With many cores driven by high memory bandwidth, today’s graphics processing unit (GPU) has involved into an absolute computing workhorse. More and more scientists, researchers and software developers are using GPUs to accelerate their algorithms and applications. Developing complex programs and software on the GPU, however, is still far from easy with existing tools provided by hardware vendors. This article introduces our recent research efforts to make GPU software development much easier. Speci?cally, we designed BSGP, a high-level programming language for generalpurpose computation on the GPU. A BSGP program looks much the same as a sequential C program, and is thus easy to read, write and maintain. Its performance on the GPU is guaranteed by a well-designed compiler that converts the program to native GPU code. We also developed an effective debugging system for BSGP programs based on the GPU interrupt, a unique feature of BSGP that allows calling CPU functions from inside GPU code. Moreover, using BSGP, we developed GPU algorithms for constructing several widely-used spatial hierarchies for high-performance graphics applications.

Evolutionary computation (EC) is one of the fastest growing areas in computer science that solves intractable optimization problems by emulating biologic evolution and organizational behave iors in nature. To design an EC algorithm, one needs to determine a set of algorithmic configurations like operator selections and parameter settings. How to design an effective and efficient adaptation scheme for adjusting the configurations of EC algorithms has become a significant and promising research topic in the EC research community. This paper intends to provide a comprehensive survey on this rapidly growing field. We present a classification of adaptive EC (AEC) algorithms from the perspective of how an adaptation scheme is designed, involving the adaptation objects, adaptation evidences, and adaptation methods. In particular, by analyzing the population distribution characteristics of EC algorithms, we discuss why and how the evolutionary state information of EC can be estimated and utilized for designing effective EC adaptation schemes. Two AEC algorithms using the idea of evolutionary state estimation, including the clustering-based adaptive genetic algorithm and the adaptive particle swarm optimization algorithm are presented in detail. Some potential directions for the research of AECs are also discussed in this paper.

The aim of this paper is to provide a systematic review on the framework to analyze dynamics in recurrently connected neural networks with discontinuous right-hand sides with a focus on the authors’ works in the past three years. The concept of the Filippov solution is employed to define the solution of the neural network systems by transforming them to differential inclusions. The theory of viability provides a tool to study the existence and uniqueness of the solution and the Lyapunov function (functional) approach is used to investigate the global stability and synchronization. More precisely, we prove that the diagonal-dominant conditions guarantee the existence, uniqueness, and stability of a general class of integro-differential equations with (almost) periodic self-inhibitions, interconnection weights, inputs, and delays. This model is rather general and includes the well-known Hopfield neural networks, Cohen-Grossberg neural networks, and cellular neural networks as special cases. We extend the absolute stability analysis of gradient-like neural network model by relaxing the analytic constraints so that they can be employed to solve optimization problem with non-smooth cost functions. Furthermore, we study the global synchronization problem of a class of linearly coupled neural network with discontinuous right-hand sides.

Automatic target recognition (ATR) is an important function for modern radar. High resolution range profile (HRRP) of target contains target structure signatures, such as target size, scatterer distribution, etc., which is a promising signature for ATR. Statistical modeling of target HRRPs is the key stage for HRRP statistical recognition, including model selection and parameter estimation. For statistical recognition algorithms, it is generally assumed that the test samples follow the same distribution model as that of the training data. Since the signal-to-noise ratio (SNR) of the received HRRP is a function of target distance, the assumption may be not met in practice. In this paper, we present a robust method for HRRP statistical recognition when SNR of test HRRP is lower than that of training samples. The noise is assumed independent Gaussian distributed, while HRRP is modeled by probabilistic principal component analysis (PPCA) model. Simulated experiments based on measured data show the effectiveness of the proposed method.

This article develops a polytopic linear parameter varying (LPV) model and presents a non-fragile H₂ gain-scheduled control for a flexible air-breathing hypersonic vehicle (FAHV). First, the polytopic LPV model of the FAHV can be obtained by using Jacobian linearization and tensor-product (TP) model transformation approach, simulation verification illustrates that the polytopic LPV model captures the local nonlinearities of the original nonlinear system. Second, based on the developed polytopic LPV model, a non-fragile gainscheduled control method is proposed in order to reduce the fragility encountered in controller implementation, a convex optimisation problem with linear matrix inequalities (LMIs) constraints is formulated for designing a velocity and altitude tracking controller, which guarantees H₂ control performance index. Finally, numerical simulations have demonstrated the effectiveness of the proposed approach.

