Recent approaches for few-shot 3D point cloud semantic segmentation typically require a two-stage learning process, i.e., a pre-training stage followed by a few-shot training stage. While effective, these methods face overreliance on pre-training, which hinders model flexibility and adaptability. Some models tried to avoid pre-training yet failed to capture ample information. In addition, current approaches focus on visual information in the support set and neglect or do not fully exploit other useful data, such as textual annotations. This inadequate utilization of support information impairs the performance of the model and restricts its zero-shot ability. To address these limitations, we present a novel pre-training-free network, named Efficient Point Cloud Semantic Segmentation for Few- and Zero-shot scenarios. Our EPSegFZ incorporates three key components. A Prototype-Enhanced Registers Attention (ProERA) module and a Dual Relative Positional Encoding (DRPE)-based cross-attention mechanism for improved feature extraction and accurate query-prototype correspondence construction without pre-training. A Language-Guided Prototype Embedding (LGPE) module that effectively leverages textual information from the support set to improve few-shot performance and enable zero-shot inference.Extensive experiments show that our method outperforms the state-of-the-art method by 5.68% and 3.82% on the S3DIS and ScanNet benchmarks, respectively.

Large pre-trained language models have demonstrated impressive capabilities, but there is still much to learn about how they operate.In this study, we conduct an investigation of the autoregressive transformer's ability to perform basic addition operations.Specifically, by using causal analysis we found that a few different attention heads in the middle layers control the addition carry, with each head processing carries of different lengths.Due to the lack of global focus on the sequence within these attention heads, the model struggles to handle long-sequence addition tasks.By performing inference intervention on mistral-7B, partial task performance can be restored, with the accuracy on 20-digit long-sequence additions from 2% to 38%.Through fine-tuning, a new mechanism branches out for handling complex cases, yet it still faces challenges with length generalization.Our research reveals how the models perform basic arithmetic task, and further provides insights into the debate on whether these models are merely statistical.

Recent 6D pose estimation methods demonstrate notable performance but still face some practical limitations. For instance, many of them rely heavily on sensor depth, which may fail with challenging surface conditions, such as transparent or highly reflective materials. In the meantime, RGB-based solutions provide less robust matching performance in low-light and texture-less scenes due to the lack of geometry information. Motivated by these, we propose **SingRef6D**, a lightweight pipeline requiring only a **single RGB** image as a reference, eliminating the need for costly depth sensors, multi-view image acquisition, or training view synthesis models and neural fields. This enables SingRef6D to remain robust and capable even under resource-limited settings where depth or dense templates are unavailable. Our framework incorporates two key innovations. First, we propose a token-scaler-based fine-tuning mechanism with a novel optimization loss on top of Depth-Anything v2 to enhance its ability to predict accurate depth, even for challenging surfaces. Our results show a 14.41% improvement (in $\delta_{1.05}$) on REAL275 depth prediction compared to Depth-Anything v2 (with fine-tuned head). Second, benefiting from depth availability, we introduce a depth-aware matching process that effectively integrates spatial relationships within LoFTR, enabling our system to handle matching for challenging materials and lighting conditions. Evaluations of pose estimation on the REAL275, ClearPose, and Toyota-Light datasets show that our approach surpasses state-of-the-art methods, achieving a 6.1% improvement in average recall.

This article presents a novel dynamic programming approach to determine the robust controllability of Boolean control networks (BCNs) subject to stochastic disturbances. By applying Bellman's optimality principle, we derive the recurrence relation for computing the optimal time matrix, a crucial concept characterizing robust reachability between two arbitrary states. We develop a finite-termination dynamic programming algorithm to calculate the optimal time matrix exactly and efficiently, with a rigorously certified iteration count. Sufficient and necessary conditions for robust controllability are then established based on the optimal time matrix. Furthermore, for any pair of reachable states, we construct time-optimal state feedback control laws to steer the system from the initial state to the target state, regardless of disturbances. Finally, extensive numerical experiments with biological networks validate the effectiveness of the proposed approach, showing significant improvements in computational efficiency. Additionally, we introduce a Q-learning-based algorithm and compare its performance, highlighting the advantages of our dynamic programming approach in terms of both efficiency and solution quality.

