Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
| Entity Passport | |
| Registry ID | arxiv-paper--unknown--1708.07747 |
| License | ArXiv |
| Provider | semantic_scholar |
Cite this paper
Academic & Research Attribution
@misc{arxiv_paper__unknown__1708.07747,
author = {Han Xiao, Kashif Rasul, Roland Vollgraf},
title = {Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms Paper},
year = {2026},
howpublished = {\url{https://free2aitools.com/paper/arxiv-paper--unknown--1708.07747}},
note = {Accessed via Free2AITools Knowledge Fortress}
}🔬Technical Deep Dive
Full Specifications [+]▾
⚖️ Nexus Index V2.0
💬 Index Insight
FNI V2.0 for Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms: Semantic (S:50), Authority (A:90), Popularity (P:73), Recency (R:100), Quality (Q:45).
Verification Authority
📝 Executive Summary
❝ Cite Node
@article{Unknown2026Fashion-MNIST:,
title={Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms},
author={},
journal={arXiv preprint arXiv:arxiv-paper--unknown--1708.07747},
year={2026}
}Abstract & Analysis
[1708.07747] Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
Han Xiao Zalando Research Mühlenstraße 25, 10243 Berlin han.xiao@zalando.de \And Kashif Rasul Zalando Research Mühlenstraße 25, 10243 Berlin kashif.rasul@zalando.de \And Roland Vollgraf Zalando Research Mühlenstraße 25, 10243 Berlin roland.vollgraf@zalando.de
Abstract
We present Fashion-MNIST, a new dataset comprising of 28 × 28 28 28 28\times 28 grayscale images of 70 , 000 70 000 70,000 fashion products from 10 10 10 categories, with 7 , 000 7 000 7,000 images per category. The training set has 60 , 000 60 000 60,000 images and the test set has 10 , 000 10 000 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 .
1
Introduction
The MNIST dataset comprising of 10-class handwritten digits, was first introduced by LeCun et al. ( 1998 ) in 1998. At that time one could not have foreseen the stellar rise of deep learning techniques and their performance. Despite the fact that today deep learning can do so much the simple MNIST dataset has become the most widely used testbed in deep learning, surpassing CIFAR-10 (Krizhevsky and Hinton, 2009 ) and ImageNet (Deng et al., 2009 ) in its popularity via Google trends 1 1 1 https://trends.google.com/trends/explore?date=all&q=mnist,CIFAR,ImageNet . Despite its simplicity its usage does not seem to be decreasing despite calls for it in the deep learning community.
The reason MNIST is so popular has to do with its size, allowing deep learning researchers to quickly check and prototype their algorithms. This is also complemented by the fact that all machine learning libraries (e.g. scikit-learn) and deep learning frameworks (e.g. Tensorflow, Pytorch) provide helper functions and convenient examples that use MNIST out of the box.
Our aim with this work is to create a good benchmark dataset which has all the accessibility of MNIST, namely its small size, straightforward encoding and permissive license. We took the approach of sticking to the 10 10 10 classes 70 , 000 70 000 70,000 grayscale images in the size of 28 × 28 28 28 28\times 28 as in the original MNIST. In fact, the only change one needs to use this dataset is to change the URL from where the MNIST dataset is fetched. Moreover, Fashion-MNIST poses a more challenging classification task than the simple MNIST digits data, whereas the latter has been trained to accuracies above 99.7% as reported in Wan et al. ( 2013 ); Ciregan et al. ( 2012 ) .
We also looked at the EMNIST dataset provided by Cohen et al. ( 2017 ) , an extended version of MNIST that extends the number of classes by introducing uppercase and lowercase characters. However, to be able to use it seamlessly one needs to not only extend the deep learning framework’s MNIST helpers, but also change the underlying deep neural network to classify these extra classes.
2
Fashion-MNIST Dataset
Fashion-MNIST is based on the assortment on Zalando’s website 2 2 2 Zalando is the Europe’s largest online fashion platform. http://www.zalando.com . Every fashion product on Zalando has a set of pictures shot by professional photographers, demonstrating different aspects of the product, i.e. front and back looks, details, looks with model and in an outfit. The original picture has a light-gray background (hexadecimal color: #fdfdfd ) and stored in 762 × 1000 762 1000 762\times 1000 JPEG format. For efficiently serving different frontend components, the original picture is resampled with multiple resolutions, e.g. large, medium, small, thumbnail and tiny.
