Catastrophic forgetting remains a central challenge in lifelong learning, where newly acquired knowledge interferes with previously learned tasks, degrading performance over time. Mitigation strategies such as rehearsal and regularization have been proposed, but both introduce limitations, either by retaining old data or by constraining model updates in ways that may impair learning. Complicating matters, recent findings show that feature-space overlap between tasks can produce similar performance drops even in models that memorize data, making it difficult to distinguish true forgetting from representational interference. Current accuracy-based metrics fail to disentangle these effects, undermining diagnostic clarity. In this work, we introduce the Overlap Index, an incremental cluster validity index adapted from the inter-cluster component of the iCONN index, which quantifies overlap between feature representations in input or latent space. We then introduce the Overshadowing and Forgetting Index, an online meta-metric that leverages the Overlap Index to attribute performance degradation to catastrophic forgetting, class overshadowing, or both. Our experimental results demonstrate that these tools enable more precise online and batch-mode evaluation of continual learning systems, paving the way for more targeted mitigation strategies.

The accelerating impact of AI in biomedical research is driving significant advances in precision medicine. As these systems increasingly shape health outcomes, the imperative to develop trustworthy, reliable, and ethically grounded AI becomes more pressing, particularly in addressing concerns related to data integrity, patient safety, and equitable outcomes. While the potential of AI to transform biomedical research is clear, its responsible integration depends on more than technological capability. Ensuring that these systems are aligned with societal values requires a dual commitment: the operationalization of ethical principles throughout the AI life cycle and the establishment of robust regulatory mechanisms. Ethics provides the normative vision for fairness, accountability, and human dignity, whereas regulation translates these ideals into enforceable standards. This paper explores the convergence of these domains as a necessary foundation for developing trustworthy human-centered AI in biomedical contexts. We provide practical guidance for AI developers and researchers on integrating proactive governance and translating ethical principles into actionable strategies to support equitable and responsible innovation.

Infrared imaging has emerged as a robust solution for urban object detection under low-light and adverse weather conditions, offering significant advantages over traditional visible-light cameras. However, challenges such as class imbalance, thermal noise, and computational constraints can significantly hinder model performance in practical settings. To address these issues, we evaluate multiple YOLO variants on the FLIR ADAS V2 dataset, ultimately selecting YOLOv8 as our baseline due to its balanced accuracy and efficiency. Building on this foundation, we present MS-YOLO (MobileNetv4 and SlideLoss based on YOLO), which replaces YOLOv8’s CSPDarknet backbone with the more efficient MobileNetV4, reducing computational overhead by 1.5% while sustaining high accuracy. In addition, we introduce SlideLoss, a novel loss function that dynamically emphasizes under-represented and occluded samples, boosting precision without sacrificing recall. Experiments on the FLIR ADAS V2 benchmark show that MS-YOLO attains competitive mAP and superior precision while operating at only 6.7 GFLOPs. These results demonstrate that MS-YOLO effectively addresses the dual challenge of maintaining high detection quality while minimizing computational costs, making it well-suited for real-time edge deployment in urban environments.

This article addresses the stabilization challenges of nonholonomic systems under the threat of false data injection (FDI) attacks, which compromise the integrity of state information. A novel adaptive control strategy using Nussbaum-type gains is proposed to ensure the asymptotic stability of the closed-loop system while maintaining signal boundedness. The approach extends conventional Nussbaum designs to handle multiple unknown control directions. It integrates online learning mechanisms to mitigate the impact of FDI attacks. Additionally, adaptive backstepping and fuzzy-logic systems are utilized to approximate and compensate for unknown nonlinear dynamics. The methodology transforms nonholonomic systems into equivalent cascade structures to address inherent constraints and enable secure control input design. Simulation studies validate the effectiveness and resilience of the proposed control strategy, demonstrating significant improvements in stability and robustness in the presence of FDI attacks.

