Recent studies in image retrieval task have shown that ensembling different models and combining multiple global descriptors lead to performance improvement. However, training different models for the ensemble is not only difficult but also inefficient with respect to time and memory. In this paper, we propose a novel framework that exploits multiple global descriptors to get an ensemble effect while it can be trained in an end-to-end manner. The proposed framework is flexible and expandable by the global descriptor, CNN backbone, loss, and dataset. Moreover, we investigate the effectiveness of combining multiple global descriptors with quantitative and qualitative analysis. Our extensive experiments show that the combined descriptor outperforms a single global descriptor, as it can utilize different types of feature properties. In the benchmark evaluation, the proposed framework achieves the state-of-theart performance on the CARS196, CUB200-2011, In-shop Clothes, and Stanford Online Products on image retrieval tasks. Our model implementations and pretrained models are publicly available 1 .

The ways in which organic solar cells (OSCs) are measured and characterized are inefficient: many substrates must be coated with expensive or otherwise precious materials to test the effects of a single variable in processing. This serial, sample-by-sample approach also takes significant amounts of time on the part of the researcher. Combinatorial approaches to research OSCs generally do not permit microstructural characterization on the actual films from which photovoltaic measurements were made, or they require specialized equipment that is not widely available. This paper describes the formation of one- and two-dimensional gradients in morphology and thickness. Gradients in morphology are formed using gradient annealing, and gradients in thickness are formed using asymmetric spin coating. Use of a liquid metal top electrode, eutectic gallium–indium (EGaIn), allows reversible contact with the organic semiconductor film. Reversibility of contact permits subsequent characterization of the specific areas of the semiconductor film from which the photovoltaic parameters are obtained. Microstructural data from UV–vis experiments extracted using the weakly interacting H-aggregate model, along with atomic force microscopy, are correlated to the photovoltaic performance. The technique is used first on the model bulk heterojunction system comprising regioregular poly­(3-hexylthiophene) (P3HT) and the soluble fullerene derivative [6,6]-phenyl C61 butyric acid methyl ester (PCBM). To demonstrate that the process can be used to optimize the thickness and annealing temperature using only small (≤10 mg) amounts of polymer, the technique was then applied to a bulk heterojunction blend comprising a difficult-to-obtain low-bandgap polymer. The combination of the use of gradients and a nondamaging top electrode allows for significant reduction in the amount of materials and time required to understand the effects of processing parameters and morphology on the performance of OSCs.

Mucosal mast cell (MMC) responses and worm recovery rates in rats infected with Echinostoma hortense were investigated from day 3 to day 56 post-infection (p.i.). Experimental infected group showed apparently higher number of MMC in each part of the small intestine than that of the control group. The number of MMC in the duodenum increased gradually after the infection and reached a peak on day 35 p.i. Thereafter, the number of MMC continued to decrease at a slow pace. The kinetics of MMC responses in the upper and lower jejunum were similar to that of the duodenum, but the number of MMC in the jejunum was lower. The worm recovery rate decreased with respect to time of which it was markedly reduced on day 49 and 56 p.i. The duration in which a high number of MMC appeared was similar to that in which a low rate in worm recovery was recorded. These results indicate that intestinal mastocytosis may play an important role in the expulsion of E. hortense.

Cloud computing services have emerged as a cost-effective alternative for cluster systems as the number of genomes and required computation power to analyze them increased in recent years. Here we introduce the Microsoft Azure platform with detailed execution steps and a cost comparison with Amazon Web Services.