Three-dimensional point cloud semantic segmentation is a fundamental task in computer vision. As the fully supervised approaches suffer from the generalization issue with limited data, few-shot point cloud segmentation models have been proposed to address the flexible adaptation. Nevertheless, due to the class-agnostic nature of the few-shot pretraining, its pretrained feature extractor is hard to capture the class-related intrinsic and abstract information. Therefore, we introduce the new concept of shallow and deep similarities and propose a shallow-deep similarity learning network (SDSimPoint) that aims to learn both shallow (superficial geometry, color, etc.) and deep similarities (intrinsic context and semantics, etc.) between the support and query samples, thereby boosting the performance. Moreover, we design a beyond-episode attention module (BEAM) to enlarge the region of the attention mechanism from a single episode to the entire dataset by utilizing the memory units, which enhances the extraction ability to better capture the shallow and deep similarities. Furthermore, our distance metric function is learnable in the proposed framework, which can better adapt to complex data distributions. Our proposed SDSimPoint consistently demonstrates substantial improvements compared to baseline approaches across various datasets in diverse few-shot point cloud semantic segmentation settings.

Automating the processing of health insurance claims to achieve "Straight-Through Processing" is one of the holy grails that all insurance companies aim to achieve. One of the major impediments to this automation is the difficulty in establishing the relationship between the underwriting exclusions that a policy has and the incoming claim's diagnosis information. Typically, policy underwriting exclusions are captured in free-text such as "Respiratory illnesses are excluded due to a pre-existing asthma condition". A medical claim coming from a hospital would have the diagnosis represented using the International Classification of Disease (ICD) codes from the World Health Organization. The complex and labour-intensive task of establishing the relationship between free-text underwriting exclusions in health insurance policies and medical diagnosis codes from health insurance claims is critical towards determining if a claim should be rejected due to underwriting exclusions. In this work, we present a novel framework that leverages both explicit and implicit domain knowledge present in medical ontologies and pre-trained language models respectively, to effectively establish the relationship between free-text describing medical conditions present in underwriting exclusions and the ICD-10CM diagnosis codes in health insurance claims. Termed KAMEL (Knowledge Aware Medical Entity Linkage), our proposed framework addresses the limitations faced by prior approaches when evaluated on real-world health insurance claims data. Our proposed framework have been deployed in several multi-national health insurance providers to automate their health insurance claims.

Battery health monitoring and prediction are critically important in the era of electric mobility with a huge impact on safety, sustainability, and economic aspects. Existing research often focuses on prediction accuracy but tends to neglect practical factors that may hinder the technology’s deployment in real-world applications. In this paper, we address these practical considerations and develop models based on the Bayesian neural network for predicting battery end-of-life. Our models use sensor data related to battery health and apply distributions, rather than single-point, for each parameter of the models. This allows the models to capture the inherent randomness and uncertainty of battery health, which leads to not only accurate predictions but also quantifiable uncertainty. We conducted an experimental study and demonstrated the effectiveness of our proposed models, with a prediction error rate averaging 13.9%, and as low as 2.9% for certain tested batteries. Additionally, all predictions include quantifiable certainty, which improved by 66% from the initial to the mid-life stage of the battery. This research has practical values for battery technologies and contributes to accelerating the technology adoption in the industry.

In this article, an event-triggered state-dependent switching method is proposed to address adaptive fuzzy control problem for a class of multi-input multi-output switched nonlinear systems with IOCs. IOCs mean that output constraints can occur in some intermittent time intervals rather than for all time. Unlike traditional continuously monitored state-dependent switching laws, an event-triggered state-dependent switching law is constructed to overcome the difficulty of the corresponding problem of subsystems being unsolvable, which is caused by unknown control gains of subsystems going through zero. Meanwhile, a positive lower bound of consecutive switching instants is derived. Also, it is achieved that adaptive fuzzy controller and switched update laws are event-triggered by designing several new switching event-triggering mechanisms, which results in the reduction of communication and computation loads. Furthermore, by developing some modified shifting functions and barrier functions, more general intermittent output constraints are handled. The effectiveness and applicability of the method proposed are verified by a mass-spring-damper system.

