Low-rank matrix approximation plays an important role in the area of computer vision and image processing. Most of the conventional low-rank matrix approximation methods are based on the l2 -norm (Frobenius norm) with principal component analysis (PCA) being the most popular among them. However, this can give a poor approximation for data contaminated by outliers (including missing data), because the l2 -norm exaggerates the negative effect of outliers. Recently, to overcome this problem, various methods based on the l1 -norm, such as robust PCA methods, have been proposed for low-rank matrix approximation. Despite the robustness of the methods, they require heavy computational effort and substantial memory for high-dimensional data, which is impractical for real-world problems. In this paper, we propose two efficient low-rank factorization methods based on the l1 -norm that find proper projection and coefficient matrices using the alternating rectified gradient method. The proposed methods are applied to a number of low-rank matrix approximation problems to demonstrate their efficiency and robustness. The experimental results show that our proposals are efficient in both execution time and reconstruction performance unlike other state-of-the-art methods.

Recovering a non-rigid 3D structure from a series of 2D observations is still a difficult problem to solve accurately. Many constraints have been proposed to facilitate the recovery, and one of the most successful constraints is smoothness due to the fact that most real-world objects change continuously. However, many existing methods require to determine the degree of smoothness beforehand, which is not viable in practical situations. In this paper, we propose a new probabilistic model that incorporates the smoothness constraint without requiring any prior knowledge. Our approach regards the sequence of 3D shapes as a simple stationary Markov process with Procrustes alignment, whose parameters are learned during the fitting process. The Markov process is assumed to be stationary because deformation is finite and recurrent in general, and the 3D shapes are assumed to be Procrustes aligned in order to discriminate deformation from motion. The proposed method outperforms the state-of-the-art methods, even though the computation time is rather moderate compared to the other existing methods.

Tensor decomposition is frequently used in image processing and machine learning for its ability to express higher order characteristics of data. Among tensor decomposition methods, N-mode singular value decomposition (SVD) is widely used owing to its simplicity. However, the data dimension often becomes too large to perform N-mode SVD directly due to memory limitation. An incremental method to N-mode SVD can be used to resolve this issue, but existing approaches only provide a result, which is just enough to solve discriminative problems, not the full factorization result. In this paper, we present a complete derivation of the incremental N-mode SVD, which can be applied to generative models, accompanied by a technique that can reduce the computational cost by reordering calculations. The proposed incremental N-mode SVD can also be used effectively to update the current result of N-mode SVD when new training data is received. The proposed method provides a very good approximation of N -mode SVD for the experimental data, and requires much less computation in updating a multilinear model.

Non-rigid structure from motion is a fundamental problem in computer vision, which is yet to be solved satisfactorily. The main difficulty of the problem lies in choosing the right constraints for the solution. In this paper, we propose new constraints that are more effective for non-rigid shape recovery. Unlike the other proposals which have mainly focused on restricting the deformation space using rank constraints, our proposal constrains the motion parameters so that the 3D shapes are most closely aligned to each other, which makes the rank constraints unnecessary. Based on these constraints, we define a new class of probability distribution called the Procrustean normal distribution and propose a new NRSfM algorithm, EM-PND. The experimental results show that the proposed method outperforms the existing methods, and it works well even if there is no temporal dependence between the observed samples.

Albedo estimation from a facial image is crucial for various computer vision tasks, such as 3-D morphable-model fitting, shape recovery, and illumination-invariant face recognition, but the currently available methods do not give good estimation results. Most methods ignore the influence of cast shadows and require a statistical model to obtain facial albedo. This paper describes a method for albedo estimation that makes combined use of image intensity and facial depth information for an image with cast shadows and general unknown light. In order to estimate the albedo map of a face, we formulate the albedo estimation problem as a linear programming problem that minimizes intensity error under the assumption that the surface of the face has constant albedo. Since the solution thus obtained has significant errors in certain parts of the facial image, the albedo estimate needs to be compensated. We minimize the mean square error of albedo under the assumption that the surface normals, which are calculated from the facial depth information, are corrupted with noise. The proposed method is simple and the experimental results show that this method gives better estimates than other methods.