The transferability of adversarial examples is of central importance to transfer-based black-box adversarial attacks. Previous works for generating transferable adversarial examples focus on attacking given pretrained surrogate models while the connections between surrogate models and adversarial trasferability have been overlooked. In this paper, we propose Lipschitz Regularized Surrogate (LRS) for transfer-based black-box attacks, a novel approach that transforms surrogate models towards favorable adversarial transferability. Using such transformed surrogate models, any existing transfer-based black-box attack can run without any change, yet achieving much better performance. Specifically, we impose Lipschitz regularization on the loss landscape of surrogate models to enable a smoother and more controlled optimization process for generating more transferable adversarial examples. In addition, this paper also sheds light on the connection between the inner properties of surrogate models and adversarial transferability, where three factors are identified: smaller local Lipschitz constant, smoother loss landscape, and stronger adversarial robustness. We evaluate our proposed LRS approach by attacking state-of-the-art standard deep neural networks and defense models. The results demonstrate significant improvement on the attack success rates and transferability. Our code is available at https://github.com/TrustAIoT/LRS.

The capacity to generalize to future unseen data stands as one of the utmost crucial attributes of deep neural networks. Sharpness-Aware Minimization (SAM) aims to enhance the generalizability by minimizing worst-case loss using one-step gradient ascent as an approximation. However, as training progresses, the non-linearity of the loss landscape increases, rendering one-step gradient ascent less effective. On the other hand, multi-step gradient ascent will incur higher training cost. In this paper, we introduce a normalized Hessian trace to accurately measure the curvature of loss landscape on both training and test sets. In particular, to counter excessive non-linearity of loss landscape, we propose Curvature Regularized SAM (CR-SAM), integrating the normalized Hessian trace as a SAM regularizer. Additionally, we present an efficient way to compute the trace via finite differences with parallelism. Our theoretical analysis based on PAC-Bayes bounds establishes the regularizer's efficacy in reducing generalization error. Empirical evaluation on CIFAR and ImageNet datasets shows that CR-SAM consistently enhances classification performance for ResNet and Vision Transformer (ViT) models across various datasets. Our code is available at https://github.com/TrustAIoT/CR-SAM.

This article presents a novel efficient experience-replay-based adaptive dynamic programming (ADP) for the optimal control problem of a class of nonlinear dynamical systems within the Hamiltonian-driven framework. The quasi-Hamiltonian is presented for the policy evaluation problem with an admissible policy. With the quasi-Hamiltonian, a novel composite critic learning mechanism is developed to combine the instantaneous data with the historical data. In addition, the pseudo-Hamiltonian is defined to deal with the performance optimization problem. Based on the pseudo-Hamiltonian, the conventional Hamilton-Jacobi-Bellman (HJB) equation can be represented in a filtered form, which can be implemented online. Theoretical analysis is investigated in terms of the convergence of the adaptive critic design and the stability of the closed-loop systems, where parameter convergence can be achieved under a weakened excitation condition. Simulation studies are investigated to verify the efficacy of the presented design scheme.

Adversarial examples (AE) with good transferability enable practical black-box attacks on diverse target models, where insider knowledge about the target models is not required. Previous methods often generate AE with no or very limited transferability; that is, they easily overfit to the particular architecture and feature representation of the source, white-box model and the generated AE barely work for target, black-box models. In this paper, we propose a novel approach to enhance AE transferability using Gradient Norm Penalty (GNP). It drives the loss function optimization procedure to converge to a flat region of local optima in the loss landscape. By attacking 11 state-of-the-art (SOTA) deep learning models and 6 advanced defense methods, we empirically show that GNP is very effective in generating AE with high transferability. We also demonstrate that it is very flexible in that it can be easily integrated with other gradient based methods for stronger transfer-based attacks.

This article investigates the local stability and local convergence of a class of neural network (NN) controllers with error integrals as inputs for reference tracking. It is formally proved that if the input of the NN controller consists exclusively of error terms, the control system shows a non-zero steady-state error for any constant reference except for one specific point, for both single-layer and multi-layer NN controllers. It is further proved that adding error integrals to the input of the (single- and multi-layers) NN controller is one sufficient way to remove the steady-state error for any constant reference. Due to the nonlinearity of the NN controllers, the NN control systems are linearized at the equilibrium points. We provide proof that if all the eigenvalues of the linearized NN control system have negative real parts, local asymptotic stability and local exponential convergence are guaranteed. Two case studies were explored to verify the theoretical results: a single-layer NN controller in a 1-D system and a four-layer NN controller in a 2-D system applied to renewable energy integration. Simulations demonstrate that when NN controllers and the corresponding generalized proportional-integral (PI) controllers have the same eigenvalues, all control systems exhibit almost the same responses in a small neighborhood of their respective equilibrium points.