A home energy management system exploits the time-varying electricity tariff and renewable energy profiles to lower residents' electricity bills via wise scheduling of various domestic appliances. This study targets the rather typical case of a general household with solar panels. All four classes of loads are considered, while many existing studies only investigate a restricted subset. Considering the high stochasticity in real-time pricing and solar power generation, we propose an online approach in a hybrid semidecentralized framework, where each shiftable load is controlled by a deep neural network (DNN), and all adjustable loads are coordinated together by fast online optimization. We train each DNN via efficient and effective imitation learning (IL) instead of popular reinforcement learning (RL). This framework allows adjustable loads to react properly to possibly poor actions of shiftable loads via online one-step optimization to alleviate their adverse impact. Numerical experiments with real-world data show that, compared with RL, our approach can reduce the training time significantly, while its execution time is only slightly affected. Moreover, our approach outperforms the traditional day-ahead optimization method and the fully decentralized multiagent RL and multiagent IL methods by a wide margin, attaining an average cost fairly close to the theoretical minimum.

This study investigates economical scheduling of charging for an electric vehicle (EV) in a typical household with an intelligent charging management system. The problem formulation considers rooftop solar power generation, time-varying domestic energy consumption, real-time pricing of electricity, and user preferences. This task traditionally takes the form of a mixed-integer linear programming (MILP) problem, but we demonstrate its equivalence to linear programming (LP) to reduce computational complexity. The LP problem can be solved to global optimality if all future information is known, which is unrealistic in practice and replaced with forecasting. Learning-based methods such as deep reinforcement learning (DRL) eliminate the need for a forecaster and make online decisions rapidly using a learned policy. We propose an approach based on imitation learning that leverages the knowledge of an LP expert by learning from its optimal demonstrations instead of learning from scratch in DRL. Our approach trains a deep neural network (DNN) based policy efficiently in a supervised manner and incorporates a safety post-processing mechanism that enforces strict constraint satisfaction. Numerical studies on real-world data show that the proposed approach achieves$23~ \sim ~220$times speedup compared to DRL for DNN training, and the total electricity cost is far lower than DRL as well, which is strikingly close to the lower bound in theory. Our implementation code can be found athttps://github.com/ZhenhaoH/IL_EVCS.

This article proposes a switching event-triggered (ET) adaptive neural network (NN) output-feedback control scheme for a family of networked switched nonlinear systems under hybrid deception and denial-of-service (DoS) attacks in sensor-to-controller channel. The concept of “effective” DoS attacks is introduced for removing the assumption of the time sequences of DoS off/on and on/off transitions being known. In the active and inactive intervals of effective DoS attacks, an ET dual-switched NN observer, a dual-switched update law, common coordinate transformations in backstepping and a switching adaptive NN controller of each subsystem are constructed. Then, hybrid attacks are coped with and the difficulty in stability analysis caused by different coordinate transformations is overcome. Moreover, by designing a new switching dynamic event-triggering mechanism and a new Lyapunov function dependent on the switching signal of controller and DoS attacks, asynchronous switching between candidate subsystems and candidate observers and controllers is handled, and the convergence of tracking error to a small neighborhood around the origin is proved under a new class of switching signals with average dwell time. The effectiveness and applicability of the scheme proposed are illustrated by a switched one-link robotic manipulator system.

In recent years, the combination of event cameras and computer vision has shown increasingly excellent performance. Due to high sensitivity, event cameras are capable of addressing the issue of motion blur in conventional cameras, and are well-suited for analyzing fast-moving objects, making them highly suitable for depth estimation in UAV applications This paper focuses on methods for depth estimation using events generated by event cameras. Due to the asynchronicity of events, it is difficult to directly transmit events to the depth estimation network. So the method to preprocess events is important. Unlike existing processing methods, this paper creatively proposes the idea of adaptive stacks, which can change the size of weighted stacks in real time according to the events generation rate. In this way, we can solve the problems caused by traditional processing methods, and better utilize the effective information of events. Then, a depth estimation network corresponding to the adaptive stacks is designed to form a complete end-to-end events depth estimation model: Adaptive Stacks Depth Estimation Network (ASNet). Compared with other models, ASNet has demonstrated excellent depth estimation accuracy and has great application prospects.

This article proposes a novel approach utilizing imitation learning to address the computational challenge of multistep model predictive torque control (MPTC). The long prediction horizon of multistep MPTC usually contributes to better performance for induction motor control. Nevertheless, the computational burden increases exponentially with the length of the prediction horizon, leading to considerable difficulties in its real time implementation. MPTC essentially solves an optimization problem, which is time-consuming for a long prediction horizon. To overcome the computational difficulty, we replace the expensive numerical solver with a cheap deep neural network (DNN) following the idea of imitation learning. The DNN's output approximates the optimal solution after training. The proposed method achieves comparable steady-state performance to the ideal multistep MPTC, with lower computational complexity, especially when the switching frequency is limited. Moreover, this strategy demonstrates notable control performance improvements in contrast to the conventional one-step MPTC. Overall, the proposed method has great potential for real-time multistep MPTC implementation in typical induction motor drives.