Bipedal walking robots are inherently hybrid systems due to their intermittent, switching dynamics resulting from the impact between the robot foot and the ground as the robot foot lands on the ground. It is well known that stable (passive) limit cycles for the biped robots can be induced on shallow slopes without actuation. Recently the studies in passive dynamic walking have considered the robots with knee and point or curved feet. In this paper, we study the passive dynamic walking for biped robots with knee and fixed flat feet, which includes heel and toe rocking motions and the effect of foot length on the passive limit cycles. We derive the dynamic equations of motion for this model. We show by simulation that the proposed robot model can walk down a slope passively and also verify the stability of this walking by calculating the eigenvalues of the Jacobian of the Poincarè map. By using a numerical search method, we find the initial conditions of the stable limit cycles for various slope angles and foot lengths.

We propose a new biased discriminant analysis (BDA) using composite vectors for eye detection. A composite vector consists of several pixels inside a window on an image. The covariance of composite vectors is obtained from their inner product and can be considered as a generalization of the covariance of pixels. The proposed composite BDA (C-BDA) method is a BDA using the covariance of composite vectors. We construct a hybrid cascade detector for eye detection, using Haar-like features in the earlier stages and composite features obtained from C-BDA in the later stages. The proposed detector runs in real time; its execution time is 5.5 ms on a typical PC. The experimental results for the CMU PIE database and our own real-world data set show that the proposed detector provides robust performance to several kinds of variations such as facial pose, illumination, eyeglasses, and partial occlusion. On the whole, the detection rate per pair of eyes is 98.0% for the 3604 face images of the CMU PIE database and 95.1% for the 2331 face images of the real-world data set. In particular, it provides a 99.7% detection rate for the 2120 CMU PIE images without glasses. Face recognition performance is also investigated using the eye coordinates from the proposed detector. The recognition results for the real-world data set show that the proposed detector gives similar performance to the method using manually located eye coordinates, showing that the accuracy of the proposed eye detector is comparable with that of the ground-truth data.

We propose a probing-based link adaptation scheme that is incorporated into the aggregation with fragment retransmission (AFR) scheme for high data rate wireless LANs (WLANs). Thanks to the frame structure of multiple fragments in AFR, the proposed scheme can significantly reduce the probing overhead by transmitting a few fragments in a frame at a higher data rate. At the same time, it can easily estimate the fragment error rate in the MAC layer. To maximize the throughput and maintain the average fragment error rate below the target value, the data rate can be switched to a higher data rate based on the probing result or a more robust data rate based on the estimated fragment error rate. The proposed scheme does not need a calibration process, which is a major difficulty in implementing the exponential effective SNR mapping based fast link adaptation (FLA/EESM) scheme. Moreover, it is less affected by imperfect channel matrix estimation compared to the FLA/EESM scheme. In the performance evaluation for several channel models, the proposed scheme provides 3~10% and 5~15% more throughput than the FLA/EESM scheme with perfect and imperfect channel matrix estimations, respectively, while maintaining the average fragment error rate below the target value.

For high-speed 802.11n WLANs, we propose a joint rate and fragment size adaptation scheme incorporated into the aggregation scheme, AFR (Aggregation with Fragment Retransmission), in which multiple packets are f rstly fragmented and then aggregated into in a single large frame. We investigate how to control the frame and fragment size as well as the transmission rate in dynamic channel condition. In the proposed scheme, the transmitter estimates the fragment error probability in the MAC layer, which can characterize channel quality, and uses simple look-up table of thresholds which are obtained from the perfow throughput equation for various PHY modes and fragment sizes. Through extensive simulations, we show that the proposed scheme has comparable throughput performance with the ideal performance which is obtained under the assumption that perfect channel information can be available at the transmitter as well as has higher throughput than the existing open-loop rate adaptation scheme.

