The rapid growth of deep learning models and the increasing demand for large-scale datasets have posed unprece dented challenges for data loading and memory management. Existing frameworks (e.g., PyTorch, TensorFlow) often encounter performance bottlenecks when handling large datasets resulting in inefficiencies and excessive memory usage. To address these issues, we propose Lafa, a dynamic metadata loading mechanism optimized for efficient large-scale dataset processing. Lafa introduces the .Lafa format and an adaptive loading strategy with three modes to balance memory usage and loading performance, along with a local shuffle approach that reduces memory overhead and computational complexity while preserving data randomness. Experimental results on GPU (RTX 3090) and Ascend (910A) platforms demonstrate that Lafa significantly improves memory efficiency compared to existing frameworks. Specifically, for every 10 million samples loaded, Lafa reduces additional memory consumption by a factor of 1.33× to 31.34× across various dataset types, relative to the most memory-efficient baseline among PyTorch, TensorFlow, and MindSpore.

With the rapid advancement of AI technology, there has been a substantial surge in the need for computational resources. Particularly in deep learning, machine learning, and large-scale data analysis, the processing of extensive datasets necessitates exceptionally high levels of computational efficacy and speed. Conventional homogeneous computing platforms, predominantly reliant on Central Processing Units (CPU), have encountered challenges in meeting the escalating demands for high-performance computing. Consequently, this study advocates for heterogeneous hardware acceleration technology, strategically migrating data operations from CPU to varied hardware components (e.g. GPU, NPU) to enhance processing efficiency and computational performance during the data preprocessing phase. We conducted experiments to evaluate the impact of utilizing hardware heterogeneous acceleration technologies on data processing speed under various workloads and system hardware configurations. By adjusting parameters like batch size and CPU utilization rates, we compared the performance of frameworks that support hardware heterogeneity with popular deep learning frameworks (e.g. PyTorch and TensorFlow) across various hardware configurations and neural network models. Empirical findings demonstrate that the system framework optimized through heterogeneous hardware acceleration technology (the preprocessing speed is improved in all the given experimental environment tests) exhibits commendable universality and superiority in performance. Codes are available at https://github.com/mindspore-ai/mindspore.

Category-level 6D object pose estimation aims to predict the position and orientation of unseen object instances, which is a fundamental problem in robotic applications . Previous works mainly focused on exploiting visual cues from RGB images , while depth images received less attention. However, depth images contain rich geometric attributes about the object’s shape, which are crucial for inferring the object’s pose. This work achieves category-level 6D object pose estimation by performing sufficient geometric learning from depth images represented by point clouds. Specifically, we present a novel geometric consistency and geometric discrepancy learning framework called CD-Pose to resolve the intra-category variation, inter-category similarity, and objects with complex structures. Our network consists of a Pose-Consistent Module and a Pose-Discrepant Module. First, a simple MLP-based Pose-Consistent Module is utilized to extract geometrically consistent pose features of objects from the pre-computed object shape priors for each category. Then, the Pose-Discrepant Module, designed as a multi-scale region-guided transformer network, is dedicated to exploring each instance’s geometrically discrepant features. Next, the NOCS model of the object is reconstructed according to the integration of consistent and discrepant geometric representations . Finally, 6D object poses are obtained by solving the similarity transformation between the reconstruction and the observed point cloud. Experiments on the benchmark datasets show that our CD-Pose produces superior results to state-of-the-art competitors.

