• 2014-2023close
• Open Access close
• Preprint close
• AE close

Two fundamental algorithm-design paradigms are Tree Search and Dynamic Programming. The techniques used therein have been shown to complement one another when solving the complete set partitioning problem, also known as the coalition structure generation problem [5]. Inspired by this observation, we develop in this paper an algorithm to solve the coalition structure generation problem on graphs, where the goal is to identifying an optimal partition of a graph into connected subgraphs. More specifically, we develop a new depth-first search algorithm, and combine it with an existing dynamic programming algorithm due to Vinyals et al. [9]. The resulting hybrid algorithm is empirically shown to significantly outperform both its constituent parts when the subset-evaluation function happens to have certain intuitive properties.

Images captured by fisheye lenses violate the pinhole camera assumption and suffer from distortions. Rectification of fisheye images is therefore a crucial preprocessing step for many computer vision applications. In this paper, we propose an end-to-end multi-context collaborative deep network for removing distortions from single fisheye images. In contrast to conventional approaches, which focus on extracting hand-crafted features from input images, our method learns high-level semantics and low-level appearance features simultaneously to estimate the distortion parameters. To facilitate training, we construct a synthesized dataset that covers various scenes and distortion parameter settings. Experiments on both synthesized and real-world datasets show that the proposed model significantly outperforms current state of the art methods. Our code and synthesized dataset will be made publicly available. Comment: 16 pages, 5 figures

We present coordinated multiwavelength observations of the high Galactic latitude (b=+50 deg) black hole X-ray binary (XRB) J1357.2-0933 in quiescence. Our broadband spectrum includes strictly simultaneous radio and X-ray observations, and near-infrared, optical, and ultraviolet data taken 1-2 days later. We detect Swift J1357.2-0933 at all wavebands except for the radio (f_5GHz < 3.9 uJy/beam). Given current constraints on the distance (2.3-6.3 kpc), its 0.5-10 keV X-ray flux corresponds to an Eddington ratio Lx/Ledd = 4e-9 -- 3e-8 (assuming a black hole mass of 10 Msun). The broadband spectrum is dominated by synchrotron radiation from a relativistic population of outflowing thermal electrons, which we argue to be a common signature of short-period quiescent BHXBs. Furthermore, we identify the frequency where the synchrotron radiation transitions from optically thick-to-thin (approximately 2-5e14 Hz, which is the most robust determination of a 'jet break' for a quiescent BHXB to date. Our interpretation relies on the presence of steep curvature in the ultraviolet spectrum, a frequency window made observable by the low amount of interstellar absorption along the line of sight. High Galactic latitude systems like Swift J1357.2-0933 with clean ultraviolet sightlines are crucial for understanding black hole accretion at low luminosities. 12 pages, 5 Figures, 1 Table. Accepted for publication in MNRAS

Some studies have demonstrated that in vivo degradation processes are influenced by the material&rsquo;s physico-chemical properties. The present study compares two hydroxyapatites manufactured on an industrial scale, deproteinized at low and high temperatures, and how physico-chemical properties can influence the mineral degradation process of material performance in bone biopsies retrieved 6 months after maxillary sinus augmentation. Residual biomaterial particles were examined by field scanning electron microscopy (SEM) and energy dispersive X-ray spectroscopy (EDX) to determine the composition and degree of degradation of the bone graft substitute material. According to the EDX analysis, the Ca/P ratio significantly lowered in the residual biomaterial (1.08&plusmn;0.32) compared to the initial composition (2.22&plusmn;0.08) for the low-temperature sintered group, which also presented high porosity, low crystallinity, low density, a large surface area, and poor stability and a high resorption rate compared to the high-temperature sintered material. This demonstrates that variations in the physico-chemical properties of bone substitute material clearly influence the degradation process. Further studies are needed to determine whether the resorption of deproteinized bone particles proceeds slowly enough to allow sufficient time for bone maturation to occur.