Albedo estimation from a facial image is crucial for various computer vision tasks such as 3D morphable model fitting, shape recovery and illumination-invariant face recognition, but it has not been addressed well. This paper describes how albedo can be estimated from the combined use of image in tensity and facial depth information even if the image is under general,unknown light with cast shadow. In order to estimate the albedo map, we formulate the albedo estimation problem as a linear programming using 1-norm under the assumption that the surface of a face has a constant albedo. The proposed method is simple and the experiment results show that the proposed approach gives better albedo estimation than other methods.

In this paper, we propose Period-Controlled Medium Access Control (PC-MAC), which can operate in pseudo-TDMA manner and achieves high throughput and fairness in simple networks. PC-MAC works like CSMA/CA initially and becomes a pseudo-TDMA scheme in a few seconds due to the periodic backoff mechanism along with the contention control that tries to maintain the number of idle slots to an optimal level. Simulation results show 10%-50% higher throughput than distributed coordination function (DCF), depending on the number of nodes, while maintaining nearly perfect fairness. Furthermore, we also show how PC-MAC can successfully be applied to complex networks.

We propose a pixel selection method in a face image based on discriminant features for face recognition. By analyzing the relationship between the pixels in a face image and features extracted from face images, pixels that contain a large amount of discriminative information are selected, while pixels with less discriminative information are discarded. The proposed method orders the pixels based on the discriminative information in face recognition, instead of selecting salient a priori regions. Comparative experiments are performed using the FERET, CMU-PIE and Yale B databases. The experimental results show that the pixel selection results in improved recognition performance, especially under illumination variation.

We propose a new discriminant analysis using composite vectors for eye detection. A composite vector consists of a number of pixels inside a window on an image. The covariance of composite vectors is obtained from their inner product and can be considered as a generalized form of the covariance of pixels. The proposed C-BDA is a biased discriminant analysis using the covariance of composite vectors. In the hybrid cascade detector constructed for eye detection, Haar-like features are used in the earlier stages and composite features obtained from C-BDA are used in the later stages. The experimental results for the CMU and Yale databases show that the proposed detector provides robust performance to several kinds of variations such as facial pose, illumination, and closed eyes. In particular, it provides a 99.4% detection rate for the CMU images without glasses.

In this paper, we propose a new kernel discriminant analysis using composite vectors (C-KDA). We show that employing composite vectors is similar to using more samples by analysis, which is a great advantage in classification problems when the size of training samples is small. Motivated by this, we apply composite vectors to kernel-based methods, which may have overfitting problems when training samples are not sufficient. Experimental results using several data sets from UCI machine learning repository show that C-KDA gives a better performance compared to other methods based on primitive input variables and linear discriminant analysis using composite vectors (C-LDA) when the training sample size is relatively small.

In wireless mesh networks, the end-to-end throughput of a flow over a multiple-hop wireless link rapidly decreases as the hop-count between the source and destination nodes increases, and the flows that travel over a path of more than 4-5 hops from its source node eventually starve. To alleviate this unfairness, we propose a weighted random early detection (RED) mechanism that has a different dropping preference according to the hop-count information. When the network is congested, it forwards packets that come from a farther source node with a higher priority. Through extensive simulations, we show that the proposed queue management policy can improve the fairness performance effectively and alleviate the starvation problem in multihop wireless networks.

In this paper, we investigate the issue of per-station fairness in TCP over IEEE 802.11-compliant wireless local area networks (WLANs), especially in Wi-Fi hot spot. It is asserted that the hot spot suffers from the unfairness among stations in exploiting the wireless medium. The source of this unfairness is analyzed from two aspects, TCP-induced asymmetry and MAC-induced asymmetry; the former causes TCP congestion control with a cumulative acknowledgment mechanism to prefer the sending stations to receiving stations, while the later exacerbates the unfairness problem in the hot spots. We investigate the interaction between TCP congestion control and MAC contention control, and propose a cross-layer feedback approach to assure per-station fairness and to ensure high channel utilization. In this approach, we introduce the notion of channel access cost to quantify the system-wide traffic load and per-station channel usage. The access cost is estimated at the MAC in an access point and conveyed to the TCP sender. Then, the TCP sender adjusts its sending rate based on the access cost, so as to assure per-station fairness. The simulation results indicate that the proposed approach can provide both per-station fairness and high channel utilization, irrespective of network configurations.

