{"work":{"id":"6714d44f-1b5e-4141-9450-ea09a7e724b0","openalex_id":"https://openalex.org/W2750384547","doi":"10.48550/arxiv.1708.07747","arxiv_id":"1708.07747","raw_key":null,"title":"Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms","authors":null,"authors_text":"Han Xiao, Kashif Rasul, Roland Vollgraf","year":2017,"venue":"cs.LG","abstract":"We present Fashion-MNIST, a new dataset comprising of 28x28 grayscale images of 70,000 fashion products from 10 categories, with 7,000 images per category. The training set has 60,000 images and the test set has 10,000 images. Fashion-MNIST is intended to serve as a direct drop-in replacement for the original MNIST dataset for benchmarking machine learning algorithms, as it shares the same image size, data format and the structure of training and testing splits. The dataset is freely available at https://github.com/zalandoresearch/fashion-mnist","external_url":"https://arxiv.org/abs/1708.07747","cited_by_count":6091,"metadata_source":"pith","metadata_fetched_at":"2026-08-05T02:28:24.338817+00:00","pith_arxiv_id":"1708.07747","created_at":"2026-05-09T03:40:44.608842+00:00","updated_at":"2026-08-05T02:28:24.338817+00:00","title_quality_ok":true,"display_title":"Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms","render_title":"Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms"},"hub":{"state":{"work_id":"6714d44f-1b5e-4141-9450-ea09a7e724b0","tier":"super_hub","tier_reason":"100+ Pith inbound or 10,000+ external citations","pith_inbound_count":240,"external_cited_by_count":6091,"distinct_field_count":25,"first_pith_cited_at":"2018-02-09T19:39:33+00:00","last_pith_cited_at":"2026-07-08T13:41:31+00:00","author_build_status":"needed","summary_status":"needed","contexts_status":"needed","graph_status":"needed","ask_index_status":"needed","reader_status":"not_needed","recognition_status":"not_needed","updated_at":"2026-08-23T02:29:34.104565+00:00","tier_text":"super_hub"},"tier":"super_hub","role_counts":[{"context_role":"dataset","n":13},{"context_role":"background","n":7}],"polarity_counts":[{"context_polarity":"use_dataset","n":13},{"context_polarity":"background","n":6},{"context_polarity":"unclear","n":1}],"runs":{"ask_index":{"job_type":"ask_index","status":"succeeded","result":{"title":"Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms","claims":[{"claim_text":"We present Fashion-MNIST, a new dataset comprising of 28x28 grayscale images of 70,000 fashion products from 10 categories, with 7,000 images per category. The training set has 60,000 images and the test set has 10,000 images. Fashion-MNIST is intended to serve as a direct drop-in replacement for the original MNIST dataset for benchmarking machine learning algorithms, as it shares the same image size, data format and the structure of training and testing splits. The dataset is freely available at https://github.com/zalandoresearch/fashion-mnist","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"when donor selection is suboptimal, respectively. The global model is updated: θ′ G = 1P i λi X i∈Tc∪R∪Sr λi ˜θi,(8) where ˜θi =θ i fori∈ T c ∪ Rand ˜θi is constructed via Eq. (6) fori∈ Sr. 5 Experimental Setup and Results Datasets and model architectures.We evaluate on four image-classification benchmarks of increasing complexity: MNIST [5], Fashion-MNIST [24], CIFAR- 10 [14], and CIFAR-100 [14]. To simulate heterogeneous client data distributions typical of FL, we partition each dataset using ","claim_type":"dataset","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"ration as a practical adversarial defense mechanism for QML models by embedding robustness directly into the quantum encoding stage. •We demonstrate that the proposed method is applicable to both angle encoding and amplitude encoding, making it suitable for both hybrid and fully quantum classifiers. •We evaluate applicability across three datasets (MNIST [22], FashionMNIST [23], and KMNIST [24]), two adversarial attack settings (FGSM and PGD), and three QML models (QNN, QCNN, and VQC) to demonst","claim_type":"dataset","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"should be interpreted strictly within the synthetic symmetric- query