We propose a neural network component, the regional aggregation layer, that makes it possible to train a pixel-level density estimator using only coarse-grained density aggregates, which re ect the number of objects in an image region. Our approach is simple to use and does not require domain-speci c assumptions about the nature of the density function. We evaluate our approach on several synthetic datasets. In addition, we use this approach to learn to estimate high-resolution population and housing density from satellite imagery. In all cases, we nd that our approach results in be er density estimates than a commonly used baseline. We also show how our housing density estimator can be used to classify buildings as residential or non-residential.
Individuals are increasingly turning to the web to seek and share healthcare information and this trend in online health information has resulted in a proliferation of user generated health centric content, especially online physician reviews. Physician rating websites can play a major role in empowering patients to make informed choices while selecting healthcare providers for advice and treatment. Given the wealth of information hidden in unstructured narratives such as online ratings, comments and clinical documents, there is a critical need for building efficient and accurate text classifiers for biomedicine corpus. In this paper, we analyze patient (dis)satisfaction using performance reviews of doctors and predict their ratings on various measures such as 'Knowledgeability', 'Staff' and 'Helpfulness'. We explore solutions for the same problem using Convolutional Neural Networks trained on various optimization and loss functions. We analyze the 35000 user reviews available at "www.ratemds.com" for more than 10000 doctors. The proposed model obtained an accuracy of 93% for positive/negative binary classification of patient reviews. Moreover, we obtained a mean absolute error of 0.525 in predicting rating on a 5-point scale, thu, significantly improving upon the state of the art's error rate of 0.71.Index Terms-Text classification, sentiment analysis, convolutional neural networks, dropout, physician review.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.