Person re-identification has achieved great progress with deep convolutional neural networks. However, most previous methods focus on learning individual appearance feature embedding, and it is hard for the models to handle difficult situations with different illumination, large pose variance and occlusion. In this work, we take a step further and consider employing context information for person search. For a probe-gallery pair, we first propose a contextual instance expansion module, which employs a relative attention module to search and filter useful context information in the scene. We also build a graph learning framework to effectively employ context pairs to update target similarity. These two modules are built on top of a joint detection and instance feature learning framework, which improves the discriminativeness of the learned features. The proposed framework achieves state-of-the-art performance on two widely used person search datasets. To appear in CVPR 2019

We propose a data-driven approach for deep convolutional neural network compression that achieves high accuracy with high throughput and low memory requirements. Current network compression methods either find a low-rank factorization of the features that requires more memory, or select only a subset of features by pruning entire filter channels. We propose the Cascaded Projection (CaP) compression method that projects the output and input filter channels of successive layers to a unified low dimensional space based on a low-rank projection. We optimize the projection to minimize classification loss and the difference between the next layer's features in the compressed and uncompressed networks. To solve this non-convex optimization problem we propose a new optimization method of a proxy matrix using backpropagation and Stochastic Gradient Descent (SGD) with geometric constraints. Our cascaded projection approach leads to improvements in all critical areas of network compression: high accuracy, low memory consumption, low parameter count and high processing speed. The proposed CaP method demonstrates state-of-the-art results compressing VGG16 and ResNet networks with over 4x reduction in the number of computations and excellent performance in top-5 accuracy on the ImageNet dataset before and after fine-tuning.

AbstractOnly a few studies have assessed the long-term duration of the humoral immune response against severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2).In this nationwide longitudinal study from the Faroe Islands with close to full participation of all individuals on the Islands with PCR confirmed COVID-19 during the two waves of infections in the spring and autumn 2020 (n=172 & n=233), samples were drawn at three longitudinal time points (3, 7 and 12 months and 1, 3 and 7 months after disease onset, respectively).Serum was analyzed with a direct quantitative IgG antibody binding ELISA to detect anti–SARS-CoV-2 spike RBD antibodies and a commercially available qualitative sandwich RBD ELISA kit measuring total antibody binding.The seropositive rate in the convalescent individuals was above 95 % at all sampling time points for both assays. There was an overall decline in IgG titers over time in both waves (p < 0.001). Pairwise comparison showed that IgG declined significantly from the first sample until approximately 7 months in both waves (p < 0.001). After that, the antibody level still declined significantly (p < 0.001), but decelerated with an altered slope remaining fairly stable from 7 months to 12 months after infection. Interestingly, the IgG titers followed a U-shaped curve with higher antibody levels among the oldest (67+) and the youngest (0– 17) age groups compared to intermediate groups (p < 0.001).Our results indicate that COVID-19 convalescent individuals are likely to be protected from reinfection up to 12 months after symptom onset and maybe even longer. We believe our results can add to the understanding of natural immunity and the expected durability of SARS-CoV-2 vaccine immune responses.

Enzymatic degradation of organic pollutants is a new and promising remediation approach. Peroxidases are one of the most commonly used classes of enzymes to degrade organic pollutants. However, it is generally assumed that all peroxidases behave similarly and produce similar degradation products. In this study, we conducted detailed studies of the degradation of a model aromatic pollutant, Sulforhodamine B dye (SRB dye), using two peroxidases&mdash;soybean peroxidase (SBP) and chloroperoxidase (CPO). Our results show that these two related enzymes had different optimum conditions (pH, temperature, H2O2 concentration...etc.) for efficiently degrading SRB dye. High-performance liquid chromatography and LC-mass spectrometry analyses confirmed that both SBP and CPO transformed the SRB dye into low molecular weight intermediates. While most of the intermediates produced by the two enzymes were the same, the CPO treatment produced at least one different intermediate. Furthermore, toxicological evaluation using lettuce (Lactuca sativa) seeds demonstrated that the SBP-based treatment was able to eliminate the phytotoxicity of SRB dye, but the CPO-based treatment did not. Our results show, for the first time, that while both of these related enzymes can be used to efficiently degrade organic pollutants, they have different optimum reaction conditions and may not be equally efficient in detoxification of organic pollutants.