Re-identifying the figures in portrait Thangka images is of great significance to the protection and dissemination of Thangka. Existing portrait Thangka image retrieval methods are limited in extracting features of multiple granularities, and consequently can not distinguish between Thangka figures which have different identities but subtle visual differences. To cope with this challenging task, we propose a Multi-Granularity Feature Learning (MGFL) framework consisting of part and global branches. The part branch splits feature maps along the vertical dimension with inspiration by the composition rules of portrait Thangka images. The global branch integrates inception feature pyramid, channel and spatial attention mechanisms to capture discriminative features of various granularities. To combine the multi-granularity features, we propose to use step-wise graph convolution to better exploit the relationships among different composition elements in portrait Thangka images. Extensive qualitative and quantitative experiments on a dataset with 6,208 portrait Thangka images of 145 identities collected by ourselves demonstrate the superiority of our method. We will release the dataset and code to the public for research purpose.

Data-and model-driven computer simulations are increasingly critical in many application domains. These simulations may track 10s or 100s of parameters, affected by complex inter-dependent dynamic processes. Moreover, decision makers usually need to run large simulation ensembles, containing 1000s of simulations. In this paper, we rely on a tensor-based framework to represent and analyze patterns in large simulation ensemble data sets to obtain a high-level understanding of the dynamic processes implied by a given ensemble of simulations.We, further, note that the inherent sparsity of the simulation ensembles (relative to the space of potential simulations one can run) constitutes a significant problem in discovering these underlying patterns. To address this challenge, we propose a partition-stitch sampling scheme, which divides the parameter space into subspaces to collect several lower modal ensembles, and complement this with a novel Multi-Task Tensor Decomposition (M2TD), technique which helps effectively and efficiently stitch these subensembles back. Experiments showed that, for a given budget of simulations, the proposed structured sampling scheme leads to significantly better overall accuracy relative to traditional sampling approaches, even when the user does not have a perfect information to help guide the structured partitioning process.

Data- and model-driven computer simulations are increasingly critical in many application domains. These simulations may track 100s or 1000s of inter-dependent parameters, spanning multiple layers and spatial-temporal frames, affected by complex dynamic processes operating at different resolutions. Because of the size and complexity of the data and the varying spatial and temporal scales at which the key processes operate, experts often lack the means to analyze results of large simulation ensembles, understand relevant processes, and assess the robustness of conclusions driven from the resulting simulations. Moreover, data and\nmodels dynamically evolve over time requiring continuous adaptation of simulation ensembles. The simDMS platform aims to address the key challenges underlying the creation and use of large simulation ensembles and enables (a) execution, storage, and indexing of large ensemble simulation data sets and the corresponding models; and (b) search, analysis, and exploration of ensemble simulation data sets to enable ensemble-based decision support.

Tensor decomposition is used for many web and user data analysis operations from clustering, trend detection, anomaly detection, to correlation analysis. However, many of the tensor decomposition schemes are sensitive to noisy data, an inevitable problem in the real world that can lead to false conclusions. The problem is compounded by over-fitting when the user data is sparse. Recent research has shown that it is possible to avoid over-fitting by relying on probabilistic techniques. However, these have two major deficiencies: (a) firstly, they assume that all the data and intermediary results can fit in the main memory, and (b) they treat the entire tensor uniformly, ignoring potential non-uniformities in the noise distribution. In this paper, we propose a Noise-Profile Adaptive Tensor Decomposition (nTD) method, which aims to tackle both of these challenges. In particular, nTD leverages a grid-based two-phase decomposition strategy for two complementary purposes: firstly, the grid partitioning helps ensure that the memory footprint of the decomposition is kept low; secondly (and perhaps more importantly) any a priori knowledge about the noise profiles of the grid partitions enable us to develop a sample assignment strategy (or s-strategy) that best suits the noise distribution of the given tensor. Experiments show that nTD's performance is significantly better than conventional CP decomposition techniques on noisy user data tensors.