Illumination and pose variations that occur on face images degrade the performance of face recognition. In this paper, we propose a novel approach for handling illumination and pose variations for face recognition simultaneously. We use the two-dimensional view-based face recognition method and the shadow compensation method to deal with both variations. We construct a subspace for each pose and use the relationship between facial feature points to identify the poses. Since most human faces are similar in shape, we can find the shadow characteristics that the illumination variation makes on a face depending on the direction of light. By using these characteristics, we can compensate for illumination variation in face images. The proposed method is simple and requires much less computational effort than the other methods based on 3D models, and at the same time, provides a comparable recognition rate.

TFRC is a unicast transport layer protocol, which is designed for media streaming. Since TFRC is designed to provide nearly the same amount of throughput as that of TCP on wired networks, TFRC considers that all the losses are caused by congestion. Consequently, TFRC flow over wireless link experiences performance degradation. To improve TFRC performance over wireless links, we propose TFRC with additional sequence number (TFRC-ASN) scheme. TFRC-ASN is designed to discriminate wireless losses from congestion losses when the receiver estimates the packet loss interval. In TFRC-ASN, additional sequence number (ASN) is used at an access point or a base station to count the number of packets sent over wireless link. In contrast to existing schemes, TFRC-ASN can discriminate accurately wireless losses and also can estimate packet error rate in wireless link by observing TFRC sequence number and ASN. In addition to performance improvement, the loss discrimination algorithm used in TFRC-ASN does not need any information from the layers below the transport layer.

As public deployment of wireless local area networks (WLANs) has increased and various applications with different service requirements have emerged, fairness and quality of service (QoS) are two imperative issues in allocating wireless channels. This study proposes a fair QoS agent (FQA) to simultaneously provide per-class QoS enhancement and per-station fair channel sharing in WLAN access networks. FQA implements two additional components above the 802.11 MAC: a dual service differentiator and a service level manager. The former is intended to improve QoS for different service classes by differentiating service with appropriate scheduling and queue management algorithms, while the latter is to assure fair channel sharing by estimating the fair share for each station and dynamically adjusting the service levels of packets. FQA assures (weighted) fairness among stations in terms of channel access time without decreasing channel utilization. Furthermore, it can provide quantitative service assurance in terms of queuing delay and packet loss rate. FQA neither resorts to any complex fair scheduling algorithm nor requires maintaining per-station queues. Since the FQA algorithm is an add-on scheme above the 802.11 MAC, it does not require any modification of the standard MAC protocol. Extensive ns-2 simulations confirm the effectiveness of the FQA algorithm with respect to the per class QoS enhancement and per-station fair channel sharing

Differentiated service (DiffServ) networks have been proposed to assure the achievable minimum bandwidth to aggregate flows. However, analyses in the literature show that the current DiffServ networks are biased in favor of ah aggregate flow that has a smaller committed information rate (CIR) when aggregate flows with different CIRs share a bottleneck link. In order to mitigate this unfairness problem, we propose an adaptive marking scheme which provides the relative bandwidth assurance in proportion to the CIRs of the aggregates. By introducing a virtual target rate (VTR) and adjusting it depending on the provision level of the network, each aggregate can obtain its fair share of the bandwidth, regardless of traffic load. This scheme is based on a feedback approach. It utilizes only two-bit feedback information conveyed in the packet header and can be implemented in a distributed manner. Furthermore, the proposed scheme does not require calculating fair shares of aggregates or any additional signaling protocol. Using steady state analysis and extensive simulations, we show that the scheme can provide aggregate flows with their fair shares of bandwidth, which is proportional to the CIRs, under various network conditions

In WLANs, fairness is an important issue because the channel is shared by many users. This paper proposes a dual queue based scheme in an access point (AP) for TCP fairness among uplink flows and downlink flows. First, we discuss the unfairness problem of TCP flows due to different responses to data packet drop and TCP ACK packet drop at the AP. The proposed scheme employs two queues, one for the data packets of downlink TCP flows and another for the ACK packets corresponding to uplink TCP flows. The performance of the proposed scheme is evaluated by simulation and the results are presented. The dual queue scheme is simple and effective for resolving the TCP unfairness problem.