Category-level object pose estimation aims to predict the 6D pose and 3D metric size of objects from given categories. Due to significant intra-class shape variations among different instances, existing methods have mainly focused on estimating dense correspondences between observed point clouds and their canonical representations, i.e., normalized object coordinate space (NOCS). Subsequently, a similarity transformation is applied to recover the object pose and size. Despite these efforts, current approaches still cannot fully exploit the intrinsic geometric features to individual instances, thus limiting their ability to handle objects with complex structures (i.e., cameras). To overcome this issue, this paper introduces GPT-COPE, which leverages a graph-guided point transformer to explore distinctive geometric features from the observed point cloud. Specifically, our GPT-COPE employs a Graph-Guided Attention Encoder to extract multiscale geometric features in a local-to-global manner and utilizes an Iterative Non-Parametric Decoder to aggregate the multiscale geometric features from finer scales to coarser scales without learnable parameters. After obtaining the aggregated geometric features, the object NOCS coordinates and shape are regressed through the shape prior adaptation mechanism, and the object pose and size are obtained using the Umeyama algorithm. The multiscale network design enables perceiving the overall shape and structural information of the object, which is beneficial to handle objects with complex structures. Experimental results on the NOCS-REAL and NOCS-CAMERA datasets demonstrate that our GPT-COPE achieves state-of-the-art performance and significantly outperforms existing methods. Furthermore, our GPT-COPE shows superior generalization ability compared to existing methods on the large-scale in-the-wild dataset Wild6D and achieves better performance on the REDWOOD75 dataset, which involves objects with unconstrained orientations.

The pseudocode-to-code task has two main stages: code translation and search synthesis. It faces two main challenges: First, the generated candidate code pieces need to be more accurate. Second, there are problems with search efficiency and accuracy. To address the above challenges, this work proposes a novel approach: For the encoder of code translation, a new multi-scale pyramid feature extractor is proposed to obtain multi-scale local information, which is combined with the global information to improve the accuracy of code translation. For the search synthesis stage, this paper designs the intra-line attention, the inter-line attention, and the code-errMsg attention, which are adaptively integrated with the graph attention to effectively fuse global and local information. Under a budget of 100 program compilations, our final model, AGL-Code, outperforms the previous state-of-the-art models, achieving 46.1%/63.5% synthesis success rate on the TestP/TestW of the SPoC dataset, respectively.

This paper presents 6D vision transformer (6D-ViT), a transformer-based instance representation learning network suitable for highly accurate category-level object pose estimation based on RGB-D images. Specifically, a novel two-stream encoder-decoder framework is dedicated to exploring complex and powerful instance representations from RGB images, point clouds, and categorical shape priors. The whole framework consists of two main branches, named Pixelformer and Pointformer. Pixelformer contains a pyramid transformer encoder with an all-multilayer perceptron (MLP) decoder to extract pixelwise appearance representations from RGB images, while Pointformer relies on a cascaded transformer encoder and an all-MLP decoder to acquire the pointwise geometric characteristics from point clouds. Then, dense instance representations (i.e., correspondence matrix and deformation field) for NOCS model reconstruction are obtained from a multisource aggregation (MSA) network with shape prior, appearance and geometric information as inputs. Finally, the instance 6D pose is computed by solving the similarity transformation between the observed point clouds and the reconstructed NOCS representations. Extensive experiments with synthetic and real-world datasets demonstrate that the proposed framework achieves state-of-the-art performance for both datasets. Code is available at https://github.com/luzzou/6D-ViT.

Summary The emergence of high‐speed long‐distance networks has promoted the development of various new types of applications that are required to transfer bulk data efficiently. Because of the conservative additive increase and multiplicative decrease strategy, traditional TCP has a severe problem in utilizing bandwidth, which provides an opening for new classes of UDP‐based protocols. This paper proposes a high‐speed bulk data transfer protocol, Double Cubic UDP (henceforth DCUDP), which is built on top of UDP with rate control and reliability control. DCUDP improves the scalability of the protocol over high bandwidth‐delay product networks through cubic‐shaped rate control functions and employs a randomized algorithm to alleviate the impacts of continuous loss and loss synchronization. In addition, DCUDP introduces a novel high‐precision user‐space timer system to facilitate the data transfer. The experimental results demonstrate that DCUDP performs well in terms of bandwidth utilization, CPU utilization, intra‐protocol fairness and round‐trip time fairness, and exhibits favorable friendliness to standard TCP and high‐speed TCP variants. Copyright © 2016 John Wiley & Sons, Ltd.