regime studied in the theory, where the true label is hidden from the learner and only the queried subset together with the binary membership feedback is observed. B. Datasets and Experimental Setup We evaluate the method on six standardk= 10image classification benchmarks: MNIST [26], FashionMNIST [27], KMNIST [28], USPS [29], SVHN [30], and CIFAR-10 [31]. Table I summarizes the dataset statistics and model backbones. We addit","claim_type":"dataset","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"45 million pairs of clean and noisy images based on a fraction of QuickDraw [46], which consisted of simple but highly diverse binary images. We demonstrated that the diffractive network pre-trained on this dataset could be universally fine-tuned to denoise broad categories of images, ranging from handwritten digits and fashion prod- ucts (i.e., EMNIST [47] and Fashion-MNIST [48]) to human faces (i.e., CelebA [49]) and vehicle plates (i.e., CBLPRD [50]). Remarkably, this knowledge-transfer strat","claim_type":"dataset","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"• RQ3: What is the impact of mutating different types of quantum gates on the model performance? 4.1 Target Datasets and Models We choose two datasets commonly used in prior QML works for image classification: MNIST and FashionMNIST. The MNIST dataset [29] contains 70,000 grayscale images of handwritten digits (0-9), each with a resolution of28 × 28. The FashionMNIST dataset [60] has the same format but comprises images from ten categories of clothing items, such as T-shirts and trousers. Proc. ","claim_type":"dataset","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"tested at the primary scale: moderate (α server =α client = 0.5) and severe (α server = 0.1,α client = 0.5); the scaling study uses the moderate setting to match the headline configuration of Table II. Experiments use three datasets: 1)CIFAR-10[36]: 50K/10K train/test, 10 classes,32×32 RGB,T=200rounds. 2)SVHN[37]: 73K/26K train/test, 10 classes,32×32RGB, T=150rounds. 3)Fashion-MNIST[38]: 60K/10K train/test, 10 classes, 28×28grayscale,T=100rounds. 2) Model and Training:All methods use LeNet-5 (ap","claim_type":"dataset","confidence":0.95,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms because it crossed a citation-hub threshold. Current citing contexts most often use it as dataset evidence (10 contexts).","role_counts":[{"n":10,"context_role":"dataset"},{"n":6,"context_role":"background"}]},"error":null,"updated_at":"2026-05-19T05:21:24.141447+00:00"},"author_expand":{"job_type":"author_expand","status":"succeeded","result":{"authors_linked":[{"id":"5f6a4fd2-f832-4d0a-a57a-0dd42c359722","orcid":null,"display_name":"Han Xiao"},{"id":"a0684cbd-e183-42f6-a7ee-1d8dff2dff14","orcid":null,"display_name":"Kashif Rasul"},{"id":"d1bcec17-471c-4c1f-8b35-89073706e663","orcid":null,"display_name":"Roland Vollgraf"}]},"error":null,"updated_at":"2026-05-19T05:21:24.132573+00:00"},"context_extract":{"job_type":"context_extract","status":"succeeded","result":{"enqueued_papers":25},"error":null,"updated_at":"2026-05-14T08:47:51.716343+00:00"},"graph_features":{"job_type":"graph_features","status":"succeeded","result":{"co_cited":[{"title":"Adam: A Method for Stochastic Optimization","work_id":"1910796d-9b52-4683-bf5c-de9632c1028b","shared_citers":10},{"title":"Deep Learning for Classical Japanese Literature","work_id":"2f44c732-4934-4ceb-a58b-2e3cabc4fa79","shared_citers":8},{"title":"Gradient-based learning applied to document recognition.Proceedings of the IEEE, 86(11):2278–2324","work_id":"0a3595ca-57f9-43f8-8e2f-aface7154b99","shared_citers":6},{"title":"PennyLane: Automatic differentiation of hybrid quantum-classical computations","work_id":"83078d0b-6c02-4fc5-822d-4da4204fd057","shared_citers":5},{"title":"UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction","work_id":"54c15172-6304-4008-a3b6-c4cc0803c054","shared_citers":5},{"title":"Explaining and Harnessing Adversarial Examples","work_id":"2cedf8f6-7539-4c49-8136-f42a20487146","shared_citers":4},{"title":"Layer Normalization","work_id":"20a2d720-0046-4c7c-bcd6-327ec8143f69","shared_citers":4},{"title":"Auto-Encoding