This paper focuses on the unfairness problem between TCP uplink and downlink flows in the 802.11 Wi-Fi hot spots and shows that the service is prone to be unfair. The cause of unfairness is analyzed from two aspects: TCP-induced asymmetry and MAC-induced asymmetry. Due to the asymmetric behavior of TCP congestion control with a cumulative acknowledgment mechanism between uplink and downlink flows, the service is biased toward the uplink flow and the downlink flow tends to starve. The contention-based channel access mechanism of 802.11 MAC exacerbates this unfairness problem because it intends to provide fair access opportunity only to the sending stations. Next, the analysis of the interaction between congestion control of TCP and contention control of MAC reveals interesting and counter-intuitive results: (i) Even when a station has a sufficiently large amount of traffic to send, it does not always participate in the MAC-layer contention, its opportunity for MAC-layer contention is controlled by the TCP congestion control, (ii) The aggregate throughput remains almost constant with respect to the number of stations sending/receiving TCP traffic. (Hi) Both TCP-induced unfairness and MAC-induced unfairness can be resolved if packet loss due to buffer overflow in an access point does not occur.

Despite years of research and development, pioneering deployments of multihop wireless networks have not proven successful. The performance of routing and transport is often unstable due to contention-induced packet losses, especially when the network is large and the offered load is high. In this paper we propose RAIN, a reliable wireless network architecture for large-scale multihop wireless networks. A RAIN network enforces contention control by limiting the queue length at intermediate wireless routers to the minimum. To keep the queue short a RAIN network enforces congestion control through in-network implicit back-pressure. RAIN congestion control is built on wireless datalink layer mechanisms, e.g., mandatory per-frame acknowledgement and inter-frame backoff in popular CSMA/CA wireless transceivers, therefore very efficient and effective compared with those defined at the network or transport layer for the wired Internet. As a result of the built-in contention and congestion control, RAIN presents the end hosts a highly reliable network service model, even more reliable than that of the wired Internet. The end hosts only need to deal with packet losses due to router or routing failures. Therefore, the transport protocol can be significantly simplified. This is in stark contrast to the existing approach of adding more and more complexity to adapt TCP for multihop wireless networks. We propose the details of RAIN datalink layer protocol, and a simple transport protocol at the end hosts. Performance evaluation through intensive simulations shows that RAIN improves the throughput by up to 92% and fairness by up to 48%, with packet losses due to contention and congestion significantly reduced.

This paper proposes a combined subspace method using both global and local features for face recognition. The global and local features are obtained by applying the LDA-based method to either the whole or part of a face image, respectively. The combined space is constructed with the projection vectors corresponding to large eigenvalues of the between-class scatter matrix in each subspace. It is based on the fact that the eigenvectors corresponding to larger eigenvalues have more discriminating power. The combined subspace is evaluated in view of the Bayes error, which shows how well samples can be classified. The combined subspace gives small Bayes error than the subspaces composed of either the global or local features. Comparative experiments are also performed using the color FERET database of facial images. The experimental results show that the combined subspace method gives better recognition rate than other methods.