We use the front look thumbnail images of 70 , 000 70 000 70,000 unique products to build Fashion-MNIST. Those products come from different gender groups: men, women, kids and neutral. In particular, white-color products are not included in the dataset as they have low contrast to the background. The thumbnails ( 51 × 73 51 73 51\times 73 ) are then fed into the following conversion pipeline, which is visualized in Figure 1 .
Converting the input to a PNG image.
Trimming any edges that are close to the color of the corner pixels. The “closeness” is defined by the distance within 5 % percent 5 5% of the maximum possible intensity in RGB space.
Resizing the longest edge of the image to 28 28 28 by subsampling the pixels, i.e. some rows and columns are skipped over.
Sharpening pixels using a Gaussian operator of the radius and standard deviation of 1.0 1.0 1.0 , with increasing effect near outlines.
Extending the shortest edge to 28 28 28 and put the image to the center of the canvas.
Negating the intensities of the image.
Converting the image to 8-bit grayscale pixels.
Figure 1: Diagram of the conversion process used to generate Fashion-MNIST dataset. Two examples from dress and sandals categories are depicted, respectively. Each column represents a step described in section 2 .
Table 1: Files contained in the Fashion-MNIST dataset.
Name Description
Examples
Size
train-images-idx3-ubyte.gz Training set images 60 , 000 60 000 60,000
25 25 25 MBytes
train-labels-idx1-ubyte.gz Training set labels 60 , 000 60 000 60,000
140 140 140 Bytes
t10k-images-idx3-ubyte.gz Test set images 10 , 000 10 000 10,000
4.2 4.2 4.2 MBytes
t10k-labels-idx1-ubyte.gz Test set labels 10 , 000 10 000 10,000
92 92 92 Bytes
For the class labels, we use the silhouette code of the product. The silhouette code is manually labeled by the in-house fashion experts and reviewed by a separate team at Zalando. Each product contains only one silhouette code. Table 2 gives a summary of all class labels in Fashion-MNIST with examples for each class.
Finally, the dataset is divided into a training and a test set. The training set receives a randomly-selected 6 , 000 6 000 6,000 examples from each class. Images and labels are stored in the same file format as the MNIST data set, which is designed for storing vectors and multidimensional matrices. The result files are listed in Table 1 . We sort examples by their labels while storing, resulting in smaller label files after compression comparing to the MNIST. It is also easier to retrieve examples with a certain class label. The data shuffling job is therefore left to the algorithm developer.
Table 2: Class names and example images in Fashion-MNIST dataset.
Label Description Examples
0 0 T-Shirt/Top
1 1 1 Trouser
2 2 2 Pullover
3 3 3 Dress
4 4 4 Coat
5 5 5 Sandals
6 6 6 Shirt
7 7 7 Sneaker
8 8 8 Bag
9 9 9 Ankle boots
3
Experiments
We provide some classification results in LABEL:tbl:benchmark to form a benchmark on this data set. All algorithms are repeated 5 5 5 times by shuffling the training data and the average accuracy on the test set is reported. The benchmark on the MNIST dataset is also included for a side-by-side comparison. A more comprehensive table with explanations on the algorithms can be found on https://github.com/zalandoresearch/fashion-mnist .
Table 3: Benchmark on Fashion-MNIST (Fashion) and MNIST.