This paper reviews our recent fMRI and psychophysical finding on: 1) perceived size representation in V1; 2) border ownership representation in V2; and 3) neural processing of partially occluded face. These findings demonstrate that the human early visual cortex not only performs local feature analyses, but also contributes significantly to high-level visual computation with assistance of attention-enabled cortical feedback. Moreover, by taking advantage of recent findings on early visual cortex from neuroscience and cognitive science, we build a biologically plausible attention model that can well predict human scanpaths on natural images.

It is attractive to formulate problems in computer vision and related fields in term of probabilistic estimation where the probability models are defined over graphs, such as grammars. The graphical structures, and the state variables defined over them, give a rich knowledge representation which can describe the complex structures of objects and images. The probability distributions defined over the graphs capture the statistical variability of these structures. These probability models can be learnt from training data with limited amounts of supervision. But learning these models suffers from the difficulty of evaluating the normalization constant, or partition function, of the probability distributions which can be extremely computationally demanding. This paper shows that by placing bounds on the normalization constant we can obtain computationally tractable approximations. Surprisingly, for certain choices of loss functions, we obtain many of the standard max-margin criteria used in support vector machines (SVMs) and hence we reduce the learning to standard machine learning methods. We show that many machine learning methods can be obtained in this way as approximations to probabilistic methods including multi-class max-margin, ordinal regression, max-margin Markov networks and parsers, multipleinstance learning, and latent SVM. We illustrate this work by computer vision applications including image labeling, object detection and localization, and motion estimation. We speculate that better results can be obtained by using better bounds and approximations.

The existing auditory computational models for evaluating speech intelligibility can only account for energetic masking, and the effect of informational masking is rarely described in these models. This study was aimed to make a computational model considering the mechanism of informational masking. Several psychoacoustic experiments were conducted to test the effect of informational masking on speech intelligibility by manipulating the number of masking talker, speech rate, and the similarity of F0 contour between target and masker. The results showed that the speech reception threshold for the target increased as the F0 contours of the masker became more similar to that of the target, suggesting that the difficulty in segregating the target harmonics from the masker harmonics may underlie the informational masking effect. Based on these studies, a new auditory computational model was made by inducing the auditory function of harmonic extraction to the traditional model of speech intelligibility index (SII), named as harmonic extraction (HF) model. The predictions of the HF model are highly consistent with the experimental results.

Over the past few decades, latent variable model (LVM)-based algorithms have attracted considerable attention for the purpose of data dimensionality reduction, which plays an important role in machine learning, pattern recognition, and computer vision. LVM is an effective tool for modeling density of the observed data. It has been used in dimensionality reduction for dealing with the sparse observed samples. In this paper, two LVM-based dimensionality reduction algorithms are presented firstly, i.e., supervised Gaussian process latent variable model and semi-supervised Gaussian process latent variable model. Then, we propose an LVMbased transfer learning model to cope with the case that samples are not independent identically distributed. In the end of each part, experimental results are given to demonstrate the validity of the proposed dimensionality reduction algorithms.

As a supplementary of [Xu L. Front. Electr. Electron. Eng. China, 2010, 5(3): 281-328], this paper outlines current status of efforts made on Bayesian Ying-Yang (BYY) harmony learning, plus gene analysis applications. At the beginning, a bird’s-eye view is provided via Gaussian mixture in comparison with typical learning algorithms and model selection criteria. Particularly, semi-supervised learning is covered simply via choosing a scalar parameter. Then, essential topics and demanding issues about BYY system design and BYY harmony learning are systematically outlined, with a modern perspective on Yin-Yang viewpoint discussed, another Yang factorization addressed, and coordinations across and within Ying-Yang summarized. The BYY system acts as a unified framework to accommodate unsupervised, supervised, and semi-supervised learning all in one formulation, while the best harmony learning provides novelty and strength to automatic model selection. Also, mathematical formulation of harmony functional has been addressed as a unified scheme for measuring the proximity to be considered in a BYY system, and used as the best choice among others. Moreover, efforts are made on a number of learning tasks, including a mode-switching factor analysis proposed as a semi-blind learning framework for several types of independent factor analysis, a hidden Markov model (HMM) gated temporal factor analysis suggested for modeling stationary temporal dependence, and a two-level hierarchical Gaussian mixture extended to cover semi-supervised learning, as well as a manifold learning modified to facilitate automatic model selection. Finally, studies are applied to the problems of gene analysis, such as genome-wide association, exome sequencing analysis, and gene transcriptional regulation.