Abstract: The present invention provides an apparatus for producing labelled cursive handwritten text samples with context style variability, and method of training and using the same. The apparatus comprises: a word embedding units configured for encoding an input text to obtain an input text word embedding; a word embedding unit configured for encoding a handwriting context style description into a context style description embedding; a transformer configured for transforming the context style description embedding to obtain a handwriting context style feature embedding; a feature embedding combiner configured for combining the handwriting context style feature embedding and the input text word embeddings to form a combined feature embedding; a generator configured for generating a synthetic image containing characters occurred in the input text with the handwriting context style defined in the handwriting context style description and context style variability introduced by the generator.

Abstract: A method for processing electronic documents, comprising: receiving an electronic document; recognizing one or more content components in the electronic document; identifying a content type for each of the recognized content components; creating, by a layer separator, one or more logical layers from the recognized content components such that each of the logical layer contains only the content components of the same content type; and invoking a content-type specific content handler for each of the logical layers created. The layer separator comprises a machine learning (ML) model based on a modified U-Net convolutional neural network and trained to classify the content types of the content components. The modified U-Net CNN is improved over traditional U-Net CNN with transformers at each layer to achieve high recovery rate.

Abstract: A method for optimizing a workflow-based neural network including an attention layer is provided. The method comprises: training the workflow-based neural network to predict a result from input elements under a prediction model with the attention layer assigning attention placements and weights, based on an original attention function, to the input elements; obtaining an original attention mask pattern and a proposed attention mask pattern; creating an attention mask updating function based on the original attention mask pattern and the proposed attention mask pattern; and combining the attention mask updating function with the original attention function to form an updated attention function.

Abstract: A method for extracting information from a table includes steps as follows. Characters of a table are extracted. The characters are merged into n-gram characters. The n-gram characters are merged into words and text lines through a two-stage GNN mode. The two-stage GNN mode comprises sub steps as: spatial features, semantic features, CNN image features are extracted from a target source; a first GNN stage is processed to output graph embedding spatial features from the spatial features; and a second GNN stage is processed to output graph embedding semantic features and graph embedding CNN image features from the semantic features and the CNN image features, respectively. The text lines are merged into cells. The cells are grouped into rows, columns, and key-value pairs based on one or more adjacency matrices, a row relationship among the cells, a column relationship among the cells, and a key-value relationship among the cells.

Abstract: The present invention provides an anomaly detection method and apparatus based on a neural network which can be trained on undamaged normal vehicle images and able to detect unknown/unseen vehicle damages of stochastic types and extents from images which are taken in various contexts. The provided method and apparatus are implemented with functional units which are trained to perform the anomaly detection under a GCAN model with a training dataset containing images of undamaged vehicles, intact-vehicle frame images and augmented vehicle frame images of the vehicles.

Abstract: The present invention provides an anomaly detection method and apparatus based on a neural network which can be trained on undamaged normal vehicle images and able to detect unknown/unseen vehicle damages of stochastic types and extents from images which are taken in various contexts. The provided method and apparatus are implemented with functional units which are trained to perform the anomaly detection under a GCAN model with a training dataset containing images of undamaged vehicles, intact-vehicle frame images and augmented vehicle frame images of the vehicles.

Abstract: A method for extracting information from a table includes steps as follows. Characters of a table are extracted. The characters are merged into n-gram characters. The n-gram characters are merged into words and text lines through a two-stage GNN mode. The two-stage GNN mode comprises sub steps as: spatial features, semantic features, CNN image features are extracted from a target source; a first GNN stage is processed to output graph embedding spatial features from the spatial features; and a second GNN stage is processed to output graph embedding semantic features and graph embedding CNN image features from the semantic features and the CNN image features, respectively. The text lines are merged into cells. The cells are grouped into rows, columns, and key based on one or more adjacency matrices, a row relationship among the cells, a column relationship among the cells, and a key-value relationship among the cells.

Abstract: A method for character recognition and semantic for natural language processing comprising extracting a sequence of feature vectors from a sequence of input character images by a convolutional neural network (CNN) feature extractor. The sequence of feature vectors comprises a plurality of feature vectors, each feature vector representing an approximate-match of its corresponding input character in the sequence of input character images. The method further comprises applying a sequential classifier sequentially as a sliding window of a size of a plurality consecutive feature vectors upon the sequence of feature vectors from a first feature vector in the sequence of feature vectors to the last feature vector in the sequence of feature vectors; and recognizing an output character for a targeted feature vector among the applied-upon consecutive feature vectors within the sliding window as it is sliding across the sequence of feature vectors.

Abstract: A method for character recognition and semantic for natural language processing comprising extracting a sequence of feature vectors from a sequence of input character images by a convolutional neural network (CNN) feature extractor. The sequence of feature vectors comprises a plurality of feature vectors, each feature vector representing an approximate-match of its corresponding input character in the sequence of input character images. The method further comprises applying a sequential classifier sequentially as a sliding window of a size of a plurality consecutive feature vectors upon the sequence of feature vectors from a first feature vector in the sequence of feature vectors to the last feature vector in the sequence of feature vectors; and recognizing an output character for a targeted feature vector among the applied-upon consecutive feature vectors within the sliding window as it is sliding across the sequence of feature vectors.