Test Accuracy
Classifier Parameter Fashion MNIST
DecisionTreeClassifier criterion= entropy max_depth= 10 10 10 splitter= best
0.798 0.798 0.798
0.873 0.873 0.873 criterion= entropy max_depth= 10 10 10 splitter= random
0.792 0.792 0.792
0.861 0.861 0.861 criterion= entropy max_depth= 50 50 50 splitter= best
0.789 0.789 0.789
0.886 0.886 0.886 criterion= entropy max_depth= 100 100 100 splitter= best
0.789 0.789 0.789
0.886 0.886 0.886 criterion= gini max_depth= 10 10 10 splitter= best
0.788 0.788 0.788
0.866 0.866 0.866 criterion= entropy max_depth= 50 50 50 splitter= random
0.787 0.787 0.787
0.883 0.883 0.883 criterion= entropy max_depth= 100 100 100 splitter= random
0.787 0.787 0.787
0.881 0.881 0.881 criterion= gini max_depth= 100 100 100 splitter= best
0.785 0.785 0.785
0.879 0.879 0.879 criterion= gini max_depth= 50 50 50 splitter= best
0.783 0.783 0.783
0.877 0.877 0.877 criterion= gini max_depth= 10 10 10 splitter= random
0.783 0.783 0.783
0.853 0.853 0.853 criterion= gini max_depth= 50 50 50 splitter= random
0.779 0.779 0.779
0.873 0.873 0.873 criterion= gini max_depth= 100 100 100 splitter= random
0.777 0.777 0.777
0.875 0.875 0.875 ExtraTreeClassifier criterion= gini max_depth= 10 10 10 splitter= best
0.775 0.775 0.775
0.806 0.806 0.806 criterion= entropy max_depth= 100 100 100 splitter= best
0.775 0.775 0.775
0.847 0.847 0.847 criterion= entropy max_depth= 10 10 10 splitter= best
0.772 0.772 0.772
0.810 0.810 0.810 criterion= entropy max_depth= 50 50 50 splitter= best
0.772 0.772 0.772
0.847 0.847 0.847 criterion= gini max_depth= 100 100 100 splitter= best
0.769 0.769 0.769
0.843 0.843 0.843 criterion= gini max_depth= 50 50 50 splitter= best
0.768 0.768 0.768
0.845 0.845 0.845 criterion= entropy max_depth= 50 50 50 splitter= random
0.752 0.752 0.752
0.826 0.826 0.826 criterion= entropy max_depth= 100 100 100 splitter= random
0.752 0.752 0.752
0.828 0.828 0.828 criterion= gini max_depth= 50 50 50 splitter= random
0.748 0.748 0.748
0.824 0.824 0.824 criterion= gini max_depth= 100 100 100 splitter= random
0.745 0.745 0.745
0.820 0.820 0.820 criterion= gini max_depth= 10 10 10 splitter= random
0.739 0.739 0.739
0.737 0.737 0.737 criterion= entropy max_depth= 10 10 10 splitter= random
0.737 0.737 0.737
0.745 0.745 0.745 GaussianNB priors= [0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1]
0.511 0.511 0.511
0.524 0.524 0.524 GradientBoostingClassifier n_estimators= 100 100 100 loss= deviance max_depth= 10 10 10
0.880 0.880 0.880
0.969 0.969 0.969 n_estimators= 50 50 50 loss= deviance max_depth= 10 10 10
0.872 0.872 0.872
0.964 0.964 0.964 n_estimators= 100 100 100 loss= deviance max_depth= 3 3 3
0.862 0.862 0.862
0.949 0.949 0.949 n_estimators= 10 10 10 loss= deviance max_depth= 10 10 10
0.849 0.849 0.849
0.933 0.933 0.933 n_estimators= 50 50 50 loss= deviance max_depth= 3 3 3
0.840 0.840 0.840
0.926 0.926 0.926 n_estimators= 10 10 10 loss= deviance max_depth= 50 50 50
0.795 0.795 0.795
0.888 0.888 0.888 n_estimators= 10 10 10 loss= deviance max_depth= 3 3 3
0.782 0.782 0.782
0.846 0.846 0.846 KNeighborsClassifier weights= distance n_neighbors= 5 5 5 p= 1 1 1
0.854 0.854 0.854
0.959 0.959 0.959 weights= distance n_neighbors= 9 9 9 p= 1 1 1
0.854 0.854 0.854
0.955 0.955 0.955 weights= uniform n_neighbors= 9 9 9 p= 1 1 1
