Search

Embracing the deep learning techniques for representation learning in clustering research has attracted broad attention in recent years, yielding a newly developed clustering paradigm, viz. the deep clustering (DC). Typically, the DC models capitalize on autoencoders to learn the intrinsic features which facilitate the clustering process in consequence. Nowadays, a generative model named variational autoencoder (VAE) has got wide acceptance in DC studies. Nevertheless, the plain VAE is insufficient to perceive the comprehensive latent features, leading to the deteriorative clustering performance. In this paper, a novel DC method is proposed to address this issue. Specifically, the generative adversarial network and VAE are coalesced into a new autoencoder called fusion autoencoder (FAE) for discerning more discriminative representation that benefits the downstream clustering task. Besides, the FAE is implemented with the deep residual network architecture which further enhances the representation learning ability. Finally, the latent space of the FAE is transformed to an embedding space shaped by a deep dense neural network for pulling away different clusters from each other and collapsing data points within individual clusters. Experiment conducted on several image datasets demonstrate the effectiveness of the proposed DC model against the baseline methods., Comment: Added the highlights section, corresponding author info, and the acknowledgments

Serverless computing, or Function-as-a-Service (FaaS), enables a new way of building and scaling applications by allowing users to deploy fine-grained functions while providing fully-managed resource provisioning and auto-scaling. Custom FaaS container support is gaining traction as it enables better control over OSes, versioning, and tooling for modernizing FaaS applications. However, providing rapid container provisioning introduces non-trivial challenges for FaaS providers, since container provisioning is costly, and real-world FaaS workloads exhibit highly dynamic patterns. In this paper, we design FAASNET, a highly-scalable middleware system for accelerating FaaS container provisioning. FAASNET is driven by the workload and infrastructure requirements of the FaaS platform at one of the world's largest cloud providers, Alibaba Cloud Function Compute. FAASNET enables scalable container provisioning via a lightweight, adaptive function tree (FT) structure. FAASNET uses an I/O efficient, on-demand fetching mechanism to further reduce provisioning costs at scale. We implement and integrate FAASNET in Alibaba Cloud Function Compute. Evaluation results show that FAASNET: (1) finishes provisioning 2;500 function containers on 1;000 virtual machines in 8.3 seconds, (2) scales 13.4× and 16.3× faster than Alibaba Cloud's current FaaS platform and a state-of-the-art P2P container registry (Kraken), respectively, and (3) sustains a bursty workload using 75:2% less time than an optimized baseline. © 2021 USENIX Annual Technical Conference. All rights reserved.

Serverless computing, or Function-as-a-Service (FaaS), enables a new way of building and scaling applications by allowing users to deploy fine-grained functions while providing fully-managed resource provisioning and auto-scaling. Custom FaaS container support is gaining traction as it enables better control over OSes, versioning, and tooling for modernizing FaaS applications. However, providing rapid container provisioning introduces non-trivial challenges for FaaS providers, since container provisioning is costly, and real-world FaaS workloads exhibit highly dynamic patterns. In this paper, we design FAASNET, a highly-scalable middleware system for accelerating FaaS container provisioning. FAASNET is driven by the workload and infrastructure requirements of the FaaS platform at one of the world's largest cloud providers, Alibaba Cloud Function Compute. FAASNET enables scalable container provisioning via a lightweight, adaptive function tree (FT) structure. FAASNET uses an I/O efficient, on-demand fetching mechanism to further reduce provisioning costs at scale. We implement and integrate FAASNET in Alibaba Cloud Function Compute. Evaluation results show that FAASNET: (1) finishes provisioning 2;500 function containers on 1;000 virtual machines in 8.3 seconds, (2) scales 13.4× and 16.3× faster than Alibaba Cloud's current FaaS platform and a state-of-the-art P2P container registry (Kraken), respectively, and (3) sustains a bursty workload using 75:2% less time than an optimized baseline. © 2021 USENIX Annual Technical Conference. All rights reserved.