The continuous growth of media-rich content calls for more efficient and secure methods for content delivery. In this paper, we will address the optimallinear network coding(LNC) design forsecure unicast streamingagainst passive attacks, under the requirement ofinformation theoretical security. The objectives include 1) satisfying the information theoretical security requirement, 2) maximizing the transmission rate of a unicast stream, 3) minimizing the number of additional random symbols, and 4) minimizing the total bandwidth cost of content delivery. To fulfill the first three objectives, we formulate aninformation theoretically secure unicast streaming(ITSUS) problem, and then solve it by transforming it to a maximum network flow problem with node-capacity constraints. Based on the solution of the ITSUS problem, we develop an efficient algorithm that can find the optimal transmission topology with minimum bandwidth cost in a polynomial amount of time. With the optimal transmission topology, we investigate the design of bothdeterministicLNC and random LNC. For thedeterministicLNC design, we not only prove that it achieves the four objectives but also analyze the size of required finite field. Moreover, for the random LNC design, we analyze the probability that a random LNC scheme satisfies the information theoretical security requirement. Finally, extensive simulation experiments have been conducted, and the results demonstrate the effectiveness of the proposed algorithms.

In this paper, the performance research on CPython's latest interpreter is presented, concluding that bytecode dispatching takes about 25 percent of total execution time on average. Based on this observation, a novel bytecode dispatching mechanism is proposed to reduce the time spent on this phase to a minimum. With this mechanism, the blocks associated with each kind of bytecodes are rewritten in hand-tuned assembly, their opcodes are renumbered, and their memory spaces are rescheduled. With these preparations, this new bytecode dispatching mechanism replaces the time-consuming memory reading operations with rapid operations on registers. This mechanism is implemented in CPython-3.3.0. Experiments on lots of benchmarks demonstrate its correctness and efficiency. The comparison between original CPython and optimized CPython shows that this new mechanism achieves about 8.5 percent performance improvement on average. For some particular benchmarks, the maximum improvement is up to 18 percentages.

In this paper, we will address the modeling and optimal design of linear network coding (LNC) for secure unicast with multiple streams between the same source and destination pair. The objectives include 1) satisfying the weakly secure requirements, 2) maximizing the transmission data rate, and 3) minimizing the size of the finite field. To fulfill the first two objectives, we formulate a secure unicast routing problem and prove that it is equivalent to a constrained link-disjoint path problem. Based on this fact, we develop an efficient algorithm that can find the optimal unicast topology in a polynomial amount of time. With the given topology, we investigate the design of both weakly secure deterministic LNC and weakly secure random LNC. In the designs of deterministic LNC and random LNC, we prove that the required size of the finite field decreases with the decrease of the number of intermediate nodes in the topology. Therefore, to meet the third objective, we formulate a problem to minimize the number of intermediate nodes. We prove that this problem is NP-Complete and develop an approximation algorithm to solve it. Finally, extensive simulation experiments have been conducted, and the results demonstrate the effectiveness of the proposed algorithms.

The instruction-level parallelism (ILP) of Very Long Instruction (VLIW) Word DSP processor is acquired through operation partitioning and software pipeline. In the previous research of cluster, researchers always focus on reducing move operations between clusters, but rarely consider the effect of heterogeneous architecture combined with SIMD structure and some registers which should be allocated on specified cluster. And the process of building the data flow graph (DFG) has no relationships with the architecture. Therefore, we design an algorithm based on DFG for heterogeneous multicluster VLIW DSP processor to solve the problem. Firstly, using the method of SIMD to find the instructions to be disposed specially. Then the DFG is built with the information of the cluster, and it will be partitioned into several sub-graphs according to the relations among operations. At last the sub-graphs are adjusted with a heuristic method. Experiment results show that this algorithm can make the load of cluster more balanced, and achieve an average of 10% improvement compared to the traditional method.

Abstract Content‐based human motion analysis has captured extensive concerns of researchers from the domains of computer animation, human‐machine interaction, entertainment, etc. However, it is a non‐trivial task due to the spatial and temporal variations in the motion data. In this paper, we propose a generalized model (GM)‐based approach to model the variations and accurately recognize motion patterns. We partition the human character model into five parts, and extract the features of the submotions of each specific body part using clustering techniques. These features from the training trials in each class are combined to build the GM. We propose a new penalty based similarity measure for DTW to be used with the GMs for isolated motion recognition. On the other hand, from the GMs five body partition index maps are constructed and used for matching together with a flexible end point detection scheme during continuous motion recognition. In the experiments, we examine the effectiveness and efficiency of the approach in both isolated motion and continuous motion recognition. The results show that our proposed method has good performance compared with other state‐of‐the‐art methods in recognition accuracy and processing speed.