In streaming data applications, the incoming samples are processed and discarded, and therefore, intelligent decision-making is crucial for the performance of lifelong learning systems. In addition, the order in which the samples arrive may heavily affect the performance of incremental learners. The recently introduced incremental cluster validity indices (iCVIs) provide valuable aid in addressing such class of problems. Their primary use case has been cluster quality monitoring; nonetheless, they have been recently integrated in a streaming clustering method. In this context, the work presented, here, introduces the first adaptive resonance theory (ART)-based model that uses iCVIs for unsupervised and semi-supervised online learning. Moreover, it shows how to use iCVIs to regulate ART vigilance via an iCVI-based match tracking mechanism. The model achieves improved accuracy and robustness to ordering effects by integrating an online iCVI module as module B of a topological ART predictive mapping (TopoARTMAP)-thereby being named iCVI-TopoARTMAP-and using iCVI-driven postprocessing heuristics at the end of each learning step. The online iCVI module provides assignments of input samples to clusters at each iteration in accordance to any of the several iCVIs. The iCVI-TopoARTMAP maintains useful properties shared by the ART predictive mapping (ARTMAP) models, such as stability, immunity to catastrophic forgetting, and the many-to-one mapping capability via the map field module. The performance and robustness to the presentation order of iCVI-TopoARTMAP were evaluated via experiments with synthetic and real-world datasets.

This article presents an adaptive resonance theory predictive mapping (ARTMAP) model, which uses incremental cluster validity indices (iCVIs) to perform unsupervised learning, namely, iCVI-ARTMAP. Incorporating iCVIs to the decision-making and many-to-one mapping capabilities of this adaptive resonance theory (ART)-based model can improve the choices of clusters to which samples are incrementally assigned. These improvements are accomplished by intelligently performing the operations of swapping sample assignments between clusters, splitting and merging clusters, and caching the values of variables when iCVI values need to be recomputed. Using recursive formulations enables iCVI-ARTMAP to considerably reduce the computational burden associated with cluster validity index (CVI)-based offline clustering. In this work, six iCVI-ARTMAP variants were realized via the integration of one information-theoretic and five sum-of-squares-based iCVIs into fuzzy ARTMAP. With proper choice of iCVI, iCVI-ARTMAP either outperformed or performed comparably to three ART-based and four non-ART-based clustering algorithms in experiments using benchmark datasets of different natures. Naturally, the performance of iCVI-ARTMAP is subject to the selected iCVI and its suitability to the data at hand; fortunately, it is a general model in which other iCVIs can be easily embedded.

Concussions, also known as mild traumatic brain injury (mTBI), are a growing health challenge. Approximately four million concussions are diagnosed annually in the United States. Concussion is a heterogeneous disorder in causation, symptoms, and outcome making precision medicine approaches to this disorder important. Persistent disabling symptoms sometimes delay recovery in a difficult to predict subset of mTBI patients. Despite abundant data, clinicians need better tools to assess and predict recovery. Data-driven decision support holds promise for accurate clinical prediction tools for mTBI due to its ability to identify hidden correlations in complex datasets. We apply a Locality-Sensitive Hashing model enhanced by varied statistical methods to cluster blood biomarker level trajectories acquired over multiple time points. Additional features derived from demographics, injury context, neurocognitive assessment, and postural stability assessment are extracted using an autoencoder to augment the model. The data, obtained from FITBIR, consisted of 301 concussed subjects (athletes and cadets). Clustering identified 11 different biomarker trajectories. Two of the trajectories (rising GFAP and rising NF-L) were associated with a greater risk of loss of consciousness or post-traumatic amnesia at onset. The ability to cluster blood biomarker trajectories enhances the possibilities for precision medicine approaches to mTBI.