Serverless computing, or Function-as-a-Service (FaaS), enables a new way of building and scaling applications by allowing users to deploy fine-grained functions while providing fully-managed resource provisioning and auto-scaling. Custom FaaS container support is gaining traction as it enables better control over OSes, versioning, and tooling for modernizing FaaS applications. However, providing rapid container provisioning introduces non-trivial challenges for FaaS providers, since container provisioning is costly, and real-world FaaS workloads exhibit highly dynamic patterns. In this paper, we design FAASNET, a highly-scalable middleware system for accelerating FaaS container provisioning. FAASNET is driven by the workload and infrastructure requirements of the FaaS platform at one of the world's largest cloud providers, Alibaba Cloud Function Compute. FAASNET enables scalable container provisioning via a lightweight, adaptive function tree (FT) structure. FAASNET uses an I/O efficient, on-demand fetching mechanism to further reduce provisioning costs at scale. We implement and integrate FAASNET in Alibaba Cloud Function Compute. Evaluation results show that FAASNET: (1) finishes provisioning 2;500 function containers on 1;000 virtual machines in 8.3 seconds, (2) scales 13.4× and 16.3× faster than Alibaba Cloud's current FaaS platform and a state-of-the-art P2P container registry (Kraken), respectively, and (3) sustains a bursty workload using 75:2% less time than an optimized baseline. © 2021 USENIX Annual Technical Conference. All rights reserved.

Serverless computing, or Function-as-a-Service (FaaS), enables a new way of building and scaling applications by allowing users to deploy fine-grained functions while providing fully-managed resource provisioning and auto-scaling. Custom FaaS container support is gaining traction as it enables better control over OSes, versioning, and tooling for modernizing FaaS applications. However, providing rapid container provisioning introduces non-trivial challenges for FaaS providers, since container provisioning is costly, and real-world FaaS workloads exhibit highly dynamic patterns. In this paper, we design FaaSNet, a highly-scalable middleware system for accelerating FaaS container provisioning. FaaSNet is driven by the workload and infrastructure requirements of the FaaS platform at one of the world's largest cloud providers, Alibaba Cloud Function Compute. FaaSNet enables scalable container provisioning via a lightweight, adaptive function tree (FT) structure. FaaSNet uses an I/O efficient, on-demand fetching mechanism to further reduce provisioning costs at scale. We implement and integrate FaaSNet in Alibaba Cloud Function Compute. Evaluation results show that FaaSNet: (1) finishes provisioning 2500 function containers on 1000 virtual machines in 8.3 seconds, (2) scales 13.4x and 16.3x faster than Alibaba Cloud's current FaaS platform and a state-of-the-art P2P container registry (Kraken), respectively, and (3) sustains a bursty workload using 75.2% less time than an optimized baseline., Comment: This is the preprint version of a paper published in USENIX ATC 2021

Chang-shuai Zhou,1,2,* Ming-tao Feng,1,2,* Xin Chen,1,2 Yang Gao,1,2 Lei Chen,1,2 Liang-dong Li,1,2 De-heng Li,1,2 Yi-qun Cao1,2 1Department of Neurosurgery, Fudan University Shanghai Cancer Center, Shanghai, People’s Republic of China; 2Department of Oncology, Shanghai Medical College, Fudan University, Shanghai, People’s Republic of China*These authors contributed equally to this workCorrespondence: Yi-qun Cao Email yiqun_fduscc@163.comBackground: Exonuclease 1 (EXO1) has been identified to be highly expressed in different human malignancies, but its expression and prognostic role in lung adenocarcinoma (LUAD) remain unknown.Materials and Methods: Two independent cohorts extracted from public databases and one cohort from our center were analyzed in this study. Expression levels of EXO1 in LUAD tissues and paired para-cancer tissues were detected. The prognostic value of EXO1 in LUAD patients was evaluated in the three cohorts. Enrichment analyses were performed to explore the possible underlying biological pathways. Moreover, we also explored the correlations between EXO1 and tumor-infiltrating immune cells and evaluated the impact of EXO1 knock-down on the migration of lung cancer cells.Results: In this study, we found that EXO1 was highly expressed in LUAD tissues compared with para-cancerous tissues in public databases (p < 0.01), which was consistent with our data (p < 0.01). Survival analysis indicated that high expression of EXO1 was associated with poor prognosis in LUAD (p < 0.01). Enrichment analyses indicated that biological pathways like cell cycle regulation, DNA damage and repair, immune response, neuroactive ligand-receptor interaction, may be associated with EXO1 aberrant expression. Moreover, high expression of EXO1 was correlated with decreased infiltrating B cells (p < 0.01) and CD4+ T cells (p < 0.01) levels, and low infiltrating levels of B cells (p < 0.01) and dendritic cells (DCs)