0.853 0.853 0.853
0.955 0.955 0.955 weights= uniform n_neighbors= 5 5 5 p= 1 1 1
0.852 0.852 0.852
0.957 0.957 0.957 weights= distance n_neighbors= 5 5 5 p= 2 2 2
0.852 0.852 0.852
0.945 0.945 0.945 weights= distance n_neighbors= 9 9 9 p= 2 2 2
0.849 0.849 0.849
0.944 0.944 0.944 weights= uniform n_neighbors= 5 5 5 p= 2 2 2
0.849 0.849 0.849
0.944 0.944 0.944 weights= uniform n_neighbors= 9 9 9 p= 2 2 2
0.847 0.847 0.847
0.943 0.943 0.943 weights= distance n_neighbors= 1 1 1 p= 2 2 2
0.839 0.839 0.839
0.943 0.943 0.943 weights= uniform n_neighbors= 1 1 1 p= 2 2 2
0.839 0.839 0.839
0.943 0.943 0.943 weights= uniform n_neighbors= 1 1 1 p= 1 1 1
0.838 0.838 0.838
0.955 0.955 0.955 weights= distance n_neighbors= 1 1 1 p= 1 1 1
0.838 0.838 0.838
0.955 0.955 0.955 LinearSVC loss= hinge C= 1 1 1 multi_class= ovr penalty= l2
0.836 0.836 0.836
0.917 0.917 0.917 loss= hinge C= 1 1 1 multi_class= crammer_singer penalty= l2
0.835 0.835 0.835
0.919 0.919 0.919 loss= squared_hinge C= 1 1 1 multi_class= crammer_singer penalty= l2
0.834 0.834 0.834
0.919 0.919 0.919 loss= squared_hinge C= 1 1 1 multi_class= crammer_singer penalty= l1
0.833 0.833 0.833
0.919 0.919 0.919 loss= hinge C= 1 1 1 multi_class= crammer_singer penalty= l1
0.833 0.833 0.833
0.919 0.919 0.919 loss= squared_hinge C= 1 1 1 multi_class= ovr penalty= l2
0.820 0.820 0.820
0.912 0.912 0.912 loss= squared_hinge C= 10 10 10 multi_class= ovr penalty= l2
0.779 0.779 0.779
0.885 0.885 0.885 loss= squared_hinge C= 100 100 100 multi_class= ovr penalty= l2
0.776 0.776 0.776
0.873 0.873 0.873 loss= hinge C= 10 10 10 multi_class= ovr penalty= l2
0.764 0.764 0.764
0.879 0.879 0.879 loss= hinge C= 100 100 100 multi_class= ovr penalty= l2
0.758 0.758 0.758
0.872 0.872 0.872 loss= hinge C= 10 10 10 multi_class= crammer_singer penalty= l1
0.751 0.751 0.751
0.783 0.783 0.783 loss= hinge C= 10 10 10 multi_class= crammer_singer penalty= l2
0.749 0.749 0.749
0.816 0.816 0.816 loss= squared_hinge C= 10 10 10 multi_class= crammer_singer penalty= l2
0.748 0.748 0.748
0.829 0.829 0.829 loss= squared_hinge C= 10 10 10 multi_class= crammer_singer penalty= l1
0.736 0.736 0.736
0.829 0.829 0.829 loss= hinge C= 100 100 100 multi_class= crammer_singer penalty= l1
0.516 0.516 0.516
0.759 0.759 0.759 loss= hinge C= 100 100 100 multi_class= crammer_singer penalty= l2
0.496 0.496 0.496
0.753 0.753 0.753 loss= squared_hinge C= 100 100 100 multi_class= crammer_singer penalty= l1
0.492 0.492 0.492
0.746 0.746 0.746 loss= squared_hinge C= 100 100 100 multi_class= crammer_singer penalty= l2
0.484 0.484 0.484
0.737 0.737 0.737 LogisticRegression C= 1 1 1 multi_class= ovr penalty= l1
0.842 0.842 0.842
0.917 0.917 0.917 C= 1 1 1 multi_class= ovr penalty= l2
0.841 0.841 0.841
0.917 0.917 0.917 C= 10 10 10 multi_class= ovr penalty= l2
0.839 0.839 0.839
0.916 0.916 0.916 C= 10 10 10 multi_class= ovr penalty= l1
0.839 0.839 0.839
0.909 0.909 0.909 C= 100 100 100 multi_class= ovr penalty= l2
0.836 0.836 0.836
0.916 0.916 0.916 MLPClassifier activation= relu hidden_layer_sizes= [100]
0.871 0.871 0.871
0.972 0.972 0.972 activation= relu hidden_layer_sizes= [100, 10]
0.870 0.870 0.870
0.972 0.972 0.972 activation= tanh hidden_layer_sizes= [100]
0.868 0.868 0.868
