MEDBERT.de: A Comprehensive German BERT Model for the Medical Domain

This paper presents medBERTde, a pre-trained German BERT model specifically designed for the German medical domain. The model has been trained on a large corpus of 4.7 Million German medical documents and has been shown to achieve new state-of-the-art performance on eight different medical benchmarks covering a wide range of disciplines and medical document types. In addition to evaluating the overall performance of the model, this paper also conducts a more in-depth analysis of its capabilities. We investigate the impact of data deduplication on the model's performance, as well as the potential benefits of using more efficient tokenization methods. Our results indicate that domain-specific models such as medBERTde are particularly useful for longer texts, and that deduplication of training data does not necessarily lead to improved performance. Furthermore, we found that efficient tokenization plays only a minor role in improving model performance, and attribute most of the improved performance to the large amount of training data. To encourage further research, the pre-trained model weights and new benchmarks based on radiological data are made publicly available for use by the scientific community.

Machine-Generated Hierarchical Structure of Human Activities to Reveal How Machines Think

Deep-learning based computer vision models have proved themselves to be ground-breaking approaches to human activity recognition (HAR). However, most existing works are dedicated to improve the prediction accuracy through either creating new model architectures, increasing model complexity, or refining model parameters by training on larger datasets. Here, we propose an alternative idea, differing from existing work, to increase model accuracy and also to shape model predictions to align with human understandings through automatically creating higher-level summarizing labels for similar groups of human activities. First, we argue the importance and feasibility of constructing a hierarchical labeling system for human activity recognition. Then, we utilize the predictions of a black box HAR model to identify similarities between different activities. Finally, we tailor hierarchical clustering methods to automatically generate hierarchical trees of activities and conduct experiments. In this system, the activity labels on the same level will have a designed magnitude of accuracy and reflect a specific amount of activity details. This strategy enables a trade-off between the extent of the details in the recognized activity and the user privacy by masking some sensitive predictions; and also provides possibilities for the use of formerly prohibited invasive models in privacy-concerned scenarios. Since the hierarchy is generated from the machine's perspective, the predictions at the upper levels provide better accuracy, which is especially useful when there are too detailed labels in the training set that are rather trivial to the final prediction goal. Moreover, the analysis of the structure of these trees can reveal the biases in the prediction model and guide future data collection strategies.

Explainable Text-Driven Neural Network for Stock Prediction

It has been shown that financial news leads to the fluctuation of stock prices. However, previous work on news-driven financial market prediction focused only on predicting stock price movement without providing an explanation. In this paper, we propose a dual-layer attention-based neural network to address this issue. In the initial stage, we introduce a knowledge-based method to adaptively extract relevant financial news. Then, we use input attention to pay more attention to the more influential news and concatenate the day embeddings with the output of the news representation. Finally, we use an output attention mechanism to allocate different weights to different days in terms of their contribution to stock price movement. Thorough empirical studies based upon historical prices of several individual stocks demonstrate the superiority of our proposed method in stock price prediction compared to state-of-the-art methods.

Differential Diagnosis for Pancreatic Cysts in CT Scans Using Densely-Connected Convolutional Networks

The lethal nature of pancreatic ductal adenocarcinoma (PDAC) calls for early differential diagnosis of pancreatic cysts, which are identified in up to 16% of normal subjects, and some of which may develop into PDAC. Previous computer-aided developments have achieved certain accuracy for classification on segmented cystic lesions in CT. However, pancreatic cysts have a large variation in size and shape, and the precise segmentation of them remains rather challenging, which restricts the computer-aided interpretation of CT images acquired for differential diagnosis. We propose a computer-aided framework for early differential diagnosis of pancreatic cysts without pre-segmenting the lesions using densely-connected convolutional networks (Dense-Net). The Dense-Net learns high-level features from whole abnormal pancreas and builds mappings between medical imaging appearance to different pathological types of pancreatic cysts. To enhance the clinical applicability, we integrate saliency maps in the framework to assist the physicians to understand the decision of the deep learning method. The test on a cohort of 206 patients with 4 pathologically confirmed subtypes of pancreatic cysts has achieved an overall accuracy of 72.8%, which is significantly higher than the baseline accuracy of 48.1%, which strongly supports the clinical potential of our developed method.

Video Compressive Sensing for Spatial Multiplexing Cameras using Motion-Flow Models

Spatial multiplexing cameras (SMCs) acquire a (typically static) scene through a series of coded projections using a spatial light modulator (e.g., a digital micro-mirror device) and a few optical sensors. This approach finds use in imaging applications where full-frame sensors are either too expensive (e.g., for short-wave infrared wavelengths) or unavailable. Existing SMC systems reconstruct static scenes using techniques from compressive sensing (CS). For videos, however, existing acquisition and recovery methods deliver poor quality. In this paper, we propose the CS multi-scale video (CS-MUVI) sensing and recovery framework for high-quality video acquisition and recovery using SMCs. Our framework features novel sensing matrices that enable the efficient computation of a low-resolution video preview, while enabling high-resolution video recovery using convex optimization. To further improve the quality of the reconstructed videos, we extract optical-flow estimates from the low-resolution previews and impose them as constraints in the recovery procedure. We demonstrate the efficacy of our CS-MUVI framework for a host of synthetic and real measured SMC video data, and we show that high-quality videos can be recovered at roughly $60\times$ compression.

The STONE Transform: Multi-Resolution Image Enhancement and Real-Time Compressive Video

Compressed sensing enables the reconstruction of high-resolution signals from under-sampled data. While compressive methods simplify data acquisition, they require the solution of difficult recovery problems to make use of the resulting measurements. This article presents a new sensing framework that combines the advantages of both conventional and compressive sensing. Using the proposed \stone transform, measurements can be reconstructed instantly at Nyquist rates at any power-of-two resolution. The same data can then be "enhanced" to higher resolutions using compressive methods that leverage sparsity to "beat" the Nyquist limit. The availability of a fast direct reconstruction enables compressive measurements to be processed on small embedded devices. We demonstrate this by constructing a real-time compressive video camera.