Additive manufacturing enables the fabrication of periodic ceramic lattices with controllable micro-architectures. Many studies reported their catastrophic brittle fracture behaviour. However, ceramic lattices may fail by a layer-by-layer pseudo-ductile fracture mode, by controlling micro-architectures and porosities. Moreover, their fracture behaviour can be optimised by introducing strut/wall thickness gradients. This paper investigates the fracture behaviour and the fracture mode transition of ceramic triply periodic minimal surface (TPMS) structures. Alumina TPMS structures with relative densities of 0.14-0.37 are fabricated by ceramic stereolithography. Quasi-static compression tests validate a transition density range for non-graded samples: low (<0.21) and moderate (>0.25) relative density samples show layer-by-layer pseudo-ductile and catastrophic brittle fracture modes, respectively. The pseudo-ductile failure mode increases the energy absorption performance, enabling load-bearing capacity for a compressive strain up to 50%. With appropriate thickness gradients, graded structures exhibit significant increase of energy absorption without a decrease of fracture strength compared to their non-graded counterparts. © 2019 Elsevier Ltd

Reprecipitation synthesis has been demonstrated to be a simple and convenient route to fabricate high quality perovskite quantum dots toward display applications, whereas the limited chemical yields (< 10%) and difficulty of purification limited its further application. In order to overcome this issue, we here report a modified emulsion synthesis by introducing phase transfer strategy, which achieving effective extraction of newly formed perovskite quantum dots into non-polar solvent and avoiding the degradation of perovskite quantum dots to a large extent. Based on this strategy, gram-scale CH3NH3PbBr3 quantum dots were fabricated in 10 mL (~0.02 mol/L) colloidal solution with chemical yields larger than 70%. The as fabricated CH3NH3PbBr3 quantum dots exhibit an emission peak of 453 nm and a full width at half maximum of only 14 nm. Moreover, electroluminescent devices based on blue emitting CH3NH3PbBr3 quantum dots were also explored with a maximum luminance of 32 cd/m2, showing potential applications in blue light emitting devices.

Designing metallic cellular structures with triply periodic minimal surface (TPMS) sheet cores is a novel approach for lightweight, multi-functional structural applications. Different from current honeycombs and lattices, TPMS sheet structures are composed of continuous and smooth shells, allowing for large surface areas and continuous internal channels. In this paper, we investigate the mechanical properties and energy absorption abilities of three types of TPMS sheet structures (Primitive, Diamond, and Gyroid) fabricated by selective laser melting (SLM) with 316 L stainless steel and classify their failure mechanisms and printing accuracy with the help of numerical analysis. The results reveal that the properties and deformation mechanisms strongly depend on the unit cell geometry. TPMS sheet structures are found to exhibit superior stiffness, plateau stress and energy absorption ability compared to body-centred cubic lattices, with Diamond-type sheet structures performing best. Linear and post-yielding mechanical behaviour of TPMS sheet structures as predicted by explicit finite element models is in good agreement with experimental results. The simulation results also show that Diamond and Gyroid sheet structures display relatively uniform stress distributions across all lattice cells under compression, leading to stable collapse mechanisms and desired energy absorption performance. In contrast, P-type structures display rapid diagonal shear band development followed by localized wall buckling. Lastly, an energy absorption diagram is developed to facilitate a systematic way to select optimal densities of TPMS structures for energy absorbing applications.

