Extreme learning machine for regression and multiclass classification

Due to the simplicity of their implementations, least square support vector machine (LS-SVM) and proximal support vector machine (PSVM) have been widely used in binary classification applications. The conventional LS-SVM and PSVM cannot be used in regression and multiclass classification applications directly, although variants of LS-SVM and PSVM have been proposed to handle such cases. This paper shows that both LS-SVM and PSVM can be simplified further and a unified learning framework of LS-SVM, PSVM, and other regularization algorithms referred to extreme learning machine (ELM) can be built. ELM works for the "generalized" single-hidden-layer feedforward networks (SLFNs), but the hidden layer (or called feature mapping) in ELM need not be tuned. Such SLFNs include but are not limited to SVM, polynomial network, and the conventional feedforward neural networks. This paper shows the following: 1) ELM provides a unified learning platform with a widespread type of feature mappings and can be applied in regression and multiclass classification applications directly; 2) from the optimization method point of view, ELM has milder optimization constraints compared to LS-SVM and PSVM; 3) in theory, compared to ELM, LS-SVM and PSVM achieve suboptimal solutions and require higher computational complexity; and 4) in theory, ELM can approximate any target continuous function and classify any disjoint regions. As verified by the simulation results, ELM tends to have better scalability and achieve similar (for regression and binary class cases) or much better (for multiclass cases) generalization performance at much faster learning speed (up to thousands times) than traditional SVM and LS-SVM.     

Routing with multiple quality-of-services constraints: An approximation perspective

Finding a path that satisfies multiple Quality-of-Service (QoS) constraints is vital to the deployment of current emerged services. However, existing algorithms are not very efficient and effective at finding such a path. Moreover, few works focus on three or more QoS constraints. In this paper, we present an enhanced version of fully polynomial time approximation scheme (EFPTAS) for multiconstrainted path optimal (MCOP) problem. Specifically, we make four major contributions. We first allow the proposed algorithm to construct an auxiliary graph, through which the QoS parameters on each of the finding path can be guaranteed not to exceed the given constraints. Then we adopt a concept, called nonlinear definition of path constraints in EFPTAS for reducing both time and space complexity. Also, we enable EFPTAS to run iteratively to facilitate a progressive refinement of the finding result. In addition to these, we identify some “deployment” issues for proposed algorithm, the essential steps that how and when the EFPTAS takes place are presented. By analyzing the proposed algorithm theoretically, we find that the presented EFPTAS can find a (1+ε)-approximation path in the network with time complexity O(|E||V|/ε) (where |E| is the number of edges and |V| is the number of nodes), which outperforms the previous best-known algorithm for MCOP. We conduct an extensive comparison between the algorithm presented in this paper and previous best-known study experimentally, our results indicate that EFPTAS can find a path with low complexity and preferable quality.     

Symbolic dynamic filtering and language measure for behavior identification of mobile robots

This paper presents a procedure for behavior identification of mobile robots, which requires limited or no domain knowledge of the underlying process. While the features of robot behavior are extracted by symbolic dynamic filtering of the observed time series, the behavior patterns are classified based on language measure theory. The behavior identification procedure has been experimentally validated on a networked robotic test bed by comparison with commonly used tools, namely, principal component analysis for feature extraction and Bayesian risk analysis for pattern classification.     

Register allocation for write activity minimization on non-volatile main memory for embedded systems

Non-volatile memories are good candidates for DRAM replacement as main memory in embedded systems and they have many desirable characteristics. Nevertheless, the disadvantages of non-volatile memory co-exist with its advantages. First, the lifetime of some of the non-volatile memories is limited by the number of erase operations. Second, read and write operations have asymmetric speed or power consumption in non-volatile memory. This paper focuses on the embedded systems using non-volatile memory as main memory. We propose register allocation technique with re-computation to reduce the number of store instructions. When non-volatile memory is applied as the main memory, reducing store instructions will reduce write activities on non-volatile memory. To re-compute the spills effectively during register allocation, a novel potential spill selection strategy is proposed. During this process, live range splitting is utilized to split certain long live ranges such that they are more likely to be assigned into registers. In addition, techniques for re-computation overhead reduction is proposed on systems with multiple functional units. With the proposed approach, the lifetime of non-volatile memory is extended accordingly. The experimental results demonstrate that the proposed technique can efficiently reduce the number of store instructions on systems with non-volatile memory by 33% on average.     

A fast two-step block type decision algorithm for intra prediction in H.264/AVC high profile

Intra coding in H.264/AVC can significantly improve the compression efficiency but at the cost of high computational complexity. To reduce the complexity, this paper presents an efficient block type decision algorithm for intra prediction. In H.264/AVC high profile, three kinds of block types are supported. This algorithm determines the optimal block type by two steps. The first step is based on the fact that the block type of intra prediction is highly dependent on the smoothness of macroblock. An edge-based feature is introduced to characterize the smoothness, by comparing it with two thresholds impossible block type is firstly filtered out. Then the second step is based on the correlation of block type chosen for different chroma modes. Experimental results show that the proposed fast algorithm can achieve 72.8% time saving on average for encoding the all intra-frame sequence with average 0.75% bit rate increase and 0.05dB PSNR degradation when comparing with the reference software.     

The interleaved video frame distribution for P2P-based VoD system with VCR functionality

The peer-to-peer (P2P) network structure is widely employed for video streaming applications because of its high stability, flexible extensibility, and ability to distribute data stream loading among different peer nodes. Numerous P2P schemes have been proposed for video on demand (VoD) applications. High video source searching cost and long response latency are always issues in dealing with VCR functionality, such as jump and fast-forward/rewind, because of asynchronous interactive and random join/leave behaviors of end users. To overcome this bottleneck, an interleaved video frame distribution (IVFD) scheme is proposed to support full VCR functionality in a P2P environment without searching for new sources. Instead of acquiring video stream data from a single parent peer in the published schemes, each child peer in the IVFD scheme can simultaneously acquire interleaved video data from multiple parent peers. When a client peer carries out arbitrary VCR operations, such as jump or fast-forward/rewind, its parent peers are still able to provide intermittent video stream data for the client peer; thus, no video source search is necessary. Simulation results reveal excellent load distribution performance and response latency for VCR operations in the proposed IVFD scheme.     

Memoryless SMC Design Methods for Stochastic Delay Systems

Recently, the sliding mode control (SMC) method has been extended to accommodate stochastic systems. Particularly, restrictive assumption on the structure of the control systems have been removed and a practical SMC design method for stochastic systems has been presented. This paper continues this line of research and studies the SMC methods for stochastic systems with state delay in diffusion. Both delay‐independent and delay‐dependent approaches are investigated. Essentially, this paper studies SMC design methods with memoryless controllers for time delay systems, which has been a problem in the past decades.     

Exponential stabilization of uncertain time‐delay linear systems with Markovian jumping parameters

This paper studies the exponential stabilization problem of uncertain time‐delay linear systems with Markovian jumping parameters. A novel delay decomposition approach is developed to derive delay‐dependent conditions under which the closed‐loop control system is mean square exponentially stable for all admissible uncertainties. It is shown that the feedback gain matrices and the decay rate can be obtained by solving coupled linear matrix inequalities. Moreover, the difficulties arising from searching for tuning parameters in the existing methods are overcome. Copyright © 2010 John Wiley and Sons Asia Pte Ltd and Chinese Automatic Control Society