Facial Expression Synthesis Using Manifold Learning and Belief Propagation

Given a person’s neutral face, we can predict his/her unseen expression by machine learning techniques for image processing. Different from the prior expression cloning or image analogy approaches, we try to hallucinate the person’s plausible facial expression with the help of a large face expression database. In the first step, regularization network based nonlinear manifold learning is used to obtain a smooth estimation for unseen facial expression, which is better than the reconstruction results of PCA. In the second step, Markov network is adopted to learn the low-level local facial feature’s relationship between the residual neutral and the expressional face image’s patches in the training set, then belief propagation is employed to infer the expressional residual face image for that person. By integrating the two approaches, we obtain the final results. The experimental results show that the hallucinated facial expression is not only expressive but also close to the ground truth.     

Research of localization approach for the new comer in wireless sensor networks

When some sensor nodes of wireless sensor networks (WSN) can not work forever because of long-term work or failure caused by attack, a few new comers need to be put into the network. For the application of the new comer in WSN, an accurate and effective localization algorithm based on received signal strength indicator (RSSI) is proposed. Through a few necessary nodes’ participation and the collaboration between the new comer and its one-hop and two-hop neighbor nodes, the accurate localization of the new comer is achieved. Simulation results show that the localization accuracy is about 17% of sensor node’s radio frequency (RF) transmission range, when the measurement error is 10% and the standard deviation for Gauss error of original sensor nodes’ coordinate is about 20% of sensor node’s RF transmission range. Simulation results also verify nice stability and adaptability of the new comer’s location algorithm.     

A human-machine cooperative system for generating sign language animation using thermal image processing,a fuzzy algorithm,and simulated annealing

We propose a method for a sign language animation by skin region detection applied to an infrared thermal image. In a system incorporating the proposed method, a 3D CG model corresponding to a person’s characteristic posture while using sign language is generated automatically by pattern recognition of the thermal image, and then a person’s hand in the CG model is set. The hand part is made manually beforehand. If necessary, the model can be replaced manually by a more appropriate model corresponding to training key frames, and/or the same generated model can be refined manually. In our experiments, three hearing-impaired people, who were experienced in using sign language, recognized the Japanese sign language gestures of 70 words expressed as animations with 94.3% accuracy. We further improved the system by correcting the position and direction of the hand of the automatically generated model through the use of a fuzzy algorithm and simulated annealing.     

Providing advanced remote medical treatment services through pervasive environments

Advances on sensor technology, wireless environments and data mining introduce new possibilities in the healthcare sector, realizing the anytime-anywhere access to medical information. Towards this direction, integration of packet-switched networks and sensor devices can be effective in deploying assistive environments, such as home monitoring for elderly or patients. In this paper we describe a policy-based architecture that utilizes wireless sensor devices, advanced network topologies and software agents to enable remote monitoring of patients and elderly people; through the aforementioned technologies we achieve continuous monitoring of a patient’s condition and we can proceed when necessary with proper actions. We also present a software framework and network architecture that realizes the provision of remote medical services, in compliance with the imposed security and privacy requirements. A proof of concept prototype is also deployed, along with an evaluation of the overall architecture’s performance.     

Stochastic Representation and Recognition of High-Level Group Activities

This paper describes a stochastic methodology for the recognition of various types of high-level group activities. Our system maintains a probabilistic representation of a group activity, describing how individual activities of its group members must be organized temporally, spatially, and logically. In order to recognize each of the represented group activities, our system searches for a set of group members that has the maximum posterior probability of satisfying its representation. A hierarchical recognition algorithm utilizing a Markov chain Monte Carlo (MCMC)-based probability distribution sampling has been designed, detecting group activities and finding the acting groups simultaneously. The system has been tested to recognize complex activities such as ‘a group of thieves stealing an object from another group’ and ‘a group assaulting a person’. Videos downloaded from YouTube as well as videos that we have taken are tested. Experimental results show that our system recognizes a wide range of group activities more reliably and accurately, as compared to previous approaches.     