In this paper, we study the optimal design of linear network coding (LNC) for secure unicast against passive attacks, under the requirement of information theoretical security (ITS). The objectives of our optimal LNC design include (1) satisfying the ITS requirement, (2) maximizing the transmission rate of a unicast stream, and (3) minimizing the number of additional random symbols. We first formulate the problem that maximizes the secure transmission rate under the requirement of ITS, which is then transformed to a constrained maximum network flow problem.We devise an efficient algorithm that can find the optimal transmission topology. Based on the transmission topology, we then design a deterministic LNC which satisfies the aforementioned objectives and provide a constructive upper bound of the size of the finite field. In addition, we also study the potential of random LNC and derive the low bound of the probability that a random LNC is information theoretically secure.

Flow untraceability is one critical requirement for anonymous communication with network coding, which prevents malicious attackers with wiretapping and traffic analysis abilities from relating the senders to the receivers, using linear dependency of the received packets. There have recently been proposals advocating encryptions on the Global Encoding Vectors (GEV) of network coding to thwart such attacks [1], [2]. Nevertheless, there has been no exploration of the capability of networking coding itself, to constitute more efficient and effective algorithms which guarantee anonymity. In this paper, we design a novel, simple, and effective linear network coding mechanism (ALNCode) to achieve flow untraceability in a communication network with multiple unicast flows. With solid theoretical analysis, we first show that linear network coding (LNC) can be applied to thwart traffic analysis attacks without the need of encrypting GEVs. Our key idea is to mix multiple flows at their intersection nodes by generating downstream GEVs from the common basis of upstream GEVs belonging to multiple flows, in order to hide the correlation of upstream and downstream GEVs in each flow. We then design a deterministic LNC scheme to implement our idea, by which the downstream GEVs produced are guaranteed to obfuscate their correlation with the corresponding upstream GEVs. We also give extensive theoretical analysis on the intersection probability of GEV bases and the influential factors to the effectiveness of our scheme, as well as the algorithm complexity to support its efficiency.

Abstract In this paper, we present an interactive dancing game based on motion capture technology. We address the problem of real‐time recognition of the user's live dance performance in order to determine the interactive motion to be rendered by a virtual dance partner. The real‐time recognition algorithm is based on a human body partition indexing scheme with flexible matching to determine the end of a move as well as to detect unwanted motion. We show that the system can recognize the live dance motions of users with good accuracy and render the interactive dance move of the virtual partner. Copyright © 2011 John Wiley & Sons, Ltd.

In this paper, a novel concept of segmental singular value decomposition (SegSVD) is proposed to represent a motion pattern with a hierarchical structure. The similarity measure based on the SegSVD representation is also proposed. SegSVD is capable of capturing the temporal information of the time series. It is effective in matching patterns in a time series in which the start and end points of the patterns are not known in advance. We evaluate the performance of our method on both isolated motion classification and continuous motion recognition for dance movements. Experiments show that our method outperforms existing work in terms of higher recognition accuracy.

Linear network coding is a promising technology that can maximize the throughput capacity of communication network. Despite this salient feature, there are still many challenges to be addressed, and security is clearly one of the most important challenges. In this paper, we will address the design of secure linear network coding. Specifically, we will investigate the network coding design that can both satisfy the weakly secure requirements and maximize the transmission data rate of multiple unicast streams between the same source and destination pair, which has not been addressed in the literature. In our study, we first prove that the secure unicast routing problem is equivalent to a constrained link-disjoint path problem. We then develop efficient algorithm that can find the optimal unicast topology in a polynomial amount of time. Based on the topology, we design deterministic linear network code that is weakly secure and can be constructed at the source node. And finally, we investigate the potential of random linear code for weakly secure unicast and prove the low bound of the probability that a random linear code is weakly secure.