Human visual experience is formed through concurrent bottom-up and top-down processes, this close relationship between the two processes makes it hard to investigate their isolated effects. Visual phenomena formed by the top-down processes without direct retinal stimulation have been defined as phantom vision. The current thesis used an objective measurement, binocular rivalry, to quantify the functional effects of different forms of phantom vision induced by top-down processes. In Chapter 2, it is shown that imagined visual motion primes the stimuli of the same direction in subsequent motion rivalry perception, and this effect can be attenuated by concurrent non-directional perceptual motion. Perceived motion stimuli also bias rivalry perception, and the direction of the bias depends on the contrast of prior motion. Chapter 3 investigates the functional effects of colour experience induced by the neon colour spreading illusion. Results show that neon phantom colour has a suppressive effect on subsequent colour rivalry perception, and the suppression effect can be attenuated by concurrently presented uniform luminance and strengthened by longer presentation durations. Significant correlation and functional interaction between neon phantom colour and voluntarily generated colour imagery are reported. We further show individual differences for the interaction. Phantom colour can be induced by other process, such as the McCollough effect, in which individuals experience colour while observing horizontal or vertical achromatic gratings after adapting to orthogonal gratings in complementary colours. Using the binocular rivalry method, we show that phantom colour induced by the McCollough effect has suppressive effects on colour rivalry perception, and can be attenuated by the uniform luminance. The strength of McCollough effect induced colour experience is influenced by the contrast of grating inducers. The correlation between colour imagery and neon phantom colour is re

Reprecipitation synthesis has been demonstrated to be a simple and convenient route to fabricate high quality perovskite quantum dots toward display applications, whereas the limited chemical yields (< 10%) and difficulty of purification limited its further application. In order to overcome this issue, we here report a modified emulsion synthesis by introducing phase transfer strategy, which achieving effective extraction of newly formed perovskite quantum dots into non-polar solvent and avoiding the degradation of perovskite quantum dots to a large extent. Based on this strategy, gram-scale CH3NH3PbBr3 quantum dots were fabricated in 10 mL (~0.02 mol/L) colloidal solution with chemical yields larger than 70%. The as fabricated CH3NH3PbBr3 quantum dots exhibit an emission peak of 453 nm and a full width at half maximum of only 14 nm. Moreover, electroluminescent devices based on blue emitting CH3NH3PbBr3 quantum dots were also explored with a maximum luminance of 32 cd/m2, showing potential applications in blue light emitting devices.

Designing metallic cellular structures with triply periodic minimal surface (TPMS) sheet cores is a novel approach for lightweight, multi-functional structural applications. Different from current honeycombs and lattices, TPMS sheet structures are composed of continuous and smooth shells, allowing for large surface areas and continuous internal channels. In this paper, we investigate the mechanical properties and energy absorption abilities of three types of TPMS sheet structures (Primitive, Diamond, and Gyroid) fabricated by selective laser melting (SLM) with 316 L stainless steel and classify their failure mechanisms and printing accuracy with the help of numerical analysis. The results reveal that the properties and deformation mechanisms strongly depend on the unit cell geometry. TPMS sheet structures are found to exhibit superior stiffness, plateau stress and energy absorption ability compared to body-centred cubic lattices, with Diamond-type sheet structures performing best. Linear and post-yielding mechanical behaviour of TPMS sheet structures as predicted by explicit finite element models is in good agreement with experimental results. The simulation results also show that Diamond and Gyroid sheet structures display relatively uniform stress distributions across all lattice cells under compression, leading to stable collapse mechanisms and desired energy absorption performance. In contrast, P-type structures display rapid diagonal shear band development followed by localized wall buckling. Lastly, an energy absorption diagram is developed to facilitate a systematic way to select optimal densities of TPMS structures for energy absorbing applications.

Accurate localization for mobile nodes has been an important and fundamental problem in underwater acoustic sensor networks (UASNs). The detection information returned from a mobile node is meaningful only if its location is known. In this paper, we propose two localization algorithms based on color filtering technology called PCFL and ACFL. PCFL and ACFL aim at collaboratively accomplishing accurate localization of underwater mobile nodes with minimum energy expenditure. They both adopt the overlapping signal region of task anchors which can communicate with the mobile node directly as the current sampling area. PCFL employs the projected distances between each of the task projections and the mobile node, while ACFL adopts the direct distance between each of the task anchors and the mobile node. Also the proportion factor of distance is proposed to weight the RGB values. By comparing the nearness degrees of the RGB sequences between the samples and the mobile node, samples can be filtered out. And the normalized nearness degrees are considered as the weighted standards to calculate coordinates of the mobile nodes. The simulation results show that the proposed methods have excellent localization performance and can timely localize the mobile node. The average localization error of PCFL can decline by about 30.4% than the AFLA method., Comment: 18 pages, 11 figures, 2 tables