Tensors are multi-dimensional arrays - consequently, tensor decomposition operations (CP and Tucker) are the bases for many high-dimensional data analysis tasks, from clustering, trend detection, anomaly detection, to correlation analysis in various application domains, including science and engineering1. One key problem with tensor decomposition is its computational complexity and space requirements. Especially, as the relevant data sets get denser, in-memory schemes for tensor decomposition become increasingly ineffective; therefore out-of-core (secondary-memory supported, potentially parallel) computing is necessitated. However, existing techniques do not consider the I/O and network data exchange costs that out-of-core execution of the tensor decomposition operation will incur. In this paper, we note that when this operation is implemented with the help of secondary-memory and/or multiple servers to tackle the memory limitations, we would need intelligent buffer-management and task-scheduling techniques which take into account the cost of bringing the relevant blocks into the buffer to minimize I/O in the system. In this paper, we introduce 2PCP, a two-phase, block-based CP decomposition system with intelligent buffer sensitive task scheduling and buffer management mechanisms. 2PCP aims to reduce I/O costs in the analysis of relatively dense tensors common in scientific and engineering applications. Experiment results compare with current state of art tensor decomposition algorithms and show that our algorithms can significantly reduce the amount of I/O and execution time while maintaining decomposition accuracy.

Network based recommendation systems leverage the topology of the underlying graph and the current user context to rank objects in the database. Random-walk based techniques, such as PageRank, encode the structure of the graph in the form of a transition matrix of a stochastic process from which the significances of the nodes in the graph are inferred. Personalized PageRank (PPR) techniques complement this with a seed node set which serves as the personalization context. In this paper, we note (and experimentally show) that PPR algorithms that do not differentiate among the seed nodes may not properly rank nodes in situations where the seed set is incomplete and/or noisy. To tackle this problem, we propose alternative robust personalized PageRank (RPR) strategies, which are insensitive to noise in the set of seed nodes and in which the rankings are not overly biased towards the seed nodes. In particular, we show that novel teleportation discounting and seed-set maximal PPR techniques help eliminate harmful bias of individual seed nodes and provide effective seed differentiation to lead to more accurate rankings.

Tensor decomposition operation is the basis for many data analysis tasks from clustering, trend detection, anomaly detection, to correlation analysis. One key problem with tensor decomposition, however, is its computational complexity -- especially for dense data sets, the decomposition process takes exponential time in the number of tensor modes; the process is relatively faster for sparse tensors, but decomposition is still a major bottleneck in many applications. While it is possible to reduce the decomposition time by trading performance with decomposition accuracy, a drop in accuracy may not always be acceptable. In this paper, we first recognize that in many applications, the user may have a focus of interest -- i.e., part of the data for which the user needs high accuracy -- and beyond this area focus, accuracy may not be as critical. Relying on this observation, we propose a novel Personalized Tensor Decomposition(PTD) mechanism for accounting for the user's focus: PTD takes as input one or more areas of focus and performs the decomposition in such a way that, when reconstructed, the accuracy of the tensor is boosted for these areas of focus. We discuss alternative ways PTD can be implemented. Experiments show that PTD helps boost accuracy at the foci of interest, while reducing the overall tensor decomposition time.

Existing RDF query languages and RDF stores fail to support a large class of knowledge applications which associate utilities or costs on the available knowledge statements. A recent proposal includes (a) a ranked RDF (R2DF) specification to enhance RDF triples with an application specific weights and (b) a SPA Rank QL query language specification, which provides novel primitives on top of the SPARQL language to express top-k queries using traditional query patterns as well as novel flexible path predicates. We introduce and demonstrate R2DB, a database system for querying weighted RDF graphs. R2DB relies on the AR2Q query processing engine, which leverages novel index structures to support efficient ranked path search and includes query optimization strategies based on proximity and sub-result inter-arrival times. In addition to being the first data management system for the R2DF data model, R2DB also provides an innovative features-of-interest (FoI) based method for visualizing large sets of query results (i.e., sub graphs of the data graph).

This article describes a middleware used in display cluster of real time visual simulation system. This middleware which based on TCP/IP protocol provides the communication infrastructure for visual simulation system. It provides unified process logic and interfaces for various status and frame data and the packing/unpacking functions of simulation data without concerning the details of the data. We found two classes of display synchronization problems and provided the resolution of them. The middleware provides a group of APIs which hide the details of data packing/unpacking, processing and transmission and meets the requirement of flexibility, efficiency and extensibility. This middleware has been successfully implemented to the display cluster of several tower simulation system.