In this article, we consider an iterative adaptive dynamic programming (ADP) algorithm within the Hamiltonian-driven framework to solve the Hamilton-Jacobi-Bellman (HJB) equation for the infinite-horizon optimal control problem in continuous time for nonlinear systems. First, a novel function, "min-Hamiltonian," is defined to capture the fundamental properties of the classical Hamiltonian. It is shown that both the HJB equation and the policy iteration (PI) algorithm can be formulated in terms of the min-Hamiltonian within the Hamiltonian-driven framework. Moreover, we develop an iterative ADP algorithm that takes into consideration the approximation errors during the policy evaluation step. We then derive a sufficient condition on the iterative value gradient to guarantee closed-loop stability of the equilibrium point as well as convergence to the optimal value. A model-free extension based on an off-policy reinforcement learning (RL) technique is also provided. Finally, numerical results illustrate the efficacy of the proposed framework.

In this article, a data-driven distributed control method is proposed to solve the cooperative optimal output regulation problem of leader-follower multiagent systems. Different from traditional studies on cooperative output regulation, a distributed adaptive internal model is originally developed, which includes a distributed internal model and a distributed observer to estimate the leader's dynamics. Without relying on the dynamics of multiagent systems, we have proposed two reinforcement learning algorithms, policy iteration and value iteration, to learn the optimal controller through online input and state data, and estimated values of the leader's state. By combining these methods, we have established a basis for connecting data-distributed control methods with adaptive dynamic programming approaches in general since these are the theoretical foundation from which they are built.

This article investigates hardware implementation of hierarchical temporal memory (HTM), a brain-inspired machine learning algorithm that mimics the key functions of the neocortex and is applicable to many machine learning tasks. Spatial pooler (SP) is one of the main parts of HTM, designed to learn the spatial information and obtain the sparse distributed representations (SDRs) of input patterns. The other part is temporal memory (TM) which aims to learn the temporal information of inputs. The memristor, which is an appropriate synapse emulator for neuromorphic systems, can be used as the synapse in SP and TM circuits. In this article, a memristor-based SP (MSP) circuit structure is designed to accelerate the execution of the SP algorithm. The presented MSP has properties of modeling both the synaptic permanence and the synaptic connection state within a single synapse, and on-device and parallel learning. Simulation results of statistic metrics and classification tasks on several real-world datasets substantiate the validity of MSP.

This paper introduces inverse ontology cogency, a concept recognition process and distance function that is biologically-inspired and competitive with alternative methods. The paper introduces inverse ontology cogency as a new alternative method. It is a novel distance measure used in selecting the optimum mapping between ontology-specified concepts and phrases in free-form text. We also apply a multi-layer perceptron and text processing method for named entity recognition as an alternative to recurrent neural network methods. Automated named entity recognition, or concept recognition, is a common task in natural language processing. Similarities between confabulation theory and existing language models are discussed. This paper provides comparisons to MetaMap from the National Library of Medicine (NLM), a popular tool used in medicine to map free-form text to concepts in a medical ontology. The NLM provides a manually annotated database from the medical literature with concepts labeled, a unique, valuable source of ground truth, permitting comparison with MetaMap performance. Comparisons for different feature set combinations are made to demonstrate the effectiveness of inverse ontology cogency for entity recognition. Results indicate that using both inverse ontology cogency and corpora cogency improved concept recognition precision 20% over the best published MetaMap results. This demonstrates a new, effective approach for identifying medical concepts in text. This is the first time cogency has been explicitly invoked for reasoning with ontologies, and the first time it has been used on medical literature where high-quality ground truth is available for quality assessment.

This article presents a model-based hybrid adaptive dynamic programming (ADP) framework consisting of continuous feedback-based policy evaluation and policy improvement steps as well as an intermittent policy implementation procedure. This results in an intermittent ADP with a quantifiable performance and guaranteed closed-loop stability of the equilibrium point. To investigate the effect of aperiodic sampling on the communication bandwidth and the control performance of the intermittent ADP algorithms, we use a Hamiltonian-driven unified framework. With such a framework, it is shown that there is a tradeoff between the communication burden and the control performance. We finally show that the developed policies exhibit Zeno-free behaviors. Simulation examples show the efficiency of the proposed framework along with quantifiable comparisons of the policies with different intermittent information.