In this paper, we consider the problem of how to represent the locations of Internet hosts in a Cartesian coordinate system to facilitate estimation of network distances among arbitrary Internet hosts. We envision an infrastructure that consists of beacon nodes and provides the service of estimating network distance between pairs of hosts without direct delay measurement. We show that the principal component analysis (PCA) technique can effectively extract topological information from delay measurements between beacon hosts. Based on PCA, we devise a transformation method that projects the raw distance space into a new coordinate system of (much) smaller dimensions. The transformation retains as much topological information as possible and yet enables end hosts to determine their coordinates in the coordinate system. The resulting new coordinate system is termed as the Internet Coordinate System (ICS). As compared to existing work (e.g., IDMaps and GNP), ICS incurs smaller computation overhead in calculating the coordinates of hosts and smaller measurement overhead (required for end hosts to measure their distances to beacon hosts). Finally, we show via experiments with both real-life and synthetic data sets that ICS makes robust and accurate estimates of network distances, incurs little computational overhead, and its performance is not susceptible to the number of beacon nodes (as long as it exceeds a certain threshold) and the network topology.

IEEE 802.11 has become a dominant standard for wireless local area network (WLAN), and the users' demand for high throughput of IEEE 802.11 is continuously increasing. We propose a new medium access control scheme, TDM-based coordination function (TCF), which can be used when all the stations are within the radio transmission range. TCF uses information on the number of active stations explicitly to eliminate the contention period in the DCF (distributed coordination function) of IEEE 802.11. TCF can improve overall throughput and provide fair sharing of resources among active stations. Also, TCF is simple and can be implemented distributively. Simulation is performed to compare TCF with DCF and FCR (fast collision recovery) in terms of performance.

By analyzing the steady state throughput of TCP flows in differentiated service (DiffServ) networks, we show that current DiffServ networks are biased in favor of those flows that have a smaller target rate, which results in unfair bandwidth allocation. In order to solve this unfairness problem, we propose an adaptive marking scheme, which allocates bandwidth in a manner which is proportional to the target rates of the aggregate TCP flows in the DiffServ network. This scheme adjusts the target rate according to the congestion level of the network, so that the aggregate flow can obtain its fair share of the bandwidth. Since it utilizes edge-to-edge feedback information without measuring or keeping any per-flow state, this scheme is scalable and does not require any additional signaling protocol or any significant changes to the current TCP/IP protocol. It can he implemented in a distributed manner using only two-bit feedback information, which is carried in the TCP acknowledgement. Using extensive simulations, we show that the proposed scheme can provide each aggregate flow with its fair share of the bandwidth, which is proportional to the target rate, under various network conditions.

We propose an adaptive token bucket algorithm for achieving proportional sharing of bandwidth among aggregate flows in differentiated service (DiffServ) networks. By observing the simulation results obtained in a study of the throughput of TCP flows in a DiffServ network, we note that the aggregate flow with a lower target rate occupies more bandwidth than its fair share, while the aggregate flow with a higher target rate gets less than its fair share. The proposed algorithm solves this unfairness problem by adjusting the target rate according to the edge-to-edge feedback information. This algorithm does not require any additional signaling protocol or measurement of pen-flow states, since it can be implemented in a distributed manner using only two-bit feedback information carried in the TCP acknowledgement. Using ns-2 simulations, we show that the proposed algorithm provides fair bandwidth sharing under various network conditions.

In this paper, we consider the problem of how to represent the locations of Internet hosts in a Cartesian coordinate system to facilitate estimate of the network distance between two arbitrary Internet hosts. We envision an infrastructure that consists of beacon nodes and provides the service of estimating network distance between two hosts without direct delay measurement. We show that the principal component analysis (PCA) technique can effectively extract topological information from delay measurements between beacon hosts. Based on PCA, we devise a transformation method that projects the distance data space into a new coordinate system of (much) smaller dimensions. The transformation retains as much topological information as possible and yet enables end hosts to easily determine their locations in the coordinate system. The resulting new coordinate system is termed as the Internet Coordinate System (ICS). As compared to existing work (e.g., IDMaps [1] and GNP [2]), ICS incurs smaller computation overhead in calculating the coordinates of hosts and smaller measurement overhead (required for end hosts to measure their distances to beacon hosts). Finally, we show via experimentation with real-life data sets that ICS is robust and accurate, regardless of the number of beacon nodes (as long as it exceeds certain threshold) and the complexity of network topology.

