Search

In this article, the cluster synchronization control problem is studied for discrete-time complex dynamical networks when the data transmission is subject to constrained bit rate. A bit-rate model is presented to quantify the limited network bandwidth, and the effects from the constrained bit rate onto the control performance of the cluster synchronization are evaluated. A sufficient condition is first proposed to guarantee the ultimate boundedness of the error dynamics of the cluster synchronization, and then, a bit-rate condition is established to reveal the fundamental relationship between the bit rate and the certain performance index of the cluster synchronization. Subsequently, two optimization problems are formulated to design the desired synchronization controllers with aim to achieve two distinct synchronization performance indices. The codesign issue for the bit-rate allocation protocol and the controller gains is further discussed to reduce the conservatism by locally minimizing a certain asymptotic upper bound of the synchronization error dynamics. Finally, three illustrative simulation examples are utilized to validate the feasibility and effectiveness of the developed synchronization control scheme.

This paper studies the distributed state estimation problem for large-scale power grid under mixed-type attacks. The topology is introduced to describe the relationship among distributed sensor nodes. Two separate Bernoulli sequences are used to describe the situations that denial of service attack and deception attack occur randomly and independently of each other. Based on the Lyapunov method and stochastic technique, a sufficient condition is obtained which guarantees the mean-square stability and H ∞ performance of the nonlinear large-scale power grid system under the mixed-type attacks. Distributed state estimators are designed to maintain a small estimation error under mixed-type attacks. Finally, the effectiveness of the designed estimators is verified by numerical simulation.

Tracking the rules matched by the packets of a flow when they traverse multiple switches in a network is very important and useful. In this paper, we design and implement a bandwidth- efficient In-Band Network Telemetry (INT) system that can track the rules matched by the packets of a flow in real time or in the past. We have evaluated the real performance of our system on a testbed composed of several hardware P4 switches. Our experimental results show that the system works correctly and its traffic reduction scheme can reduce the rate of generated INT reports by a factor of 39, 394, 2,055, or even up to 12,500, depending on the type of network states monitored and the threshold used to filter out less important INT reports.

Emerging nonvolatile memory (eNVM) have aroused extensive attention due to their low power and high speed. Recent advances have further moved eNVM to the forefront as key enablers of nonvolatile logics (nvLogics) for IoT devices and computing-in-memory (CIM) for AI chips. In this paper, we firstly examine the circuit-device-interaction (CDI) issues to implement high-performance memory macro. Then we review examples of emerging eNVM-based nvLogics for nonvolatile processors and CIM macro for AI chips with an emphasis on the challenges required CDI.

This paper investigates the maneuvering vehicle tracking problem over an energy harvesting sensor networks. The energy harvesting process of sensor equipped with a sensor harvester is assumed to be periodic. According to the periodic property of the energy harvest and the relationship between energy and SER, α θ(k) (k) and M β(k) are proposed to describe the packet dropout rates that depend both on the power levels and the sensor scheduling communication strategy, respectively. Periodic estimators are proposed to estimate the states of the system for the purpose of target tracking. Furthermore, the finite-horizon H ∞ performance is also guaranteed through the designed estimators. Finally, an example is used to clarify the effectiveness of the target tracking.

This paper concentrates on the issue of finite-time controller design for periodic systems with random transmission delays over networked control systems. The definitions of stochastic finite-time stability and finite-time boundedness are respectively given at the beginning of this paper. Then, based on the Lyapunov function and the linear matrix inequalities (LMIs) technique, the sufficient conditions that can guarantee the system to be finite-time stable are derived via state feedback. At last, a numerical example is presented to illustrate the effectiveness of the proposed method.

Web-based image search is a specialized data search used to find images from a database of digital images, which is a new usage model in search technology. Researchers are proactively working on image search development for new user experiences and are promoting a GPGPU solution to improve performance of image searches. The key image search algorithms are Union and Sort in typical image search application, which mainly contain compare operations (integer intensive) and heavy random memory access. To compare with GPGPU, we evaluate the performance of the Knights Corner (KNC) processor, which is the first generation of the Intel Many Integrated Core architecture(MIC). KNC is capable of modern GPU architectures. Besides a variety of floating point vector instructions, there are also many integer vector instructions supported like packed compare and gather/scatter which provided another choice to optimize Union and Sort on KNC. In this paper, based on the union and sort code from alibaba, we conducted several optimizations like cache blocking, utilize gather/scatter to optimize random memory access on Intel MIC architecture, further, we design an hybrid parallel paradigm to utilize all cores on the KNC processor.