Biclustering is a special case of subspace clustering that has become viable in several domains. Particularly, in genomic data analysis, biclustering has been used to identify conditions under which a subset of genes are highly co-expressed, while topological data analysis has been used to analyze disease-specific subgroups, evolution, and disease progression. In this work, we combine biclustering with topological data analysis to achieve the best of both methods. We present TopoBARTMAP - produced by hybridizing BARTMAP, an adaptive resonance theory (ART)-based biclustering method, with TopoART, a topology learning ART network - in order to identify topological associations between biclusters. TopoBARTMAP outperformed both TopoART and BARTMAP in the experimental analysis on six benchmark blood cancer data sets. In some cases, BARTMAP may nevertheless be preferred due to implementation simplicity.

This paper focuses on current control in a permanent-magnet synchronous motor (PMSM). This paper has two main objectives: the first objective is to develop a neural-network (NN) vector controller to overcome the decoupling inaccuracy problem associated with the conventional proportional-integral-based vector-control methods. The NN is developed using the full dynamic equation of a PMSM, and trained to implement optimal control based on approximate dynamic programming. The second objective is to evaluate the robust and adaptive performance of the NN controller against that of the conventional standard vector controller under motor parameter variation and dynamic control conditions by: 1) simulating the behavior of a PMSM typically used in realistic electric vehicle applications and 2) building an experimental system for hardware validation as well as combined hardware and simulation evaluation. The results demonstrate that the NN controller outperforms conventional vector controllers in both simulation and hardware implementation.

This article develops a novel distributed intermittent control framework with the ultimate goal of reducing the communication burden in containment control of multiagent systems communicating via a directed graph. Agents are assumed to be under disturbance and communicate on a directed graph. Both static and dynamic intermittent protocols are proposed. Intermittent H∞containment control design is considered to attenuate the effect of the disturbance and the game algebraic Riccati equation (GARE) is employed to design the coupling and feedback gains for both static and dynamic intermittent feedback. A novel scheme is then used to unify continuous, static, and dynamic intermittent containment protocols. Finally, simulation results verify the efficacy of the proposed approach.

In recent years, a gradient of the n-step temporal-difference [TD(λ)] learning has been developed to present an advanced adaptive dynamic programming (ADP) algorithm, called value-gradient learning [VGL(λ)]. In this paper, we improve the VGL(λ) architecture, which is called the “single adaptive actor network [SNVGL(λ)]” because it has only a single approximator function network (critic) instead of dual networks (critic and actor) as in VGL(λ). Therefore, SNVGL(λ) has lower computational requirements when compared to VGL(λ). Moreover, in this paper, a recurrent hybrid neuro-fuzzy (RNF) and a first-order Takagi-Sugeno RNF (TSRNF) are derived and implemented to build the critic and actor networks. Furthermore, we develop the novel study of the theoretical convergence proofs for both VGL(λ) and SNVGL(λ) under certain conditions. In this paper, mobile robot simulation model (model based) is used to solve the optimal control problem for affine nonlinear discrete-time systems. Mobile robot is exposed various noise levels to verify the performance and to validate the theoretical analysis.

In problems with complex dynamics and challenging state spaces, the dual heuristic programming (DHP) algorithm has been shown theoretically and experimentally to perform well. This was recently extended by an approach called value gradient learning (VGL). VGL was inspired by a version of temporal difference (TD) learning that uses eligibility traces. The eligibility traces create an exponential decay of older observations with a decay parameter ( λ ). This approach is known as TD( λ ), and its DHP extension is known as VGL( λ ), where VGL(0) is identical to DHP. VGL has presented convergence and other desirable properties, but it is primarily useful for batch learning. Online learning requires an eligibility-trace-work-space matrix, which is not required for the batch learning version of VGL. Since online learning is desirable for many applications, it is important to remove this computational and memory impediment. This paper introduces a dual-critic version of VGL, called N -step VGL (NSVGL), that does not need the eligibility-trace-work-space matrix, thereby allowing online learning. Furthermore, this combination of critic networks allows an NSVGL algorithm to learn faster. The first critic is similar to DHP, which is adapted based on TD(0) learning, while the second critic is adapted based on a gradient of n -step TD( λ ) learning. Both networks are combined to train an actor network. The combination of feedback signals from both critic networks provides an optimal decision faster than traditional adaptive dynamic programming (ADP) via mixing current information and event history. Convergence proofs are provided. Gradients of one- and n -step value functions are monotonically nondecreasing and converge to the optimum. Two simulation case studies are presented for NSVGL to show their superior performance.