Variational Bayes","work_id":"97d95295-30e1-42b4-bbf6-85f0fa4edb44","shared_citers":3},{"title":"Can you really backdoor federated learning?","work_id":"e3828d1f-038f-4371-a8d9-73a2916ee4b5","shared_citers":3},{"title":"Flower: A friendly federated learning research framework","work_id":"396ea1bf-02a2-41cf-8454-93dff73af450","shared_citers":3},{"title":"Towards Deep Learning Models Resistant to Adversarial Attacks","work_id":"b20a57fa-4b7d-40ec-8b6a-ce48234630de","shared_citers":3},{"title":"Very Deep Convolutional Networks for Large-Scale Image Recognition","work_id":"1c4b4409-c14b-488b-a086-c57a5aab8a29","shared_citers":3},{"title":"arXiv preprint arXiv:1710.04759 , year=","work_id":"4ce9bd85-2ff7-4f4f-a656-abff0bb0e232","shared_citers":2},{"title":"arXiv preprint arXiv:1808.04866 (2018)","work_id":"b28dd258-3d87-4409-a6d0-98758a4e91a2","shared_citers":2},{"title":"arXiv preprint arXiv:1909.05125 (2019)","work_id":"84489b48-3414-48a9-b628-ad775a8d1b70","shared_citers":2},{"title":"arXiv preprint arXiv:2012.13995 (2020)","work_id":"0364103d-9db6-4c1c-a200-06789828f79f","shared_citers":2},{"title":"arXiv preprint arXiv:2501.16496 , year=","work_id":"f55f2189-55b1-4a1c-acfb-a5fa7bfa9e86","shared_citers":2},{"title":"Decoupled Weight Decay Regularization","work_id":"07ef7360-d385-4033-83f7-8384a6325204","shared_citers":2},{"title":"Distilling the Knowledge in a Neural Network","work_id":"d927ab1f-17b8-4002-9d09-c3d55764fbad","shared_citers":2},{"title":"Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation","work_id":"1fe8c7c8-aff7-4b94-9096-e549d7e60789","shared_citers":2},{"title":"Gaussian Error Linear Units (GELUs)","work_id":"0466fd22-03a1-4a61-af0a-a900e77bb023","shared_citers":2},{"title":"International Conference on Learning Representations , year =","work_id":"74e5da63-cffe-4f1a-a090-e0b98d805e10","shared_citers":2},{"title":"Necessary but not sufficient: Limitations of projection quality metrics","work_id":"8e01f880-b79b-43ac-9d6c-f6961ec51f5e","shared_citers":2},{"title":"Neural Computation , volume =","work_id":"892059af-f7a4-4863-957e-f0d06c174de9","shared_citers":2}],"time_series":[{"n":2,"year":2018},{"n":67,"year":2026}],"dependency_candidates":[]},"error":null,"updated_at":"2026-05-14T08:47:56.398277+00:00"},"identity_refresh":{"job_type":"identity_refresh","status":"succeeded","result":{"items":[{"title":"Qwen3 Technical Report","outcome":"unchanged","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","resolver":"local_arxiv","confidence":0.98,"old_work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e"}],"counts":{"fixed":0,"merged":0,"unchanged":1,"quarantined":0,"needs_external_resolution":0},"errors":[],"attempted":1},"error":null,"updated_at":"2026-05-14T08:47:54.047032+00:00"},"role_polarity":{"job_type":"role_polarity","status":"succeeded","result":{"title":"Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms","claims":[{"claim_text":"We present Fashion-MNIST, a new dataset comprising of 28x28 grayscale images of 70,000 fashion products from 10 categories, with 7,000 images per category. The training set has 60,000 images and the test set has 10,000 images. Fashion-MNIST is intended to serve as a direct drop-in replacement for the original MNIST dataset for benchmarking machine learning algorithms, as it shares the same image size, data format and the structure of training and testing splits. The dataset is freely available at https://github.com/zalandoresearch/fashion-mnist","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"when donor selection is suboptimal, respectively. The global model is updated: θ′ G = 1P i λi X i∈Tc∪R∪Sr λi ˜θi,(8) where ˜θi =θ i fori∈ T c ∪ Rand ˜θi is constructed via Eq. (6) fori∈ Sr. 5 Experimental Setup and Results Datasets and model architectures.We evaluate on four image-classification benchmarks of increasing complexity: MNIST [5], Fashion-MNIST [24], CIFAR- 10 [14], and CIFAR-100 [14]. To simulate heterogeneous client data distributions typical of FL, we partition each dataset using ","claim_type":"dataset","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"ration as a practical adversarial defense mechanism for QML models by embedding