Weixin Yang

Early detection of sepsis is key to ensure timely clinical intervention. Since very few end-to-end pipelines are publicly available, fair comparisons between methodologies are difficult if not impossible. Progress is further limited by discrepancies in the reconstruction of sepsis onset time. This retrospective cohort study highlights the variation in performance of predictive models under three subtly different interpretations of sepsis onset from the sepsis-III definition and compares this against inter-model differences. The models are chosen to cover tree-based, deep learning, and survival analysis methods. Using the MIMIC-III database, between 867 and 2178 intensive care unit admissions with sepsis were identified, depending on the onset definition. We show that model performance can be more sensitive to differences in the definition of sepsis onset than to the model itself. Given a fixed sepsis definition, the best performing method had a gain of 1–5% in the area under the receiver operating characteristic (AUROC). However, the choice of onset time can cause a greater effect, with variation of 0–6% in AUROC. We illustrate that misleading conclusions can be drawn if models are compared without consideration of the sepsis definition used which emphasizes the need for a standardized definition for sepsis onset.

Inspired by the great success of recurrent neural networks (RNNs) in sequential modeling, we introduce a novel RNN system to improve the performance of online signature verification. The training objective is to directly minimize intra-class variations and to push the distances between skilled forgeries and genuine samples above a given threshold. By back-propagating the training signals, our RNN network produced discriminative features with desired metrics. Additionally, we propose a novel descriptor, called the length-normalized path signature (LNPS), and apply it to online signature verification. LNPS has interesting properties, such as scale invariance and rotation invariance after linear combination, and shows promising results in online signature verification. Experiments on the publicly available SVC-2004 dataset yielded state-of-the-art performance of 2.37% equal error rate (EER).

The path signature feature (PSF) which was initially introduced in rough paths theory as a branch of stochastic analysis, has recently been successfully applied to the field of pattern recognition for extracting sufficient quantity of information contained in a finite trajectory, but with potentially high dimension. In this paper, we propose a variation of path signature representation, namely the dyadic path signature feature (D-PSF), to fully characterize the trajectory using a hierarchical structure to solve the rotation-free online handwritten character recognition (OLHCR) problem. We adopt the deep neural network (DNN) as classifier, and investigate three hanging normalization methods to improve the robustness of the DNN to rotational distortions. Extensive experiments on digits, English letters, and Chinese radicals demonstrated that the proposed D-PSF, jointly with hanging normalization and DNN, achieved very promising results for rotated OLHCR, significantly outperforming previous methods.

This paper presents an approach for the classification of handwritten and printed receipts based on a convolutional neural network (CNN). One of the main challenges related to such classification is the diversity of the background interference in the receipt images. To overcome this problem, we propose a new technique named "attention-based CNN" (ABCNN), inspired by the concept of "attention" in visual neuroscience. This approach helps us to focus on the receipt in an image without bounding box annotation. Our experimental results showed that the proposed ABCNN (i) significantly improves the classification accuracy compared to normal CNN (from 95% to 98.25%), and (ii) enables the network to process images directly without object detection, and (iii) it is faster to train and test the network.

The rapid adoption of touchscreen mobile terminals and pen-based interfaces has increased the demand for handwriting-based writer identification systems, particularly in the areas personal authentication and digital forensics. However, most writer identification systems yield poor performance because of insufficient data and an inability to handle the various conditions inherent in handwriting samples. To address these problems, the authors introduce the end-to-end DeepWriterID system that employs a deep convolutional neural network (CNN) and incorporates a new method called DropSegment to achieve data augmentation and improve the generalized applicability of CNN. Experiments show DeepWriterID achieves accuracy rates of 95.72 percent for Chinese text and 98.51 percent for English text.

Most existing online writer-identification systems require that the text content is supplied in advance and rely on separately designed features and classifiers. The identifications are based on lines of text, entire paragraphs, or entire documents; however, these materials are not always available. In this paper, we introduce a path-signature feature to an end-to-end text-independent writer-identification system with a deep convolutional neural network (DCNN). Because deep models require a considerable amount of data to achieve good performance, we propose a data-augmentation method named DropStroke to enrich personal handwriting. Experiments were conducted on online handwritten Chinese characters from the CASIA-OLHWDB1.0 dataset, which consists of 3,866 classes from 420 writers. For each writer, we only used 200 samples for training and the remaining 3,666 for testing. The results reveal that the path-signature feature is useful for writer identification, and the proposed DropStroke technique enhances the generalization and significantly improves performance.

Deep convolutional neural networks (DCNNs) have achieved great success in various computer vision and pattern recognition applications, including those for handwritten Chinese character recognition (HCCR). However, most current DCNN-based HCCR approaches treat the handwritten sample simply as an image bitmap, ignoring some vital domain-specific information that may be useful but that cannot be learnt by traditional networks. In this paper, we propose an enhancement of the DCNN approach to online HCCR by incorporating a variety of domain-specific knowledge, including deformation, non-linear normalization, imaginary strokes, path signature, and 8-directional features. Our contribution is twofold. First, these domain-specific technologies are investigated and integrated with a DCNN to form a composite network to achieve improved performance. Second, the resulting DCNNs with diversity in their domain knowledge are combined using a hybrid serial-parallel (HSP) strategy. Consequently, we achieve a promising accuracy of 97.20% and 96.87% on CASIA-OLHWDB1.0 and CASIA-OLHWDB1.1, respectively, outperforming the best results previously reported in the literature.