Improved Rao-Blackwellized particle filter for simultaneous robot localization and person-tracking with single mobile sensor

A probabilistic algorithm is proposed for the problem of simultaneous robot localization and people-tracking (SLAP) using single onboard sensor in situations with sensor noise and global uncertainties over the observer’s pose. By the decomposition of the joint distribution according to the Rao-Blackwell theorem, posteriors of the robot pose are sequentially estimated over time by a smoothed laser perception model and an improved resampling scheme with evolution strategies; the conditional distribution of the person’s position is estimated using unscented Kalman filter (UKF) to deal with the nonlinear dynamic of human motion. Experiments conducted in a real indoor service robot scenario validate the favorable performance of the positional accuracy as well as the improved computational efficiency.     

Helen Keller Was Never in a Chinese Room

William Rapaport, in “How Helen Keller used syntactic semantics to escape from a Chinese Room,” (Rapaport 2006), argues that Helen Keller was in a sort of Chinese Room, and that her subsequent development of natural language fluency illustrates the flaws in Searle’s famous Chinese Room Argument and provides a method for developing computers that have genuine semantics (and intentionality). I contend that his argument fails. In setting the problem, Rapaport uses his own preferred definitions of semantics and syntax, but he does not translate Searle’s Chinese Room argument into that idiom before attacking it. Once the Chinese Room is translated into Rapaport’s idiom (in a manner that preserves the distinction between meaningful representations and uninterpreted symbols), I demonstrate how Rapaport’s argument fails to defeat the CRA. This failure brings a crucial element of the Chinese Room Argument to the fore: the person in the Chinese Room is prevented from connecting the Chinese symbols to his/her own meaningful experiences and memories. This issue must be addressed before any victory over the CRA is announced.     

Constructing a SenseCam visual diary as a media process

The SenseCam is a small wearable personal device which automatically captures up to 2,500 images per day. This yields a very large personal collection of images, or in a sense a large visual diary of a person’s day. Intelligent techniques are necessary for effective structuring, searching and browsing of this image collection for locating important or significant events in a person’s life. In this paper we identify three stages in the process of capturing and structuring SenseCam images and then displaying them to an end user to review. These stages are expressed in terms of the Canonical process stages to which they correlate.     

SNR estimation algorithm based on the preamble for wireless OFDM systems

Based on the analysis of the performance of Boumard＇s SNR method for wireless orthogonal frequency division multiplexing （OFDM） systems, a new estimation algorithm of the noise variance is proposed by only using the data samples of the two training symbols in the preamble, and the second order moment of these data samples is employed to estimate the signal power. The average SNR and the SNRs on the subchannels can all be estimated by the proposed algorithm, and its performance is independent of the channel＇s frequency selectivity. Simulation results show that the performance of the proposed method is highly improved and much better than that of Boumard＇s method.     

Implementing mapping composition

Mapping composition is a fundamental operation in metadata driven applications. Given a mapping over schemas σ₁ and σ₂ and a mapping over schemas σ₂ and σ₃, the composition problem is to compute an equivalent mapping over σ₁ and σ₃. We describe a new composition algorithm that targets practical applications. It incorporates view unfolding. It eliminates as many σ₂ symbols as possible, even if not all can be eliminated. It covers constraints expressed using arbitrary monotone relational operators and, to a lesser extent, non-monotone operators. And it introduces the new technique of left composition. We describe our implementation, explain how to extend it to support user-defined operators, and present experimental results which validate its effectiveness. T.J. Green and A. Nash’s work was performed during an internship at Microsoft Research. A preliminary version of this work was published in the VLDB 2006 conference proceedings.     