In manipulating data such as in supervised learning, we often extract new features from the original features for the purpose of reducing the dimensions of feature space and achieving better performance. In this paper, we show how standard algorithms for independent component analysis (ICA) can be appended with binary class labels to produce a number of features that do not carry information about the class labels-these features will be discarded-and a number of features that do. We also provide a local stability analysis of the proposed algorithm. The advantage is that general ICA algorithms become available to a task of feature extraction for classification problems by maximizing the joint mutual information between class labels and new features, although only for two-class problems. Using the new features, we can greatly reduce the dimension of feature space without degrading the performance of classifying systems.

The virtual rate control (VRC) algorithm has been proposed for active queue management (AQM) in TCP networks. This algorithm uses an adaptive rate control instead of queue length control in order to respond quickly to traffic change with high utilization and small loss. By introducing the notion of virtual target rate, the VRC algorithm can maintain an input rate around the target rate, while attempting to regulate the queue length. In this paper, we analyze the stability of the VRC algorithm in a linearized model. From the results of our analysis, we provide a design guideline for the system to remain stable. We show the validity of our analysis and the effectiveness of the VRC algorithm compared to RED, PI, REM and AVQ algorithms through ns-2 simulations.

We have explored a new method of feature extraction for face recognition. It is based on independent component analysis (ICA), but unlike original ICA, one of the unsupervised learning methods, it is developed to be well suited for classification problems by utilizing class information. By using ICA in solving supervised classification problems, we can obtain new features which are made as independent from each other as possible and which convey the class information faithfully. We have applied this method on Yale face databases and AT and T face databases and compared the performance with those of conventional methods such as principal component analysis (PCA), Fisher's linear discriminant (FLD), and so on. The experimental results show that for both databases the proposed method outperforms the others.

Mutual information is a good indicator of relevance between variables, and have been used as a measure in several feature selection algorithms. However, calculating the mutual information is difficult, and the performance of a feature selection algorithm depends on the accuracy of the mutual information. In this paper, we propose a new method of calculating mutual information between input and class variables based on the Parzen window, and we apply this to a feature selection algorithm for classification problems.

Feature selection plays an important role in classifying systems such as neural networks (NNs). We use a set of attributes which are relevant, irrelevant or redundant and from the viewpoint of managing a dataset which can be huge, reducing the number of attributes by selecting only the relevant ones is desirable. In doing so, higher performances with lower computational effort is expected. In this paper, we propose two feature selection algorithms. The limitation of mutual information feature selector (MIFS) is analyzed and a method to overcome this limitation is studied. One of the proposed algorithms makes more considered use of mutual information between input attributes and output classes than the MIFS. What is demonstrated is that the proposed method can provide the performance of the ideal greedy selection algorithm when information is distributed uniformly. The computational load for this algorithm is nearly the same as that of MIFS. In addition, another feature selection algorithm using the Taguchi method is proposed. This is advanced as a solution to the question as to how to identify good features with as few experiments as possible. The proposed algorithms are applied to several classification problems and compared with MIFS. These two algorithms can be combined to complement each other's limitations. The combined algorithm performed well in several experiments and should prove to be a useful method in selecting features for classification problems.

We propose a model based disturbance attenuator (MBDA) with the conventional PD controller for robot manipulators. It is a generalization of the MBDA structure in Choi et al. (1999) and is applied to a robot manipulator which is nonlinear. This method does not require an accurate model of a robot manipulator and takes care of disturbances or modeling errors so that the plant output remains relatively unaffected by them. The output error due to the gravity or constant disturbance can be completely eliminated by this method in the same way as PID controllers. In addition, this can be easily implemented at a moderate computational cost. We apply this to a two-link robot manipulator and compare its performance with PD and PID controllers. Simulation results show that the proposed method is very effective in controlling robot manipulators.

