English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Safe deployment of deep learning systems in critical real world applicationsrequires models to make very few mistakes, and only under predictablecircumstances. In this work, we address this problem using an abstainingclassifier that is tuned to have $>$95% accuracy, and then identify thedeterminants of abstention using LIME. Essentially, we are training our modelto learn the attributes of pathology reports that are likely to lead toincorrect classifications, albeit at the cost of reduced sensitivity. Wedemonstrate an abstaining classifier in a multitask setting for classifyingcancer pathology reports from the NCI SEER cancer registries on six tasks ofinterest. For these tasks, we reduce the classification error rate by factorsof 2--5 by abstaining on 25--45% of the reports. For the specific task ofclassifying cancer site, we are able to identify metastasis, reports involvinglymph nodes, and discussion of multiple cancer sites as responsible for many ofthe classification mistakes, and observe that the extent and types of mistakesvary systematically with cancer site (e.g., breast, lung, and prostate). Whencombining across three of the tasks, our model classifies 50% of the reportswith an accuracy greater than 95% for three of the six tasks\edit, and greaterthan 85% for all six tasks on the retained samples. Furthermore, we show thatLIME provides a better determinant of classification than measures of wordoccurrence alone. By combining a deep abstaining classifier with featureidentification using LIME, we are able to identify concepts responsible forboth correctness and abstention when classifying cancer sites from pathologyreports. The improvement of LIME over keyword searches is statisticallysignificant, presumably because words are assessed in context and have beenidentified as a local determinant of classification.

Why I'm not Answering: Understanding Determinants of Classification of an Abstaining Classifier for Cancer Pathology Reports