In a service overlay network, the services provided by different service providers might span multiple Internet domains. A service provider failure may cause significant performance deterioration. Thus, it is desirable to provide fault tolerant service composition solutions such that the service composition can be switched to the backup service composition solution in case of a service provider failure. To provide 100% protection against a single service provider failure, fault tolerant service composition essentially requires to partition service providers into two disjoint sets, each of them can provide a service composition solution. We study a generalized fault tolerant service composition which aims to find two service composition solutions for each request to minimize the number of shared service providers. Subject to such a primary objective, we also aim to minimize the total service composition cost. We firstly prove that the problem is NP-Complete, and formulate the problem as an integer linear program. We then propose heuristic algorithms to efficiently solve the problem. Simulation results demonstrate the effectiveness of the proposed heuristic algorithms.

In this paper, we characterize the problem of how to synthesize a reliable service mediator that can guarantee reliable interaction of the mediated Web services. We first present formal models of reliable protocols for Web services and service mediators. Then, we introduce an algorithm for synthesizing reliable service mediators.

Web services composition process needs business rules to regulate the behavior of the partner services. However, designing these rules is time-consuming and error-prone, especially under the condition that current standards barely provide any abstract and high-level guidance. In this spirit, rule analysis and verification for services composition is urgently required to augment its reliability and usability. In this paper, we choose a variant of Description Logics, called ALCO(Q*), as the underlying logic, and provide a formal mapping to transform ECA rules, so that the semantics in the original ECA rules can be captured and are computationally traceable. To this end, we further investigate some important properties for business rules, namely, redundancy, termination and conflict, and propose several sound and complete algorithms to resolve them.

Currently, the natural expectation of contracting a set of Web services by virtue of their semantics is becoming more and more feasible and popular. Meanwhile, it is generally accepted that a formalism with a well-defined model-theoretic semantics (i.e. some sort of logics) should be considered as the underpinning of Semantic Web Services [1]. In this paper, concrete domain and action theory are incorporated into a very expressive DL, called ALCQO. Notably, this extension can significantly augment the expressive power for modeling and reasoning about dynamic aspects of services contracting. At the same time, the original nature and advantages of classical DLs, particularly the ability to describe "static" aspects of Web services, are also preserved to the extent possible.

In this paper, we incorporate concrete domain and action theory into a very expressive Description Logic (DL), called ALCQO. Notably, this extension can significantly augment the expressive power for modeling and reasoning about dynamic aspects of services contracting. Meanwhile, the original nature and advantages of classical DLs are also preserved to the extent possible. Categories and Subject Descriptors:

Web services can be described as local autonomous routines communicating with each other through message exchange. Hence a good understanding of the messages defined in each Web service is crucial to enabling automatic composition. This paper presents a semantic specification framework for analyzing the functional composability of autonomous Web services. As the locally described Web services contain both semantics and business protocols, we model the former using ontologies and the latter by finite-state machines. A layered approach is adopted to analyze the Web service composability, which shows how to check whether two Web services are composable or not. Based on the specification and analysis, a polynomial-time algorithm is devised for checking the composability of Web services efficiently.

In this paper, we study optical multistage interconnection networks (MINs). To increase the communication capability of parallel computing and reduce the communication latency of high-performance computing application, optical MINs represent a very important class of interconnecting schemes used for constructing optical interconnections for communication networks and multiprocessor systems. However, optical MINs still face their own challenges, e.g., cross-talk communication in optical interconnections which should be avoided to make them work properly. In this paper, we first analyze the properties of optical Cantor network and confliction conditions. Then, we show that any permutation is realizable under the constraint of avoiding cross-talk. Based on space domain approach, a routing scheme for realizing an all-to-all permutation by one pass in an optical Cantor network is also presented. Finally, we discuss the fault tolerance of this network.

In this paper, we present a context-based mechanism to facilitate Web service discovery, which takes user preferences and dislikes into account. A novel method to quantify and evaluate contexts is proposed, which lays down a possible foundation to address such practical problems like user-centered service discovery.

We describe and analyze the problems existing on RCR network, a family of scalable interconnection network topologies proposed by Sun and Cheung, which include disconnected graph, the definition of bisection width, and network diameter. And, necessary modifications on some RCR topological parameters including bisection width and network diameter are made.