Incremental few-shot learning is highly expected for practical robotics applications. On one hand, robot is desired to learn new tasks quickly and flexibly using only few annotated training samples; on the other hand, such new additional tasks should be learned in a continuous and incremental manner without forgetting the previous learned knowledge dramatically. In this work, we propose a novel Class-Incremental Few- Shot Object Detection (CI-FSOD) framework that enables deep object detection network to perform effective continual learning from just few-shot samples without re-accessing the previous training data. We achieve this by equipping the widely-used Faster-RCNN detector with three elegant components. Firstly, to best preserve performance on the pre-trained base classes, we propose a novel Dual-Embedding-Space (DES) architecture which decouples the representation learning of base and novel categories into different spaces. Secondly, to mitigate the catastrophic forgetting on the accumulated novel classes, we propose a Sequential Model Fusion (SMF) method, which is able to achieve long-term memory without additional storage cost. Thirdly, to promote inter-task class separation in feature space, we propose a novel regularization technique that extends the classification boundary further away from the previous classes to avoid misclassification. Overall, our framework is simple yet effective and outperforms the previous SOTA with a significant margin of 2.4 points in AP performance.

This study investigates the infinite-horizon optimal control (IHOC) problem for switched Boolean control networks with an average cost criterion. A primary challenge of this problem is the prohibitively high computational cost when dealing with large-scale networks. We attempt to develop a more efficient approach from a novel graph-theoretical perspective. First, a weighted directed graph structure called the optimal state transition graph (OSTG) is established, whose edges encode the optimal action for each admissible state transition between states reachable from a given initial state subject to various constraints. Then, we reduce the IHOC problem into a minimum-mean cycle (MMC) problem in the OSTG. Finally, we develop an algorithm that can quickly find a particular MMC by resorting to Karp's algorithm in the graph theory and construct an optimal switching control law based on state feedback. The time complexity analysis shows that our algorithm, albeit still running in exponential time, can outperform all the existing methods in terms of time efficiency. A 16-state-3-input signaling network in leukemia is used as a benchmark to test its effectiveness. Results show that the proposed graph-theoretical approach is much more computationally efficient and can reduce the running time dramatically: it runs hundreds or even thousands of times faster than the existing methods. The Python implementation of the algorithm is available at https://github.com/ShuhuaGao/sbcn_mmc.

Recent advances in high-throughput single-cell technologies provide new opportunities for computational modeling of gene regulatory networks (GRNs) with an unprecedented amount of gene expression data. Current studies on the Boolean network (BN) modeling of GRNs mostly depend on bulk time-series data and focus on the synchronous update scheme due to its computational simplicity and tractability. However, such synchrony is a strong and rarely biologically realistic assumption. In this study, we adopt the asynchronous update scheme instead and propose a novel framework called SgpNet to infer asynchronous BNs from single-cell data by formulating it into a multiobjective optimization problem. SgpNet aims to find BNs that can match the asynchronous state transition graph (STG) extracted from single-cell data and retain the sparsity of GRNs. To search the huge solution space efficiently, we encode each Boolean function as a tree in genetic programming and evolve all functions of a network simultaneously via cooperative coevolution. Besides, we develop a regulator preselection strategy in view of GRN sparsity to further enhance learning efficiency. An error threshold estimation heuristic is also proposed to ease tedious parameter tuning. SgpNet is compared with the state-of-the-art method on both synthetic data and experimental single-cell data. Results show that SgpNet achieves comparable inference accuracy, while it has far fewer parameters and eliminates artificial restrictions on the Boolean function structures. Furthermore, SgpNet can potentially scale to large networks via straightforward parallelization on multiple cores.