robustness directly into the quantum encoding stage. •We demonstrate that the proposed method is applicable to both angle encoding and amplitude encoding, making it suitable for both hybrid and fully quantum classifiers. •We evaluate applicability across three datasets (MNIST [22], FashionMNIST [23], and KMNIST [24]), two adversarial attack settings (FGSM and PGD), and three QML models (QNN, QCNN, and VQC) to demonst","claim_type":"dataset","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"should be interpreted strictly within the synthetic symmetric- query regime studied in the theory, where the true label is hidden from the learner and only the queried subset together with the binary membership feedback is observed. B. Datasets and Experimental Setup We evaluate the method on six standardk= 10image classification benchmarks: MNIST [26], FashionMNIST [27], KMNIST [28], USPS [29], SVHN [30], and CIFAR-10 [31]. Table I summarizes the dataset statistics and model backbones. We addit","claim_type":"dataset","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"45 million pairs of clean and noisy images based on a fraction of QuickDraw [46], which consisted of simple but highly diverse binary images. We demonstrated that the diffractive network pre-trained on this dataset could be universally fine-tuned to denoise broad categories of images, ranging from handwritten digits and fashion prod- ucts (i.e., EMNIST [47] and Fashion-MNIST [48]) to human faces (i.e., CelebA [49]) and vehicle plates (i.e., CBLPRD [50]). Remarkably, this knowledge-transfer strat","claim_type":"dataset","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"• RQ3: What is the impact of mutating different types of quantum gates on the model performance? 4.1 Target Datasets and Models We choose two datasets commonly used in prior QML works for image classification: MNIST and FashionMNIST. The MNIST dataset [29] contains 70,000 grayscale images of handwritten digits (0-9), each with a resolution of28 × 28. The FashionMNIST dataset [60] has the same format but comprises images from ten categories of clothing items, such as T-shirts and trousers. Proc. ","claim_type":"dataset","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"tested at the primary scale: moderate (α server =α client = 0.5) and severe (α server = 0.1,α client = 0.5); the scaling study uses the moderate setting to match the headline configuration of Table II. Experiments use three datasets: 1)CIFAR-10[36]: 50K/10K train/test, 10 classes,32×32 RGB,T=200rounds. 2)SVHN[37]: 73K/26K train/test, 10 classes,32×32RGB, T=150rounds. 3)Fashion-MNIST[38]: 60K/10K train/test, 10 classes, 28×28grayscale,T=100rounds. 2) Model and Training:All methods use LeNet-5 (ap","claim_type":"dataset","confidence":0.95,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms because it crossed a citation-hub threshold. Current citing contexts most often use it as dataset evidence (10 contexts).","role_counts":[{"n":10,"context_role":"dataset"},{"n":6,"context_role":"background"}]},"error":null,"updated_at":"2026-05-19T05:21:24.137867+00:00"},"summary_claims":{"job_type":"summary_claims","status":"succeeded","result":{"title":"Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms","claims":[{"claim_text":"We present Fashion-MNIST, a new dataset comprising of 28x28 grayscale images of 70,000 fashion products from 10 categories, with 7,000 images per category. The training set has 60,000 images and the test set has 10,000 images. Fashion-MNIST is intended to serve as a direct drop-in replacement for the original MNIST dataset for benchmarking machine learning algorithms, as it shares the same image size, data format and the structure of training and testing splits. The dataset is freely available at https://github.com/zalandoresearch/fashion-mnist","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms because it crossed a citation-hub threshold.","role_counts":[]},"error":null,"updated_at":"2026-05-14T08:47:58.302493+00:00"}},"summary":{"title":"Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms","claims":[{"claim_text":"We present Fashion-MNIST, a new dataset comprising of 28x28 grayscale images of 