It is shown that a multilayer perceptron can learn highly nonlinear continuous mappings more easily if the target values are thermometer-coded. Because of the similarity between sigmoidal functions and thermometer coding, a network does not need more hidden nodes to produce thermometer-coded target values when more output nodes are added to the original structure. Furthermore, the weights for the hidden layers of a network, which are trained by thermometer-coded target values, can be used in the initialization of a network which is then trained by the original target values. The reason why such a two-staged learning is possible is discussed. Experiments on synthetic data sets show that using thermometer-coded target values improves the learning performance of a network and that conversion to a single output network is more efficient and gives better results.

In classification problems, we use a set of attributes which are relevant, irrelevant or redundant. By selecting only the relevant attributes of the data as input features of a classifying system and excluding redundant ones, higher performance is expected with smaller computational effort. We propose an algorithm of feature selection that makes more careful use of the mutual informations between input attributes and others than the mutual information feature selector (MIFS). The proposed algorithm is applied in several feature selection problems and compared with the MIFS. Experimental results show that the proposed algorithm can be well used in feature selection problems.

We describe our experiments on a real-time system design, focusing on design alternatives such as scheduling jitter, sensor-to-output latency, intertask communication schemes and the system utilization. The prime objective of these experiments was to evaluate a real-time design produced using the period calibration method (Gerber et al., 1995) and thus identify the limitations of the method. We chose a computerized numerical control (CNC) machine as our target real-time system and built a realistic controller and a plant simulator. Our results were extracted from a controlled series of more than a hundred test controllers obtained by varying four test variables. This study unveils many interesting facts: average sensor-to-output latency is one of the most dominating factors in determining control quality; the effect of scheduling jitter appears only when the average sensor-to-output latency is sufficiently small; and loop processing periods are another dominating factor of performance. Based on these results, we propose a new communication scheme and a new objective function for the period calibration method.

This paper proposes a new medium access control protocol for wireless data communications in local area, called the reservation-based multiple access with variable frame length (RMAV). We design RMAV under the consideration that the population of wireless terminals and the system load frequently change and are almost unpredictable in wireless data communications. RMAV is based on the slot reservation scheme and adopts a frame structure with variable length. The frame length increases as the number of active terminals and/or the system load increases. We evaluate the performance of RMAV by analysis and computer simulation. Due to its adaptability to traffic patterns, RMAV offers short delay in light load conditions and high throughput in heavy load conditions.

This paper proposes a constructive neural network with a piecewise linear or nonlinear local interpolation capability to approximate arbitrary continuous functions. This neural network is devised by introducing a space tessellation which is a covering of the Euclidean space by nonoverlapping hyperpolyhedral convex cells. In the proposed neural network, a number of neural network granules (NNG's) are processed in parallel and repeated regularly with the same structures. Each NNG does a local mapping with an interpolation capability for a corresponding hyperpolyhedral convex cell in a tessellation. The plastic weights of the NNG can be calculated to implement the mapping for training data; consequently, this reduces training time and alleviates the difficulties of local minima in training. In addition, the interpolation capability of the NNG improves the generalization for the new data within the convex cell. The proposed network requires additional neurons for tessellation over the standard multilayer neural networks. This increases the network size but does not slow the retrieval response when implemented by parallel architecture.

In a neural network, many different sets of connection weights can approximately realize an input-output mapping. The sensitivity of the neural network varies depending on the set of weights. For the selection of weights with lower sensitivity or for estimating output perturbations in the implementation, it is important to measure the sensitivity for the weights. A sensitivity depending on the weight set in a single-output multilayer perceptron (MLP) with differentiable activation functions is proposed. Formulas are derived to compute the sensitivity arising from additive/multiplicative weight perturbations or input perturbations for a specific input pattern. The concept of sensitivity is extended so that it can be applied to any input patterns. A few sensitivity measures for the multiple output MLP are suggested. For the verification of the validity of the proposed sensitivities, computer simulations have been performed, resulting in good agreement between theoretical and simulation outcomes for small weight perturbations.