0.962 0.962 0.962 activation= tanh hidden_layer_sizes= [100, 10]
0.863 0.863 0.863
0.957 0.957 0.957 activation= relu hidden_layer_sizes= [10, 10]
0.850 0.850 0.850
0.936 0.936 0.936 activation= relu hidden_layer_sizes= [10]
0.848 0.848 0.848
0.933 0.933 0.933 activation= tanh hidden_layer_sizes= [10, 10]
0.841 0.841 0.841
0.921 0.921 0.921 activation= tanh hidden_layer_sizes= [10]
0.840 0.840 0.840
0.921 0.921 0.921 PassiveAggressiveClassifier C= 1 1 1
0.776 0.776 0.776
0.877 0.877 0.877 C= 100 100 100
0.775 0.775 0.775
0.875 0.875 0.875 C= 10 10 10
0.773 0.773 0.773
0.880 0.880 0.880 Perceptron penalty= l1
0.782 0.782 0.782
0.887 0.887 0.887 penalty= l2
0.754 0.754 0.754
0.845 0.845 0.845 penalty= elasticnet
0.726 0.726 0.726
0.845 0.845 0.845 RandomForestClassifier n_estimators= 100 100 100 criterion= entropy max_depth= 100 100 100
0.873 0.873 0.873
0.970 0.970 0.970 n_estimators= 100 100 100 criterion= gini max_depth= 100 100 100
0.872 0.872 0.872
0.970 0.970 0.970 n_estimators= 50 50 50 criterion= entropy max_depth= 100 100 100
0.872 0.872 0.872
0.968 0.968 0.968 n_estimators= 100 100 100 criterion= entropy max_depth= 50 50 50
0.872 0.872 0.872
0.969 0.969 0.969 n_estimators= 50 50 50 criterion= entropy max_depth= 50 50 50
0.871 0.871 0.871
0.967 0.967 0.967 n_estimators= 100 100 100 criterion= gini max_depth= 50 50 50
0.871 0.871 0.871
0.971 0.971 0.971 n_estimators= 50 50 50 criterion= gini max_depth= 50 50 50
0.870 0.870 0.870
0.968 0.968 0.968 n_estimators= 50 50 50 criterion= gini max_depth= 100 100 100
0.869 0.869 0.869
0.967 0.967 0.967 n_estimators= 10 10 10 criterion= entropy max_depth= 50 50 50
0.853 0.853 0.853
0.949 0.949 0.949 n_estimators= 10 10 10 criterion= entropy max_depth= 100 100 100
0.852 0.852 0.852
0.949 0.949 0.949 n_estimators= 10 10 10 criterion= gini max_depth= 50 50 50
0.848 0.848 0.848
0.948 0.948 0.948 n_estimators= 10 10 10 criterion= gini max_depth= 100 100 100
0.847 0.847 0.847
0.948 0.948 0.948 n_estimators= 50 50 50 criterion= entropy max_depth= 10 10 10
0.838 0.838 0.838
0.947 0.947 0.947 n_estimators= 100 100 100 criterion= entropy max_depth= 10 10 10
0.838 0.838 0.838
0.950 0.950 0.950 n_estimators= 100 100 100 criterion= gini max_depth= 10 10 10
0.835 0.835 0.835
0.949 0.949 0.949 n_estimators= 50 50 50 criterion= gini max_depth= 10 10 10
0.834 0.834 0.834
0.945 0.945 0.945 n_estimators= 10 10 10 criterion= entropy max_depth= 10 10 10
0.828 0.828 0.828
0.933 0.933 0.933 n_estimators= 10 10 10 criterion= gini max_depth= 10 10 10
0.825 0.825 0.825
0.930 0.930 0.930 SGDClassifier loss= hinge penalty= l2
0.819 0.819 0.819
0.914 0.914 0.914 loss= perceptron penalty= l1
0.818 0.818 0.818
0.912 0.912 0.912 loss= modified_huber penalty= l1
0.817 0.817 0.817
0.910 0.910 0.910 loss= modified_huber penalty= l2
0.816 0.816 0.816
0.913 0.913 0.913 loss= log penalty= elasticnet
0.816 0.816 0.816
0.912 0.912 0.912 loss= hinge penalty= elasticnet
0.816 0.816 0.816
0.913 0.913 0.913 loss= squared_hinge penalty= elasticnet
0.815 0.815 0.815
0.914 0.914 0.914 loss= hinge penalty= l1
0.815 0.815 0.815
0.911 0.911 0.911 loss= log penalty= l1
0.815 0.815 0.815
0.910 0.910 0.910 loss= perceptron penalty= l2
0.814 0.814 0.814
0.913 0.913 0.913 loss= perceptron penalty= elasticnet
0.814 0.814 0.814
0.912 0.912 0.912 loss= squared_hinge penalty= l2
