School of Informatics, University of Edinburgh

By Luke Nicholas Darlow, Elliot J. Crowley, Antreas Antoniou, Amos J. Storkey

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

FedML: A Research Library and Benchmark for Federated Machine Learning (2020, neutral): “CINIC-10 [139] has 4.5 times as many images as that of CIFAR-10.”
Group Knowledge Transfer: Collaborative Training of Large CNNs on the Edge (2020, methodology): “We evaluate GKT by training modern CNN architectures (ResNet-56 and ResNet-110) on three distinct datasets (CIFAR-10, CIFAR-100, and CINIC-10) and their non-IID variants.”
Similarity-Preserving Knowledge Distillation (2019, methodology): “The CINIC-10 dataset [5] is designed to be a middle option relative to CIFAR-10 and ImageNet: it is composed of 32x32 images in the style of CIFAR-10, but at a total of 270,000 images its scale is closer to that of ImageNet.”
XNAS: Neural Architecture Search with Expert Advice (2019, methodology): “In addition, using the cell found on CIFAR-10 we did transferability experiments on the well-known benchmarks ImageNet, CIFAR-100, SVHN, Fashion-MNIST, Freiburg and CINIC10.”
Discriminability-Transferability Trade-Off: An Information-Theoretic Perspective (2022, methodology): “As shown in Figure 6, increasing α to 0.5 leads to a normal classification accuracy 79.4% on CIFAR-100, but significantly better transferring results on STL-10 and CINIC-10.”
Membership Inference Attacks From First Principles (2021, methodology): “For this experiment, we use the CINIC-10 dataset [8].”
The Evolution of Out-of-Distribution Robustness Throughout Fine-Tuning (2021, methodology): “For the map from CIFAR-10 to ImageNet, we use the map as deﬁned by [4] in the construction of the CINIC-10 dataset.”
The Computational Limits of Deep Learning (2020, unknown): “…datasets, including Caltech 101 [111], Caltech 256 [112], ImageNet [24], CIFAR-10/100 [113], MNIST [114], SVHN [115], STL-10 [116], Fashion-MNIST [117], CINIC-10 [118], Flowers-102, iNaturalist [119], EMNIST-Letters [120], Kuzushiji-MNIST [121], Stanford Cars [122], and MultiMNIST [123].”
Noise as a Resource for Learning in Knowledge Distillation (2020, methodology): “To measure the out-of-distribution performance, we use the ImageNet images from the CINIC dataset [10].”
CODEs: Chamfer Out-of-Distribution Examples against Overconfidence Issue (2021, methodology): “We demonstrate CODEs could improve classification by evaluating ResNets on CINIC-10 [8] and ImageNet [9].”
Deep Learning on a Data Diet: Finding Important Examples Early in Training (2021, methodology): “We train convolutional neural networks of varying depth–ResNet18 and ResNet50 [13]–on standard vision datasets of varying difﬁculty–CIFAR-10, CIFAR-100 [14], and CINIC-10 [15].”
Bad Global Minima Exist and SGD Can Reach Them (2019, methodology): “Datasets and Architectures We ran experiments on the CIFAR [21] dataset (including CIFAR10 and CIFAR100), CINIC10 [22] and a resized Restricted ImageNet [23].2 The CIFAR training set consists of 50k data points and the test set consists of 10k data points.”
Towards Fairness in Visual Recognition: Effective Strategies for Bias Mitigation (2019, methodology): “Arguably the most interesting shift corresponds to taking images of similar classes from ImageNet [39, 10], and we focus our discussion on that one.”
RobustBench: a standardized adversarial robustness benchmark (2020, background): “1, [112]) and image source shift (CINIC-10, [31]).”
Label-Consistent Backdoor Attacks (2019, methodology): “As the target image for each interpolation we use images from the CINIC-10 dataset (Darlow et al., 2018), which contains inputs with the same size and class structure as CIFAR-10 (obtained from downsampling images from the corresponding ImageNet (Russakovsky et al.”
Stateful Detection of Black-Box Adversarial Attacks (2019, methodology): “We also evaluate against CINIC-10 [15] (210,000 images of the CIFAR-10 classes taken from the ImageNet dataset and downsampled to 32 × 32 × 3), and find the FPR to be 0.”
Human Uncertainty Makes Classification More Robust (2019, methodology): “The CINIC10 dataset collected by [7] contains both CIFAR10 images and rescaled ImageNet images from equivalent classes [7].”
Differentiable Feature Aggregation Search for Knowledge Distillation (2020, methodology): “Experiments on CIFAR-100 [18] and CINIC [6] datasets show that DFA could outperform the state-of-the-art distillation methods, demonstrating the eﬀectiveness of the feature aggregation search.”
Differentiable Branching In Deep Networks for Fast Inference (2020, background): “Results for the three datasets are summarized in Table 1 (for B-NET, we report the accuracy of the last branch).”