An energy-efficient nonvolatile intelligent processor (NIP) is proposed for battery-less energy harvesting system. This NIP employs RRAM-based nonvolatile logics (NVL) with self-write-termination (SWT) scheme and low-power processing-in-memory (PIM) to achieve energy-efficient computing against frequent power-off situations. An NIP test chip was fabricated in 150nm CMOS process using HfO RRAM. This NIP chip achieves 462GOPs/J energy efficiency at 20MHz clock frequency, showing 13× performance improvement over state-of-the-arts. This work presents the first nonvolatile processor capable of general as well as neural network computing in addition to the first integrated chip using RRAM-based PIM.

Emerging memory devices enable performance improvements in memory applications and make possible chip designs using beyond von Neumann architectures. This paper explores the use of emerging memory devices in applications of nonvolatile logics and neuromorphic computing, and provides a review of several silicon examples of nonvolatile logics. This paper also discusses the challenges involved in the design of circuits for nonvolatile logics and neuromorphic computing systems based on emerging memory devices.

This paper is concerned with the distributed l 2 — l ∞ state estimation for discrete time periodic systems with multiplicative noises. To reflect a more general practical situation, multiplicative noises are considered both in state and measurement. The desired distributed state estimators are designed such that the estimation error system is globally asymptotically stable with a guaranteed l 2 — l ∞ performance index. A sufficient condition, based on a periodical Lyapunov procedure, is obtained to guarantee the stability and the performance of the system. At last, a numerical example is provided to illustrate the applicability and the effectiveness of the proposed results.

Approximate nearest neighbor search is a good method for large-scale image retrieval. We put forward an effective deep learning framework to generate binary hash codes for fast image retrieval after knowing the recent benefits of convolutional neural networks (CNNs). Our concept is that we can learn binary codes by using a hidden layer to present the latent concepts dominating the class labels when the data labels are usable. CNN also can be used to learn image representations. Other supervised methods require pair-wised inputs for binary code learning. However, our method can be used to learn hash codes and image representations in a point-by-point manner so it is suitable for large-scale datasets. Experimental results show that our method is better than several most advanced hashing algorithms on the CIFAR-10 and MNIST datasets. We will further demonstrate its scalability and efficiency on a large-scale dataset with 1 million clothing images.

Approximate nearest neighbor search is a good method for large-scale image retrieval. We put forward an effective deep learning framework to generate binary hash codes for fast image retrieval after knowing the recent benefits of convolution neural networks (CNN). Our concept is that we can learn binary codes by using a hidden layer to present the latent concepts dominating the class labels when the data labels are usable. CNN also can be used to learn image representations. Other supervised methods require pair-wised inputs for binary code learning. However, our method can be used to learn hash codes and image representations in a point-by-point manner so it is suitable for large-scale datasets. Experimental results show that our method is better than several most advanced hashing algorithms on the CIFAR-10 and MNIST datasets. We will further demonstrate its scalability and efficiency on a largescale dataset with 1 million clothing images.

In allusion to similarity calculation difficulty caused by high maintenance of image data, this paper introduces sparse principal component algorithm to figure out embedded subspace after dimensionality reduction of image visual words on the basis of traditional spectral hashing image index method so that image high-dimension index results can be explained overall. This method is called sparse spectral hashing index. The experiments demonstrate the method proposed in this paper superior to LSH, RBM and spectral hashing index methods.

We proposes a new image Multi-Instance (MI) bag generating method, which models an image with a Gaussian Mixed Model (GMM). The generated GMM is treated as an MI bag, of which the color and locally stable invariant components (SIFT) are the instances. Agglomerative Information Bottleneck clustering is employed to transform the MIL problem into single-instance learning problem so that single-instance classifiers can be used for classification. Finally, ensemble learning is involved to further enhance classifiers' generalization ability. Experimental results demonstrate that the performance of the proposed framework for image recognition is superior to some common MI algorithms on average in a 5-category scene recognition task.

This work describes the algorithms used in a prototypical software system for automatic pronunciation assessment of Mandarin Chinese. The system uses Viterbi decoding to isolate each syllable and find the log probability of a given utterance based on HMM (hidden Markov models). The isolated syllables are then sent to a GMM (Gaussian mixture model) for tone recognition. Based on the log probability and the result from tone recognition, a parametric scoring function, using a neural network, is constructed to approximate the scoring results from human experts. The experimental results demonstrate the system can consistently gives scores that are close to those from human's subjective evaluation.