This article investigates the finite-horizon optimal control (FHOC) problem of Boolean control networks (BCNs) from a graph theory perspective. We first formulate two general problems to unify various special cases studied in the literature: 1) the horizon length is a priori fixed and 2) the horizon length is unspecified but finite for given destination states. Notably, both problems can incorporate time-variant costs, which are rarely considered in existing work, and a variety of constraints. The existence of an optimal control sequence is analyzed under mild assumptions. Motivated by BCNs' finite state space and control space, we approach the two general problems intuitively and efficiently under a graph-theoretical framework. A weighted state transition graph and its time-expanded variants are developed, and the equivalence between the FHOC problem and the shortest-path (SP) problem in specific graphs is established rigorously. Two algorithms are developed to find the SP and construct the optimal control sequence for the two problems with reduced computational complexity, though technically, a classical SP algorithm in graph theory is sufficient for all problems. Compared with existing algebraic methods, our graph-theoretical approach can achieve state-of-the-art time efficiency while targeting the most general problems. Furthermore, our approach is the first one capable of solving Problem 2) with time-variant costs. Finally, a genetic network in the bacterium E. coli and a signaling network involved in human leukemia are used to validate the effectiveness of our approach. The results of two common tasks for both networks show that our approach can dramatically reduce the running time. Python implementation of our algorithms is available at GitHub https://github.com/ShuhuaGao/FHOC.

We aim to tackle the challenging Few-Shot Object Detection (FSOD), where data-scarce categories are presented during the model learning. The failure modes of FasterRCNN in FSOD are investigated, and we find that the performance degradation is mainly due to the classification incapability (false positives) caused by category confusion, which motivates us to address FSOD from a novel aspect of classification refinement. Specifically, we address the intrinsic limitation from the aspects of both architectural enhancement and hard-example mining. We introduce a novel few-shot classification refinement mechanism where a decoupled Few-Shot Classification Network (FSCN) is employed to improve the final classification of a base detector. Moreover, we especially probe a commonly-overlooked but destructive issue of FSOD, i.e., the presence of distractor samples due to the incomplete annotations where images from the base set may contain novel-class objects but remain unlabelled. Retreatment solutions are developed to eliminate the incurred false positives. For FSCN training, the distractor is formulated as a semi-supervised problem, where a distractor utilization loss is proposed to make proper use of it for boosting the data-scarce classes, while a confidence-guided dataset pruning (CGDP) technique is developed to facilitate the few-shot adaptation of base detector. Experiments demonstrate that our proposed framework achieves state-of-the-art FSOD performance on public datasets, e.g., Pascal VOC and MS-COCO.

This paper works on multiple-pursuer single-evader (MPSE) problems with a fast evader, which means multiple pursuers try to capture one evader while the evader tries to escape from the encirclement. The biggest concern is that the maximum velocity of the evader is larger than all the pursuers. Some improved strategies for the evader and pursuers based on traditional algorithms are firstly provided. Then gene expression programming (GEP) is used to generate new strategies which are better than the traditional ones. This paper shows configurations of function set, terminal set, fitness, evaluation function, and other parameters used in the GEP method, which can be implemented in other cases or similar problems.

Collaborative motion planning for multi-agent systems is a challenging problem because of the existence of highly nonlinear and nonconvex constraints. Such difficulties also lead to inavoidable computational inefficiency, which significantly prohibits applying the existing collaborative motion planning algorithms to complex scenarios. This paper proposes a parallel computational algorithm to achieve collaborative motion planning efficiently, considering the nonlinear dynamics model and the nonconvex collision-avoidance constraints. Specifically, the alternating direction method of multipliers (ADMM) framework is elegantly incorporated to separate the large-scale cooperative nonconvex planning problem as two tractable and manageable subproblems, where the two subproblems handle the dynamics constraints and collision-free constraints, respectively. In the proposed approach, the differential dynamic programming (DDP) method is utilized to effectively solve the nonlinear subproblem with the dynamics constraints; meanwhile, the interior point (IPOPT) method is employed to address the nonconvex subproblem derived from the collision-avoidance constraints. Finally, two simulation scenarios are successfully implemented to illustrate the effectiveness of the proposed algorithm.

