Cited by 49 · cited 1×
“A related line of research learns a generative model to synthesize training data [24, 36].”
BackgroundRL Unplugged: A Suite of Benchmarks for Offline Reinforcement Learning
Gulcehre, C., Wang, Z., Novikov, A., Paine, T. L., Colmenarejo, S. G., Zolna, K., Agarwal, R., Merel, J., Mankowitz, D., Paduraru, C., et al. (2020) · arXiv · doi:10.48550/arxiv.2006.13888 · Cited by 42
Towards Good Practices for Data Augmentation in GAN Training.Preprint
Tran, N.-T., Tran, V.-H., Nguyen, N.-B., Nguyen, T., & Cheung, N. M. (2020) · arXiv (Cornell University) · Cited by 24
Cited by
Chung, J., Kamar, E., & Amershi, S. (2023) · Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2023.acl-long.34 · Cited by 72
“As this approach would extract information from large models, they would be analogous to knowledge distillation (Phuong and Lampert, 2019; Hinton et al., 2015) or dataset distillation (Wang et al., 2018; Cazenavette et al., 2022).”
MethodA collective AI via lifelong learning and sharing at the edge
Soltoggio, A., Ben-Iwhiwhu, E., Braverman, V., Eaton, E., Epstein, B., Ge, Y., Halperin, L., How, J., Itti, L., Jacobs, M. A., et al. (2024) · Nature Machine Intelligence · doi:10.1038/s42256-024-00800-2 · Cited by 51
“Distillation approaches 157,158 may involve data sharing (Fig.”
Metadata
- Authors
- Cazenavette, G., Wang, T., Torralba, A., Efros, A. A., & Zhu, J.-Y.
- Venue
- 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
- Publisher
- IEEE
- Published
- 2022-06
- Pages
- 10708-10717
- ISSN
- 1063-6919 (ISSN-L)
- OpenAlex
- W4312412605
- Cited by
- 240
- References
- 40
- FWCI
- 28.76
- Type
- conference paper
- Language
- en
- Open access
- closed
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.