Offline Handwriting Recognition with Multidimensional Recurrent Neural Networks

By Alex Graves, J. Urgen Schmidhuber. 2008

780 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Model Architectures

How researchers see this paper

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

In Computer Vision and Pattern Recognition

In Model Architectures

Citations in the corpus

Speech recognition with deep recurrent neural networks (2013, background): “The combination of Long Short-term Memory [11], an RNN architecture with an improved memory, with end-to-end training has proved especially effective, yielding state-of-the-art results in cursive handwriting recognition [12, 13].”
Pixel Recurrent Neural Networks (2016, methodology): “These layers use LSTM units in their state (Hochreiter & Schmidhuber, 1997; Graves & Schmidhuber, 2009) and adopt a convolution to compute at once all the states along one of the spatial dimensions of the data.”
Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks (2015, methodology): “We achieve state-of-the-art results on both PASCAL VOC and COCO, and find our proposed architecture is particularly effective at improving detection of small objects.”
Conditional Image Generation with PixelCNN Decoders (2016, methodology): “Two variants were proposed in the original paper: PixelRNN, where the pixel distributions are modeled with two-dimensional LSTM [7, 26], and PixelCNN, where they are modelled with convolutional networks.”
Towards Conceptual Compression (2016, methodology): “…to use latents and it-erate sequentially over the pixels, predicting a given pixel from the previous ones (from top left to bottom right) in an autoregressive fashion (Bengio & Bengio, 1999; Graves & Schmidhuber, 2009; Larochelle & Murray, 2011; Gre-gor & LeCun, 2011; van den Oord et al., 2016).”
Sequence Transduction with Recurrent Neural Networks (2012, background): “…et al., 2001), recent results have shown that they are now capable of delivering state-of-the-art results in real-world tasks such as handwriting recognition (Graves et al., 2008; Graves & Schmidhuber, 2008), text generation (Sutskever et al., 2011) and language modelling (Mikolov et al., 2010).”
Grid Long Short-Term Memory (2015, background): “These properties allow LSTM networks to process data with complex and separated interdependencies and to excel in a range of sequence learning domains such as speech recognition [14], ofﬂine hand-writing recognition [15], machine translation [35] and image-to-caption generation [36, 25].”
Beyond short snippets: Deep networks for video classification (2015, background): “For this reason, LSTMs yield state-of-the-art results in handwriting recognition [8, 10], speech recognition [9, 7], phoneme detection [5], emotion detection [25], segmentation of meetings and events [18], and evaluating programs [27].”
A Clockwork RNN (2014, background): “Stacking LSTMs into several layers (Fernandez et al., 2007; Graves & Schmidhuber, 2009) aims for hierarchical sequence processing.”
Learning Longer Memory in Recurrent Neural Networks (2014, neutral): “LSTM relies on a fairly sophisticated structure made of gates which control ﬂow of information to hidden neurons.”
Dropout Improves Recurrent Neural Networks for Handwriting Recognition (2013, methodology): “This architecture was proposed in [22], but we have adapted the ﬁlter sizes for input images at 300 dpi.”
Pose-conditioned Spatio-Temporal Attention for Human Action Recognition (2017, background): “Part-aware LSTMs [30] separate the memory cell into part-based subcells and let the network learn long-term representations individually for each part, fusing the parts for output.”
Self-Normalizing Neural Networks (2017, background): “RNNs have won handwriting recognition challenges (Chinese and Arabic handwriting) [33, 13, 6] and Kaggle challenges, such as the “Grasp-and Lift EEG” competition.”
Semantic Image Segmentation with Task-Specific Edge Detection Using CNNs and a Discriminatively Trained Domain Transform (2015, background): “Some recent work has attempted to model spatial long range dependency with recurrent networks for computer vision tasks [17, 41, 35, 4, 43].”
On Learning to Think: Algorithmic Information Theory for Novel Combinations of Reinforcement Learning Controllers and Recurrent Neural World Models (2015, background): “2) will develop high-level, abstract, spatio-temporal feature detectors that may be active for thousands of time steps, as long as those memories are useful to predict (and thus compress) future observations [62, 61, 189, 79].”
Self-Delimiting Neural Networks (2012, methodology): “Long Short-Term Memory (LSTM) overcame these problems, outperforming early RNNs in many applications [27, 14, 15, 65, 20, 21, 23, 22].”
Recurrent Convolutional Neural Networks for Scene Labeling (2014, methodology): “RNNs have been successfully applied to wide variety of tasks, including in natural language processing (Stoianov et al., 1997), speech processing (Robinson, 1994) and image processing (Graves & Schmidhuber, 2008).”
A Fixed-Size Encoding Method for Variable-Length Sequences with its Application to Neural Network Language Models (2015, background): “For example, the long short term memory (LSTM) RNN [Hochreiter, 1997] is an enhanced architecture to implement the recurrent feedbacks using various learnable gates, and it has obtained promising results on hand-writing recognition [Graves, 2009] and sequence modeling [Graves, 2013].”