No abstract

Deep learning (DL) techniques are rapidly developed and have been widely adopted in practice. However, similar to traditional software, DL systems also contain bugs, which could cause serious impacts especially in safety-critical domains. Recently, much research has focused on testing DL models, while little attention has been paid for testing DL libraries, which is the basis of building DL models and directly affects the behavior of DL systems. In this work, we propose a novel approach, LEMON, to testing DL libraries. In particular, we (1) design a series of mutation rules for DL models, with the purpose of exploring different invoking sequences of library code and hard-to-trigger behaviors; and (2) propose a heuristic strategy to guide the model generation process towards the direction of amplifying the inconsistent degrees of the inconsistencies between different DL libraries caused by bugs, so as to mitigate the impact of potential noise introduced by uncertain factors in DL libraries. We conducted an empirical study to evaluate the effectiveness of LEMON with 20 release versions of 4 widely-used DL libraries, i.e., TensorFlow, Theano, CNTK, MXNet. The results demonstrate that LEMON detected 24 new bugs in the latest release versions of these libraries, where 7 bugs have been confirmed and one bug has been fixed by developers. Besides, the results confirm that the heuristic strategy for model generation indeed effectively guides LEMON in amplifying the inconsistent degrees for bugs.

Semiconductor quantum dots (QDs) with photoluminescence (PL) emission at 900−1700 nm (denoted as the second near-infrared window, NIR-II) exhibit much-depressed photon absorption and scattering, which has stimulated extensive researches in biomedical imaging and NIR devices. However, it is very challenging to develop NIR-II QDs with a high photoluminescence quantum yield (PLQY) and excellent biocompatibility. Herein, we designed and synthesized an alloyed silver gold selenide (AgAuSe) QD with a bright emission from 820 to 1170 nm and achieved a record absolute PLQY of 65.3% at 978 nm emission among NIR-II QDs without a toxic element and a long lifetime of 4.58 μs. It is proved that the high PLQY and long lifetime are mainly attributed to the prevented nonradiative transition of excitons, probably resulted from suppressing cation vacancies and crystal defects from the high mobility of Ag ions by alloying Au atoms. These high-PLQY QDs with nontoxic heavy metal exhibit great application potential in bioimaging, light emitting diodes (LEDs), and photovoltaic devices.

Background and Purpose— Circular RNAs (CircRNAs) show promise as stroke biomarkers because of their participation in various pathophysiological processes associated with acute ischemic stroke (AIS) and stability in peripheral blood. Methods— A circRNA microarray was used to identify differentially expressed circulating circRNAs in a discovery cohort (3 versus 3). Validation (36 versus 36) and replication (200 versus 100) were performed in independent cohorts by quantitative polymerase chain reaction. Platelets, lymphocytes, and granulocytes were separated from blood to examine the origins of circRNAs. Results— There were 3 upregulated circRNAs in Chinese population–based AIS patients compared with healthy controls. The combination of 3 circRNAs resulted in an area under the curve of 0.875, corresponding to a specificity of 91% and a sensitivity of 71.5% in AIS diagnosis. Furthermore, the combination of change rate in 3 circRNAs within the first 7 days of treatment showed an area under the curve of 0.960 in predicting stroke outcome. There was significant increase in lymphocytes and granulocytes for circPDS5B (circular RNA PDS5B) and only in granulocytes for circCDC14A (circular RNA CDC14A) in AIS patients compared with healthy controls. Conclusions— Three circRNAs could serve as biomarkers for AIS diagnosis and prediction of stroke outcomes. The elevated levels of circPDS5B and circCDC14A after stroke might be because of increased levels in lymphocytes and granulocytes.