Because of a powerful temporal-difference (TD) with λ [TD( λ )] learning method, this paper presents a novel n -step adaptive dynamic programming (ADP) architecture that combines TD( λ ) with regular TD learning for solving optimal control problems with reduced iterations. In contrast with a backward view learning of TD( λ ) that is required an extra parameter named eligibility traces to update at the end of each episode (offline training), the new design in this paper has forward view learning, which is updated at each time step (online training) without needing the eligibility trace parameter in various applications without mathematical models. Therefore, the new design is called the online model-free n -step action-dependent (AD) heuristic dynamic programming [NSHDP( λ )]. NSHDP( λ ) has three neural networks: the critic network (CN) with regular one-step TD [TD(0)], the CN with n -step TD learning [or TD( λ )], and the actor network (AN). Because the forward view learning does not require any extra eligibility traces associated with each state, the NSHDP( λ ) architecture has low computational costs and is memory efficient. Furthermore, the stability is proven for NSHDP( λ ) under certain conditions by using Lyapunov analysis to obtain the uniformly ultimately bounded (UUB) property. We compare the results with the performance of HDP and traditional action-dependent HDP( λ ) [ADHDP( λ )] with different λ values. Moreover, a complex nonlinear system and 2-D maze problem are two simulation benchmarks in this paper, and the third one is an inverted pendulum simulation benchmark, which is presented in the supplemental material part of this paper. NSHDP( λ ) performance is examined and compared with other ADP methods.

In this article, we present an intermittent framework for safe reinforcement learning (RL) algorithms. First, we develop a barrier function-based system transformation to impose state constraints while converting the original problem to an unconstrained optimization problem. Second, based on optimal derived policies, two types of intermittent feedback RL algorithms are presented, namely, a static and a dynamic one. We finally leverage an actor/critic structure to solve the problem online while guaranteeing optimality, stability, and safety. Simulation results show the efficacy of the proposed approach.

A continuous-time K -winners-take-all (KWTA) neural model that can identify the largest K of N inputs, where command signal is described. The model is given by a differential equation where the spike train is a sum of delta functions. A functional block-diagram of the model includes N feed-forward hard-limiting neurons and one feedback neuron, used to handle input dynamics. The existence and uniqueness of the model steady states are analyzed, the convergence analysis of the state variable trajectories to the KWTA operation is proven, the convergence time and number of spikes required are derived, as well as the processing of time-varying inputs and perturbations of the model nonlinearities are analyzed. The main advantage of the model is that it is not subject to the intrinsic convergence of speed limitations of comparable designs. The model also has an arbitrary finite resolution determined by a given parameter, low complexity, and initial condition independence. Applications of the model for parallel sorting and parallel rank-order filtering are presented. Theoretical results are derived and illustrated with computer-simulated examples that demonstrate the model's performance.

This paper presents an improved method for reducing high-order dynamical system models via clustering. Agglomerative hierarchical clustering based on performance evaluation (HC-PE) is introduced for model order reduction. This method computes the reduced order denominator of the transfer function model by clustering system poles in a hierarchical dendrogram. The base layer represents an nth order system, which is used to calculate each successive layer to reduce the model order until finally reaching a second-order system. HC-PE uses a mean-squared error (MSE) in every reduced order, which modifies the pole placement process. The coefficients for the numerator of the reduced model are calculated by using the Padé approximation (PA) or alternatively a genetic algorithm (GA). Several numerical examples of reducing techniques are taken from the literature to compare with HC-PE. Two classes of results are shown in this paper. The first sets are single-input single-output models that range from simple models to 48th order systems. The second sets of experiments are with a multi-input multioutput model. We demonstrate the best performance for HC-PE through minimum MSEs compared with other methods. Furthermore, the robustness of HC-PE combined with PA or GA is confirmed by evaluating the third-order reduced model for the triple-link inverted pendulum model by adding a disturbance impulse signal and by changing model parameters. The relevant stability proofs are provided in Appendixes A and B in the supplementary material. HC-PE with PA slightly outperforms its performance with GA, but both approaches are attractive alternatives to other published methods.

