Deep Learning for Classical Japanese Literature

By Tarin Clanuwat, Mikel Bober-Irizar, Asanobu Kitamoto, Alex Lamb, Kazuaki Yamamoto, David R Ha. 2018

864 citations recorded by the source corpus.

Much of machine learning research focuses on producing models which perform well on benchmark tasks, in turn improving our understanding of the challenges associated with those tasks. From the perspective of ML researchers, the content of the task itself is largely irrelevant, and thus there have increasingly been calls for benchmark tasks to more heavily focus on problems which are of social or cultural relevance. In this work, we introduce Kuzushiji-MNIST, a dataset which focuses on Kuzushiji (cursive Japanese), as well as two larger, more challenging datasets, Kuzushiji-49 and Kuzushiji-Kanji. Through these datasets, we wish to engage the machine learning community into the world of classical Japanese literature. Dataset available at this https URL

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

Generalized Variational Continual Learning (2020, methodology): “For our Split -MNIST experiment, in addition to the standard 5 binary classiﬁcation tasks for Split - MNIST, we add 5 more binary classiﬁcation tasks by taking characters from the KMNIST dataset (Clanuwat et al., 2018).”
Interpolation Consistency Training for Semi-Supervised Learning (2019, background): “For instance, recognizing documents in extinct languages, where a machine learning system has access to a few labels, produced by highly-skilled scholars (Clanuwat et al., 2018).”
Regularization Shortcomings for Continual Learning (2019, methodology): “ted earlier in this paper, we experiment with the "MNIST-Fellowship" dataset proposed in [12]. This dataset is composed of three datasets (Fig. 4): MNIST [10], Fashion-MNIST [29] and KMNIST [4], each composed of 10 classes, those datasets should be learned sequentially one by one. We choos”
muNet: Evolving Pretrained Deep Neural Networks into Scalable Auto-tuning Multitask Systems (2022, background): “…[Cohen et al., 2017] Creative Commons Attribution 4.0 License emnist/letters [Cohen et al., 2017] Creative Commons Attribution 4.0 License kmnist [Clanuwat et al., 2018] Attribution-ShareAlike 4.0 International mnist [LeCun et al., 1998] Creative Commons Attribution 4.0 License omniglot [Lake…”
Implicit Regularization via Neural Feature Alignment (2020, background): “000 MNIST training examples with 1000 difficult examples of 2 types: (i) examples with random labels and (ii) examples from the dataset KMNIST (Clanuwat et al., 2018).”
Understanding Uncertainty in Bayesian Neural Networks (2019, methodology): “on MNIST and evaluate them on OOD samples: rotated MNIST digits, Japanese letters from the KMNIST dataset (Clanuwat et al., 2018) and adversarial examples generated with the Fast Gradient Sign Method (FGSM) (Goodfellow et al.”
A Comprehensive Study of ImageNet Pre-Training for Historical Document Image Analysis (2019, methodology): “d consists of producing one or more descriptive labels for a given input image. In the context of historical image document analysis this task can be, for example, formulated as character recognition [21], [22], style/script classiﬁcation [23], [24] ormanuscript dating [24], [25]. We train the netwo”
Contextualizing Multiple Tasks via Learning to Decompose (2021, result): “Similar to [3, 16, 23], which plot RGB colors over the grayscale images based on the Kuzushiji-49 dataset [9] and test whether a model will be disturbed by the colors.”
Domain2Vec: Domain Embedding for Unsupervised Domain Adaptation (2020, methodology): “The foreground images are from MNIST [54], USPS [55], EMNIST [56], KMNIST [57], QMNIST [58], FashionMNIST [59].”
A Probabilistic Framework for Modular Continual Learning (2023, methodology): “Concretely, we start with the following image classification datasets: MNIST [21], Fashion MNIST (FMNIST) [45], EMNIST [5] and Kuzushiji49 (KMNIST) [4].”
Locally adaptive activation functions with slope recovery for deep and physics-informed neural networks (2020, unknown): “MNIST [21], Fashion-MNIST [22] and KMNIST [23] are the datasets with handwritten digits, images of clothing and accessories, and Japanese letters.”
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].”
Training Neural Networks with Local Error Signals (2019, methodology): “Kuzushiji-MNIST is another drop-in replacement for MNIST containing hand-drawn japanese characters (Clanuwat et al., 2018).”