Search for: All records

Time roots in applying language models for biomedical applications: models are trained on historical data and will be deployed for new or future data, which may vary from training data. While increasing biomedical tasks have employed state-of-the-art language models, there are very few studies have examined temporal effects on biomedical models when data usually shifts across development and deployment. This study fills the gap by statistically probing relations between language model performance and data shifts across three biomedical tasks. We deploy diverse metrics to evaluate model performance, distance methods to measure data drifts, and statistical methods to quantify temporal effects on biomedical language models. Our study shows that time matters for deploying biomedical language models, while the degree of performance degradation varies by biomedical tasks and statistical quantification approaches. We believe this study can establish a solid benchmark to evaluate and assess temporal effects on deploying biomedical language models. 

Biodiversity specimen collectors are on the front lines of observing biotic anomalies, some of which herald early stages of significant changes (e.g., the arrival of a new disease; Pearson and Mast 2019). Online data sharing has opened new possibilities for the discovery of anomaly descriptions on collectors’ labels, but it remains a challenge to find these needles in the haystack of many millions of specimen records available at aggregators like iDigBio and Global Biodiversity Information Facility. In a recent community survey, over 200 collectors identified 170 unique words and phrases (e.g., atypical) that they would use to describe six types of anomaly (Pearson and Mast 2019). Left unanswered was the relative efficiency with which anomaly descriptions can be found using the simple presence of these words. Here, we address that question with a focus on one type of anomaly (phenological; related to the timing of life historyevents) and ask a second question: can we further improve the efficiency of anomaly description discovery by engaging artificial intelligence (AI)? 

Monitoring and controlling the neutral and charged excitons (trions) in two-dimensional (2D) materials are essential for the development of high-performance devices. However, nanoscale control is challenging because of diffraction-limited spatial resolution of conventional far-field techniques. Here, we extend the classical tip-enhanced photoluminescence based on tip-substrate nanocavity to quantum regime and demonstrate controlled nano-optical imaging, namely, tip-enhanced quantum plasmonics. In addition to improving the spatial resolution, we use the scanning probe to control the optoelectronic response of monolayer WS 2 by varying the neutral/charged exciton ratio via charge tunneling in Au-Ag picocavity. We observe trion “hot spots” generated by varying the picometer-scale probe-sample distance and show the effects of weak and strong coupling, which depend on the spatial location. Our experimental results are in agreement with simulations and open an unprecedented view of a new range of quantum plasmonic phenomena with 2D materials that will help to design new quantum optoelectronic devices.