This paper provides the stability analysis for a model-free action-dependent heuristic dynamic programing (HDP) approach with an eligibility trace long-term prediction parameter ( λ ). HDP( λ ) learns from more than one future reward. Eligibility traces have long been popular in Q-learning. This paper proves and demonstrates that they are worthwhile to use with HDP. In this paper, we prove its uniformly ultimately bounded (UUB) property under certain conditions. Previous works present a UUB proof for traditional HDP [HDP( λ = 0 )], but we extend the proof with the λ parameter. By using Lyapunov stability, we demonstrate the boundedness of the estimated error for the critic and actor neural networks as well as learning rate parameters. Three case studies demonstrate the effectiveness of HDP( λ ). The trajectories of the internal reinforcement signal nonlinear system are considered as the first case. We compare the results with the performance of HDP and traditional temporal difference [TD( λ )] with different λ values. The second case study is a single-link inverted pendulum. We investigate the performance of the inverted pendulum by comparing HDP( λ ) with regular HDP, with different levels of noise. The third case study is a 3-D maze navigation benchmark, which is compared with state action reward state action, Q( λ ), HDP, and HDP( λ ). All these simulation results illustrate that HDP( λ ) has a competitive performance; thus this contribution is not only UUB but also useful in comparison with traditional HDP.

This paper presents a model-free solution to the robust stabilization problem of discrete-time linear dynamical systems with bounded and mismatched uncertainty. An optimal controller design method is derived to solve the robust control problem, which results in solving an algebraic Riccati equation (ARE). It is shown that the optimal controller obtained by solving the ARE can robustly stabilize the uncertain system. To develop a model-free solution to the translated ARE, off-policy reinforcement learning (RL) is employed to solve the problem in hand without the requirement of system dynamics. In addition, the comparisons between on- and off-policy RL methods are presented regarding the robustness to probing noise and the dependence on system dynamics. Finally, a simulation example is carried out to validate the efficacy of the presented off-policy RL approach.

One of the leading approaches to collaborative filtering is to use matrix factorization to discover a set of latent factors that explain the pattern of preferences. In this paper, we apply a resilient stochastic gradient descent approach that uses only the sign of the gradient, similar to the R-Prop algorithm in neural network training, to matrix factorization for collaborative filtering. We evaluate the performance of our approach on the MovieLens 1M dataset, and find that test set accuracy markedly improves compared to standard gradient descent. As a follow-up experiment, we apply clustering to the learned item-factor matrix in factor space, and attempt to manually characterize each cluster of movies.

The clustering structures formed by Adaptive Resonance Theory (ART) and many other algorithms are dependent on input presentation/permutation order. In this work, we exploit Visual Assessment of cluster Tendency (VAT) as a pre-processor for Fuzzy ART in order to mitigate this problem. This approach is a global strategy that uses similarity-based ordering before clustering. Experimental results show that this framework improved peak and average performance, reduced the number of categories, and incurred less variability in the clustering outcome. By enhancing performance and reducing sensitivity to input order presentation, this approach is recommended when it is suitable to perform off-line incremental learning.

This paper presents a data-driven method based on off-policy integral reinforcement learning to solve the semi-global output regulation of continuous-time linear systems with input saturation. A family of state feedback laws for the input constrained output regulation problem is designed based on solving an algebraic Riccati equation. In contrast to the existing methods, complete knowledge of the system dynamics is no longer required in this paper. Instead, the data collected from online implementation is efficiently utilized to design the controller. Therefore, the controller design in this paper is data-driven. It is shown that the presented method can find feedback control inputs with constraint of amplitude saturation and stabilize a given linear system with all its poles inside or on the imaginary axis. Finally, a simulation example is conducted to show the validity of the presented approach to solve the semi-global output regulation of continuous-time linear systems with input saturation.