Finite Elements in the Presence of Occasionally Binding Constraints

This paper proposes and compares in terms of speed and accuracy two alternative approximation methods employing finite elements to parameterize the true policy functions that solve for the equilibrium of an optimal growth model with leisure and irreversible investment. The occasionally binding constraint in investment is efficiently handled on one algorithm by parameterizing the expectations of the marginal benefit of future physical capital stock and on the other by modifying the planner’s problem to include a penalty function. While both methods benefit from the high speed and accuracy achieved by a finite elements approximation, the algorithm incorporating a penalty function in the planner’s problem proved being of fastest convergence over a whole range of the model’s calibrations.     

Multisensor fusion-based gas detection module

This article develops a gas detection module for the intelligent home. The module uses eight gas sensors to detect the environment of the home and building. The gas sensors of the module have an NH₃ sensor, an air pollution sensor, an alcohol sensor, an HS sensor, a smoke sensor, a CO sensor, an LPG sensor, and a natural gas sensor, and can classify more than eight types of gas using multisensor fusion algorithms. In the logical filter method, either AND or OR filters can be implemented in the gas detection module. Then we can calculate the system’s reliability using the AND and OR filters, and classify the type of gas in the environment. The controller of the gas detection module is a HOLTEK microchip. The module can communicate with the supervised computer via a wire interface or a wireless RF interface, and can caution the user via a voice module. Finally, we present some experimental results to measure unknown gases using the gas detection module on the security system of an intelligent building and home.     

Neural network training based on FPGA with floating point number format and it��s performance

In this paper, two-layered feed forward artificial neural network’s (ANN) training by back propagation and its implementation on FPGA (field programmable gate array) using floating point number format with different bit lengths are remarked based on EX-OR problem. In the study, being suitable with the parallel data-processing specification on ANN’s nature, it is especially ensured to realize ANN training operations parallel over FPGA. On the training, Virtex2vp30 chip of Xilinx FPGA family is used. The network created on FPGA is coded by using VHDL. By comparing the results to available literature, the technique developed here proved to consume less space for the subjected ANN training which has the same structure and bit length, it is shown to have better performance.     

An Electronic Commerce Application of the Bayesian Framework for MLPs: The Effect of Marginalisation and ARD

Within the Bayesian approach to the training of multi-layer perceptrons for classification problems, the interpretation of the outputs as posterior probabilities of class-membership requires us to integrate out (marginalise) the network function over the distribution of network weights. MacKay [1] suggests an approximation of such an analytically intractable integral, in which the integration is over the network output preactivations. The network predictions can be over-optimistic if this process of marginalisation is ignored. This study attempts to assess the effect of marginalisation, with the approximation mentioned above, on two Bayesian neural network models: one with a single regularisation term; and another giving way to a process known as   Automatic Relevance Determination (ARD), with multiple regularisation terms. A real-world classification problem, concerning the discrimination of online purchasers and non-purchasers using Internet’s WWW users’ opinions, is the test-bed for this assessment.     

An Equitable Bandwidth Allocation Model for Video-on-Demand Networks

Video-on-Demand (VOD) or near-VOD services are expected to grow significantly over time, providing diverse programs for home entertainment, learning and training, news-on-demand, and other applications. These services require large bandwidth resources. We present a model for bandwidth allocation in a tree network with limited link capacities, where a server at the root node repeatedly broadcasts copies of various programs. The time intervals between successive broadcasts of each program can be increased at subsequent nodes, or the video quality can be decreased, thus providing different service performance to different nodes while satisfying the capacity constraints. The model is formulated as an equitable resource allocation problem with a lexicographic minimax objective function and tree-like ordering constraints. We present a lexicographic minimax algorithm that allocates each link’s bandwidth among the programs carried on the link. The algorithm repeatedly solves minimax problems, and fixes some variables at their optimal value after the solution of each such problem. The algorithm for solving the minimax problems uses a bisection search to find the minimax solution with the minimal decision variable values. The model also provides an ordered list of links from the most critical link to the least critical link, a useful feature for capacity expansion planning decisions.     