70,000 fashion products from 10 categories, with 7,000 images per category. The training set has 60,000 images and the test set has 10,000 images. Fashion-MNIST is intended to serve as a direct drop-in replacement for the original MNIST dataset for benchmarking machine learning algorithms, as it shares the same image size, data format and the structure of training and testing splits. The dataset is freely available at https://github.com/zalandoresearch/fashion-mnist","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms because it crossed a citation-hub threshold.","role_counts":[]},"graph":{"co_cited":[{"title":"Adam: A Method for Stochastic Optimization","work_id":"1910796d-9b52-4683-bf5c-de9632c1028b","shared_citers":10},{"title":"Deep Learning for Classical Japanese Literature","work_id":"2f44c732-4934-4ceb-a58b-2e3cabc4fa79","shared_citers":8},{"title":"Gradient-based learning applied to document recognition.Proceedings of the IEEE, 86(11):2278–2324","work_id":"0a3595ca-57f9-43f8-8e2f-aface7154b99","shared_citers":6},{"title":"PennyLane: Automatic differentiation of hybrid quantum-classical computations","work_id":"83078d0b-6c02-4fc5-822d-4da4204fd057","shared_citers":5},{"title":"UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction","work_id":"54c15172-6304-4008-a3b6-c4cc0803c054","shared_citers":5},{"title":"Explaining and Harnessing Adversarial Examples","work_id":"2cedf8f6-7539-4c49-8136-f42a20487146","shared_citers":4},{"title":"Layer Normalization","work_id":"20a2d720-0046-4c7c-bcd6-327ec8143f69","shared_citers":4},{"title":"Auto-Encoding Variational Bayes","work_id":"97d95295-30e1-42b4-bbf6-85f0fa4edb44","shared_citers":3},{"title":"Can you really backdoor federated learning?","work_id":"e3828d1f-038f-4371-a8d9-73a2916ee4b5","shared_citers":3},{"title":"Flower: A friendly federated learning research framework","work_id":"396ea1bf-02a2-41cf-8454-93dff73af450","shared_citers":3},{"title":"Towards Deep Learning Models Resistant to Adversarial Attacks","work_id":"b20a57fa-4b7d-40ec-8b6a-ce48234630de","shared_citers":3},{"title":"Very Deep Convolutional Networks for Large-Scale Image Recognition","work_id":"1c4b4409-c14b-488b-a086-c57a5aab8a29","shared_citers":3},{"title":"arXiv preprint arXiv:1710.04759 , year=","work_id":"4ce9bd85-2ff7-4f4f-a656-abff0bb0e232","shared_citers":2},{"title":"arXiv preprint arXiv:1808.04866 (2018)","work_id":"b28dd258-3d87-4409-a6d0-98758a4e91a2","shared_citers":2},{"title":"arXiv preprint arXiv:1909.05125 (2019)","work_id":"84489b48-3414-48a9-b628-ad775a8d1b70","shared_citers":2},{"title":"arXiv preprint arXiv:2012.13995 (2020)","work_id":"0364103d-9db6-4c1c-a200-06789828f79f","shared_citers":2},{"title":"arXiv preprint arXiv:2501.16496 , year=","work_id":"f55f2189-55b1-4a1c-acfb-a5fa7bfa9e86","shared_citers":2},{"title":"Decoupled Weight Decay Regularization","work_id":"07ef7360-d385-4033-83f7-8384a6325204","shared_citers":2},{"title":"Distilling the Knowledge in a Neural Network","work_id":"d927ab1f-17b8-4002-9d09-c3d55764fbad","shared_citers":2},{"title":"Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation","work_id":"1fe8c7c8-aff7-4b94-9096-e549d7e60789","shared_citers":2},{"title":"Gaussian Error Linear Units (GELUs)","work_id":"0466fd22-03a1-4a61-af0a-a900e77bb023","shared_citers":2},{"title":"International Conference on Learning Representations , year =","work_id":"74e5da63-cffe-4f1a-a090-e0b98d805e10","shared_citers":2},{"title":"Necessary but not sufficient: Limitations of projection quality metrics","work_id":"8e01f880-b79b-43ac-9d6c-f6961ec51f5e","shared_citers":2},{"title":"Neural Computation , volume =","work_id":"892059af-f7a4-4863-957e-f0d06c174de9","shared_citers":2}],"time_series":[{"n":2,"year":2018},{"n":67,"year":2026}],"dependency_candidates":[]},"authors":[{"id":"5f6a4fd2-f832-4d0a-a57a-0dd42c359722","orcid":null,"display_name":"Han Xiao","source":"manual","import_confidence":0.72},{"id":"a0684cbd-e183-42f6-a7ee-1d8dff2dff14","orcid":null,"display_name":"Kashif Rasul","source":"manual","import_confidence":0.72},{"id":"d1bcec17-471c-4c1f-8b35-89073706e663","orcid":null,"display_name":"Roland Vollgraf","source":"manual","import_confidence":0.72}]}}