This paper presents a long-term object tracking framework with a moving event camera under general tracking conditions. A first of its kind for these revolutionary cameras, the tracking framework uses a discriminative representation for the object with online learning, and detects and re-tracks the object when it comes back into the field-of-view. One of the key novelties is the use of an event-basedlocal sliding windowtechnique that tracks reliably in scenes with cluttered and textured background. In addition, Bayesian bootstrapping is used to assist real-time processing and boost the discriminative power of the object representation. On the other hand, when the object re-enters the field-of-view of the camera, adata-driven, global sliding windowdetector locates the object for subsequent tracking. Extensive experiments demonstrate the ability of the proposed framework to track and detect arbitrary objects of various shapes and sizes, including dynamic objects such as a human. This is a significant improvement compared to earlier works that simply track objects as long as they are visible under simpler background settings. Using the ground truth locations for five different objects under three motion settings, namely translation, rotation and 6-DOF, quantitative measurement is reported for the event-based tracking framework with critical insights on various performance issues. Finally, real-time implementation in C++ highlights tracking ability under scale, rotation, view-point and occlusion scenarios in a lab setting.

We introduce a generic visual descriptor, termed as distribution aware retinal transform (DART), that encodes the structural context using log-polar grids for event cameras. The DART descriptor is applied to four different problems, namely object classification, tracking, detection and feature matching: (1) The DART features are directly employed as local descriptors in a bag-of-words classification framework and testing is carried out on four standard event-based object datasets (N-MNIST, MNIST-DVS, CIFAR10-DVS, NCaltech-101); (2) Extending the classification system, tracking is demonstrated using two key novelties: (i) Statistical bootstrapping is leveraged with online learning for overcoming the low-sample problem during the one-shot learning of the tracker, (ii) Cyclical shifts are induced in the log-polar domain of the DART descriptor to achieve robustness to object scale and rotation variations; (3) To solve the long-term object tracking problem, an object detector is designed using the principle of cluster majority voting. The detection scheme is then combined with the tracker to result in a high intersection-over-union score with augmented ground truth annotations on the publicly available event camera dataset; (4) Finally, the event context encoded by DART greatly simplifies the feature correspondence problem, especially for spatio-temporal slices far apart in time, which has not been explicitly tackled in the event-based vision domain.

In this paper, a system for the wide area surveillance of general urban environments using multiple Mini-VTOL UAVs is developed. Given the information of terrain and buildings in the target area, the problem of (robust) complete coverage of the urban environment is solved by a three-step approximation approach. Firstly, the target area and the observation area are discretized into two sets respectively. Secondly, the visibility between these two sets is checked. Finally, a set covering problem is solved based on the greedy approaches. Two case studies based on real-world data are carried out to demonstrate the effectiveness of our developed system.

Over the past few decades, a considerable amount of literature has been published on shape classification. Since classification of well-segmented shapes has become easy to achieve, a number of recent studies have emphasized the importance of robustness to noise and deformations. So in this paper, we undertake the task of classifying similar & noisy binary shape images, using a biologically inspired technique called log-polar transform (LPT). The LPT mapping technique achieves scale and rotation invariance by simulating the foveal mechanism of the human vision system. In order to ensure optimal shape representation in the log-polar space, an iterative method is presented for the LPT lattice design. In addition to optimal shape representation, the use of linear discriminant analysis is proposed for dimensionality reduction and elimination of noisy features. Besides eliminating noisy features, discriminant analysis plays a crucial role in differentiating between similar shape categories. The proposed shape classification framework is tested on five publicly available databases, and substantial boost in classification accuracy is reported compared to state-of-the-art methods. In addition to superior classification accuracy, real time performance is demonstrated using an efficient PC-based implementation.

The piecewise affine (PWA) model represents an attractive model structure for approximating nonlinear systems. In this paper, a procedure for obtaining the PWA autoregressive exogenous (ARX) (autoregressive systems with exogenous inputs) models of nonlinear systems is proposed. Two key parameters defining a PWARX model, namely, the parameters of locally affine subsystems and the partition of the regressor space, are estimated, the former through a least-squares-based identification method using multiple models, and the latter using standard procedures such as neural network classifier or support vector machine classifier. Having obtained the PWARX model of the nonlinear system, a controller is then derived to control the system for reference tracking. Both simulation and experimental studies show that the proposed algorithm can indeed provide accurate PWA approximation of nonlinear systems, and the designed controller provides good tracking performance.

Piecewise affine model is a useful tool for approximating nonlinear systems. In this paper, we first propose a procedure for obtaining the piecewise affine ARX models of nonlinear systems. Two parameters which fully characterize a piecewise affine ARX model, namely the parameters of the locally linear/affine subsystems, as well as the partitions of the regressor space, will be estimated, the former through a least-squares based identification method using multiple models, and the latter using standard procedures such as neural network classifier or support vector machine classifier. Based on the piecewise affine ARX model of the nonlinear system, we then proceed to derive a model-based controller to control the system for reference tracking. Simulation studies show that our algorithm can indeed provide accurate piecewise affine approximation of nonlinear systems, and that the proposed controller provides good tracking performance.