Task scheduling with ANN-based temperature prediction in a data center: a simulation-based study

High temperatures within a data center can cause a number of problems, such as increased cooling costs and increased hardware failure rates. To overcome this problem, researchers have shown that workload management, focused on a data center’s thermal properties, effectively reduces temperatures within a data center. In this paper, we propose a method to predict a workload’s thermal effect on a data center, which will be suitable for real-time scenarios. We use machine learning techniques, such as artificial neural networks (ANN) as our prediction methodology. We use real data taken from a data center’s normal operation to conduct our experiments. To reduce the data’s complexity, we introduce a thermal impact matrix to capture the spacial relationship between the data center’s heat sources, such as the compute nodes. Our results show that machine learning techniques can predict the workload’s thermal effects in a timely manner, thus making them well suited for real-time scenarios. Based on the temperature prediction techniques, we developed a thermal-aware workload scheduling algorithm for data centers, which aims to reduce power consumption and temperatures in a data center. A simulation study is carried out to evaluate the performance of the algorithm. Simulation results show that our algorithm can significantly reduce temperatures in data centers by introducing an endurable decline in performance.     

基于垂直采样的水下三维传感网覆盖算法

针对水下环境的三维传感器网络节点随机部署时存在覆盖率低的问题,设计一种基于垂直采样的水下三维传感网络覆盖算法,用于提高水下三维传感器网络覆盖率和连通性.垂直采样算法首先对三维监测区域进行垂直平面采样,然后再对该平面进行直线采样,把三维空间的覆盖问题转化为多平面内的直线覆盖优化问题,达到对整个三维网络覆盖优化的目的.仿真结果表明,在100 m×100 m×100 m的三维监测水域,垂直采样算法比三维随机部署策略可提高约4%~28%的覆盖率,在节点数为40时对覆盖率的提升程度最大.     

Soft-computing methods for robust authentication using soft-biometric data

Biometrics is the measurement of person’s physiological or behavioral characteristics. It enables authentication of a person’s identity using such measurements. Biometric-based authentication is thus becoming increasingly important in computer-based applications because the amount of sensitive data stored in such systems is growing. Particularly challenging is the implementation of biometric-based authentication in embedded computer system applications, because the resources of such systems are scarce. Reliability and performance are two primary requirements to be satisfied in embedded system applications. Single-mode and hard-feature-based biometrics do not offer enough reliability and performance to satisfy such requirements. Multimode biometrics is a primary level of improvement. Soft-biometric features can thus be considered along with hard-biometric features to further improve performance. A combination of soft-computing methods and soft-biometric data can yield more improvements in authentication performance by limiting requirements for memory and processing power. The multi-biometric approach also increases system reliability, since most embedded systems can capture more than one physiological or behavioral characteristic. A multi-biometric platform that combines voiceprint and fingerprint authentication was developed as a reference model to demonstrate the potential of soft-computing methods and soft-biometric data. Hard-computing pattern-matching algorithms were applied to match hard-biometric features. Artificial neural network (ANN) processing was applied to match soft-biometric features. Both hard-computing and soft-computing matching results are inferred by a fuzzy logic engine to perform smart authentication using a decision-fusion paradigm. The embedded implementation was based on a single-chip, floating-point, digital signal processor (DSP) to demonstrate the practical embeddability of such an approach and the improved performance that can be attained despite limited system resources.     

A Neural Network Approach to Predicting Airspeed in Helicopters

A helicopter’s airspeed and sideslip angle is difficult to measure at speeds below 50 knots. This paper describes the application of Artificial Neural Network (ANN) techniques to the helicopter low air-speed problem. Three ANN methods were applied to the problem: a linear network, a Radial Basis Function (RBF) network, and a Multi-Layer Perceptron (MLP), trained using an implementation of the Levenberg–Marquardt (L–M) algorithm. Internally available measurements, such as control positions and body attitudes and rates, were generated using a realistic simulation model of a Lynx helicopter. These measurements formed the inputs to the ANN methods. The MLP was found to be the superior method. Further testing, including a Tagu-chi analysis, indicated the validity of the method. It is concluded that ANN techniques present a promising solution to the helicopter low airspeed problem.