0.814 0.814 0.814
0.912 0.912 0.912 loss= modified_huber penalty= elasticnet
0.813 0.813 0.813
0.914 0.914 0.914 loss= log penalty= l2
0.813 0.813 0.813
0.913 0.913 0.913 loss= squared_hinge penalty= l1
0.813 0.813 0.813
0.911 0.911 0.911 SVC C= 10 10 10 kernel= rbf
0.897 0.897 0.897
0.973 0.973 0.973 C= 10 10 10 kernel= poly
0.891 0.891 0.891
0.976 0.976 0.976 C= 100 100 100 kernel= poly
0.890 0.890 0.890
0.978 0.978 0.978 C= 100 100 100 kernel= rbf
0.890 0.890 0.890
0.972 0.972 0.972 C= 1 1 1 kernel= rbf
0.879 0.879 0.879
0.966 0.966 0.966 C= 1 1 1 kernel= poly
0.873 0.873 0.873
0.957 0.957 0.957 C= 1 1 1 kernel= linear
0.839 0.839 0.839
0.929 0.929 0.929 C= 10 10 10 kernel= linear
0.829 0.829 0.829
0.927 0.927 0.927 C= 100 100 100 kernel= linear
0.827 0.827 0.827
0.926 0.926 0.926 C= 1 1 1 kernel= sigmoid
0.678 0.678 0.678
0.898 0.898 0.898 C= 10 10 10 kernel= sigmoid
0.671 0.671 0.671
0.873 0.873 0.873 C= 100 100 100 kernel= sigmoid
0.664 0.664 0.664
0.868 0.868 0.868 4
Conclusions
This paper introduced Fashion-MNIST, a fashion product images dataset intended to be a drop-in replacement of MNIST and whilst providing a more challenging alternative for benchmarking machine learning algorithm. The images in Fashion-MNIST are converted to a format that matches that of the MNIST dataset, making it immediately compatible with any machine learning package capable of working with the original MNIST dataset.
References
Ciregan et al. [2012]
D. Ciregan, U. Meier, and J. Schmidhuber.
Multi-column deep neural networks for image classification.
In Computer Vision and Pattern Recognition (CVPR), 2012 IEEE Conference on , pages 3642–3649. IEEE, 2012.
Cohen et al. [2017]
G. Cohen, S. Afshar, J. Tapson, and A. van Schaik.
Emnist: an extension of mnist to handwritten letters.
arXiv preprint arXiv:1702.05373 , 2017.
Deng et al. [2009]
J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei.
Imagenet: A large-scale hierarchical image database.
In Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on , pages 248–255. IEEE, 2009.
Krizhevsky and Hinton [2009]
A. Krizhevsky and G. Hinton.
Learning multiple layers of features from tiny images.
LeCun et al. [1998]
Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner.
Gradient-based learning applied to document recognition.
Proceedings of the IEEE , 86(11):2278–2324, 1998.
Wan et al. [2013]
L. Wan, M. Zeiler, S. Zhang, Y. L. Cun, and R. Fergus.
Regularization of neural networks using dropconnect.
In Proceedings of the 30th international conference on machine learning (ICML-13) , pages 1058–1066, 2013.
◄
Feeling lucky?
Conversion report
Report an issue
View original on arXiv ►AI Summary: Based on semantic_scholar metadata. Not a recommendation.
🛡️ Paper Transparency Report
Technical metadata sourced from upstream repositories.
🆔 Identity & Source
- id
- arxiv-paper--unknown--1708.07747
- slug
- unknown--1708.07747
- source
- semantic_scholar
- author
- Han Xiao, Kashif Rasul, Roland Vollgraf
- license
- ArXiv
- tags
- paper, research, academic
⚙️ Technical Specs
- architecture
- null
- params billions
- null
- context length
- null
- pipeline tag
📊 Engagement & Metrics
- downloads
- 0
- stars
- 0
- forks
- 0
Data indexed from public sources. Updated daily.