In this paper, output feedback adaptive neural network (NN) controls are investigated for two classes of nonlinear discrete-time systems with unknown control directions: 1) nonlinear pure-feedback systems and 2) nonlinear autoregressive moving average with exogenous inputs (NARMAX) systems. To overcome the noncausal problem, which has been known to be a major obstacle in the discrete-time control design, both systems are transformed to a predictor for output feedback control design. Implicit function theorem is used to overcome the difficulty of the nonaffine appearance of the control input. The problem of lacking a priori knowledge on the control directions is solved by using discrete Nussbaum gain. The high-order neural network (HONN) is employed to approximate the unknown control. The closed-loop system achieves semiglobal uniformly-ultimately-bounded (SGUUB) stability and the output tracking error is made within a neighborhood around zero. Simulation results are presented to demonstrate the effectiveness of the proposed control.

Recurrent neural networks, through their unconstrained synaptic connectivity and resulting state-dependent nonlinear dynamics, offer a greater level of computational ability when compared with regular feedforward neural network (FFNs) architectures. A necessary consequence of this increased capability is a higher degree of complexity, which in turn leads to gradient-based learning algorithms for RNNs being more likely to be trapped in local optima, thus resulting in sub- optimal solutions. This motivates the use of evolutionary computational methods which center about the use of population- based global-search techniques as an optimization scheme. In this article, we propose the use of a hybrid evolutionary strategy (ES) approach together with an adaptive linear observer, acting as a local search operator, as a learning mechanism for general RNN applications. Illustrative examples, though largely preliminary in nature, in solving a few system identification problems, are encouraging.

In this paper, a stochastic optimal control problem is formulated and solved for an investment and consumption model that includes stocks and bonds with transactions costs. In contrast to earlier results which considered linear transaction rate and got a non-singular feedback controls, we propose to use a quadratic transaction rate function to take into account of the liquidity of the bond and stock. The Taylor expansion is utilized to obtain an important initial condition in order to solve the nonlinear differential HJB equation numerically. The simulation studies are also carried out to quantify the effect of the transactions costs on the optimal investment-consumption policies

Stock market prediction has always been, in the past and at present, an intriguing issue. In this paper, an attempt is made at predicting the Standard & Poor's (S&P) 500 returns on a daily and weekly basis by using only historical price data. Two different types of prediction models are used for the prediction task: the auto-regressive (AR) and the neural network (NN) models. These two models are used in four different prediction systems. The first two prediction systems consist of either an AR model or a NN model. The next two prediction systems represent the novelty of the approach used in this paper. A multiple-model approach is proposed, together with the use of a trend classification algorithm, to predict the S&P 500 returns. Three models (either AR or NN) are used in each of the systems, with each model used to represent one of the three market trends (bear, choppy and bull). A decision rule is used to select one prediction from the three models, and one of two trading rules is used to make trading decisions. Three experiments were carried out to select appropriate parameters for the three-model systems. Evaluation of the models based on ARR after commission showed that the system consisting of three NNs was able to obtain approximately two times as much return as the buy-and-hold strategy in the test period when used in weekly predictions. Furthermore, the results in this paper show that non-linear systems performed better than linear ones, and three-model systems performed better than single-model ones

Extracting proper features is crucial to the performance of a pattern recognition system. Popular feature extraction techniques like principal component analysis (PCA), Fisher linear discriminant analysis (FLD), and independent component analysis (ICA) extract features that are not directly related to the classification accuracy. In this paper, we propose a new linear discriminant analysis algorithm (LDA) whose criterion function is based on the probability of classification error. The efficiency of this novel algorithm is demonstrated by application to face recognition problems.

Fisher linear discriminant (FLD) has recently emerged as a more efficient approach for extracting features for many pattern classification problems as compared to traditional principal component analysis. However, the constraint on the total number of features available from FLD has seriously limited its application to a large class of problems. In order to overcome this disadvantage, a recursive procedure of calculating the discriminant features is suggested in this paper. The new algorithm incorporates the same fundamental idea behind FLD of seeking the projection that best separates the data corresponding to different classes, while in contrast to FLD the number of features that may be derived is independent of the number of the classes to be recognized. Extensive experiments of comparing the new algorithm with the traditional approaches have been carried out on face recognition problem with the Yale database, in which the resulting improvement of the performances by the new feature extraction scheme is significant.

