Naveen Garg

Aryan Mobiny, Hien V. Nguyen, Supratik Moulik et al.

Deep neural networks (DNNs) have achieved state-of-the-art performances in many important domains, including medical diagnosis, security, and autonomous driving. In these domains where safety is highly critical, an erroneous decision can result in serious consequences. While a perfect prediction accuracy is not always achievable, recent work on Bayesian deep networks shows that it is possible to know when DNNs are more likely to make mistakes. Knowing what DNNs do not know is desirable to increase the safety of deep learning technology in sensitive applications. Bayesian neural networks attempt to address this challenge. However, traditional approaches are computationally intractable and do not scale well to large, complex neural network architectures. In this paper, we develop a theoretical framework to approximate Bayesian inference for DNNs by imposing a Bernoulli distribution on the model weights. This method, called MC-DropConnect, gives us a tool to represent the model uncertainty with little change in the overall model structure or computational cost. We extensively validate the proposed algorithm on multiple network architectures and datasets for classification and semantic segmentation tasks. We also propose new metrics to quantify the uncertainty estimates. This enables an objective comparison between MC-DropConnect and prior approaches. Our empirical results demonstrate that the proposed framework yields significant improvement in both prediction accuracy and uncertainty estimation quality compared to the state of the art.

Tejaswini Ganapathi, David Vining, Roland Bassett et al.

The purpose of this research is to evaluate the human and technical factors required to create a human-computer interface (HCI) for a structured reporting solution based on eye-gaze and speech signals. Gaze and speech signals from radiologists acquired during simulated image interpretation and dictation sessions were analyzed to determine a) variation of temporal relationship between eye gaze and speech in a dictation environment, and b) variation in eye movements for a particular image interpretation task among radiologists. Knowledge of these factors provides information regarding the complexity of the image interpretation or dictation task, and provides information that can be used to design a HCI for use in diagnostic radiology. Our ultimate goal is to use these data to create an HCI to automate the generation of a particular type of structured radiology report. Our data indicate that the a) temporal relationships between eye gaze and speech and b) scan paths substantially vary among radiologists, thus implying that an HCI system based on eye gaze and speech for automating the capture of data for structured reporting processes should be customized for each user. The image resolution and layout, image content, and order of targets during an image interpretation session are not relevant factors to consider when designing an HCI. Our findings can be applied to the design of other HCI solutions for radiological applications that involve visual inspection and verbal descriptions of image findings.