A novel recursive procedure for extracting discriminant features, termed recursive cluster-based linear discriminant (RCLD), is proposed in this paper. Compared to the traditional Fisher linear discriminant (FLD) and its variations, RCLD has a number of advantages. First of all, it relaxes the constraint on the total number of features that can be extracted. Second, it fully exploits all information available for discrimination. In addition, RCLD is able to cope with multimodal distributions, which overcomes an inherent problem of conventional FLDs, which assumes uni-modal class distributions. Extensive experiments have been carried out on various types of face recognition problems for Yale, Olivetti Research Laboratory, and JAFFE databases to evaluate and compare the performance of the proposed algorithm with other feature extraction methods. The resulting improvement of performances by the new feature extraction scheme is significant.

In this letter, we attempt to quantify the significance of increasing the number of neurons in the hidden layer of a feedforward neural network architecture using the singular value decomposition (SVD). Through this, we extend some well-known properties of the SVD in evaluating the generalizability of single hidden layer feedforward networks (SLFNs) with respect to the number of hidden layer neurons. The generalization capability of the SLFN is measured by the degree of linear independency of the patterns in hidden layer space, which can be indirectly quantified from the singular values obtained from the SVD, in a postlearning step. A pruning/growing technique based on these singular values is then used to estimate the necessary number of neurons in the hidden layer. More importantly, we describe in detail properties of the SVD in determining the structure of a neural network particularly with respect to the robustness of the selected model.

In this paper, a novel constant quality rate control (CQRC) algorithm is proposed for object-based (MPEG-4) video codecs. Instead of minimizing distortion of every single frame or minimizing average frame distortion, this controller seeks to minimize the variation of the frame distortion to achieve consistent good quality for whole video sequences. The CQRC algorithm uses a linear rate control model to estimate frame-level bit allocation based on a target distortion measure, and a quadratic rate-quantization model to calculate the quantization parameter for the current frame. The scheme is then further extended to encompass multiple (arbitrary shaped) video objects by means of a bitrate distribution algorithm. Experimental results demonstrate the ability of the CQRC algorithm to achieve a video sequence with less flickering effects and motion jerkiness compared to MPEG-4's scalable rate control scheme, even in the scenario of using imperfect segmentation masks

Two new recursive procedures for extracting discriminant features, termed recursive modified linear discriminant (RMLD) and recursive cluster-based linear discriminant (RCLD) are proposed in this paper. The two new methods, RMLD and RCLD overcome two major shortcomings of fisher linear discriminant (FLD): it can fully exploit all information available for discrimination; and it removes the constraint on the total number of features that can be extracted. Experiments of comparing the new algorithm with the traditional FLD and some of its variations have been carried out on various types of face recognition problems for Yale database, in which the resulting improvement of the performances by the new feature extraction scheme is significant

A geometrical interpretation of the multilayer perceptron (MLP) is suggested in this paper. Some general guidelines for selecting the architecture of the MLP, i.e., the number of the hidden neurons and the hidden layers, are proposed based upon this interpretation and the controversial issue of whether four-layered MLP is superior to the three-layered MLP is also carefully examined.

A decentralized neuro-controller with feedback error learning is proposed in this paper to deal with robot manipulator tracking problem. The PD + nonlinear (NL) feedback law + robustifying signal ensure global stability while the neural networks are utilized to compensate the decentralized nonlinear terms in the robot manipulator dynamics so that both robustness and good tracking performance are achieved. In addition to the theoretical proof of global stability, the effectiveness of the proposed scheme is also demonstrated by comparing the tracking performance of the neuro-controller for a two-link robot manipulator with that of the conventional decentralized adaptive controller.

The constraint on the total number of features available from the Fisher linear discriminant (FLD) has seriously limited its application to a large class of problems. In order to overcome this disadvantage of FLD, a recursive procedure for calculating the discriminant features is suggested in this paper. Extensive experiments of comparing the new algorithm with the traditional PCA and FLD approaches have been carried out on a face recognition problem, in which the resulting improvement of the performance by the new feature extraction scheme is significant.