Improved data visualization will be a significant tool to enhance cluster analysis. In this paper, an information-theoretic-based method for cluster visualization using self-organizing maps (SOMs) is presented. The information-theoretic visualization (IT-vis) has the same structure as the unified distance matrix, but instead of depicting Euclidean distances between adjacent neurons, it displays the similarity between the distributions associated with adjacent neurons. Each SOM neuron has an associated subset of the data set whose cardinality controls the granularity of the IT-vis and with which the first- and second-order statistics are computed and used to estimate their probability density functions. These are used to calculate the similarity measure, based on Renyi's quadratic cross entropy and cross information potential (CIP). The introduced visualizations combine the low computational cost and kernel estimation properties of the representative CIP and the data structure representation of a single-linkage-based grouping algorithm to generate an enhanced SOM-based visualization. The visual quality of the IT-vis is assessed by comparing it with other visualization methods for several real-world and synthetic benchmark data sets. Thus, this paper also contains a significant literature survey. The experiments demonstrate the IT-vis cluster revealing capabilities, in which cluster boundaries are sharply captured. Additionally, the information-theoretic visualizations are used to perform clustering of the SOM. Compared with other methods, IT-vis of large SOMs yielded the best results in this paper, for which the quality of the final partitions was evaluated using external validity indices.

This paper develops optimal control protocols for the distributed output synchronization problem of leader-follower multiagent systems with an active leader. Agents are assumed to be heterogeneous with different dynamics and dimensions. The desired trajectory is assumed to be preplanned and is generated by the leader. Other follower agents autonomously synchronize to the leader by interacting with each other using a communication network. The leader is assumed to be active in the sense that it has a nonzero control input so that it can act independently and update its control to keep the followers away from possible danger. A distributed observer is first designed to estimate the leader's state and generate the reference signal for each follower. Then, the output synchronization of leader-follower systems with an active leader is formulated as a distributed optimal tracking problem, and inhomogeneous algebraic Riccati equations (AREs) are derived to solve it. The resulting distributed optimal control protocols not only minimize the steady-state error but also optimize the transient response of the agents. An off-policy reinforcement learning algorithm is developed to solve the inhomogeneous AREs online in real time and without requiring any knowledge of the agents' dynamics. Finally, two simulation examples are conducted to illustrate the effectiveness of the proposed algorithm.

This work examines the validity of facial phenotypes as Autism Spectrum Disorders (ASD) biomarkers in boys with essential autism. A family-based association analysis framework is presented that uses previously identified facially-delineated (FD) clusters to examine relationship between FD clusters and known ASD genes. The hypothesis is that there are certain genetic variants, single nucleotide polymorphisms (SNP), specific to the FD clusters. Although statistical significance was not established, the results identified some candidate SNPs unique to each of the FD clusters that could indicate an underlying etiological difference. Further, recommendations are provided for larger-scale studies that could utilize the analysis framework presented.

This paper uses value gradient learning (VGL) to track a reference trajectory under uncertainties, by computing the optimal left and right torque values for a nonholonomic mobile robot. VGL is a high-performance algorithm in adaptive dynamic programming (ADP). Here, it is used as a critic function after fitting a first-order Sugeno fuzzy neural network (FNN) structure to critic and actor networks. Moreover, this work handles the impacts of unmodeled bounded disturbances with various friction values. The simulation is introduced to compare two approaches. The first uses an actor network that confirms the ability of the mobile robot dynamic model to follow a desired trajectory. This approach demonstrates a significant enhancement of the robot's capability to absorb unstructured disturbance signals and friction effects. The second type of results use a critic-optimal-control approach, calculating the optimal control signal for the affine dynamic model of the robot. This completely removes the actor network to exploit reduced computational complexity with faster responses. The simulation is introduced to compare both cases.

In this paper, based on the Hamiltonian, an alternative interpretation about the iterative adaptive dynamic programming (ADP) approach from the perspective of optimization is developed for discrete time nonlinear dynamic systems. The role of the Hamiltonian in iterative ADP is explained. The resulting Hamiltonian driven ADP is able to evaluate the performance with respect to arbitrary admissible policies, compare two different admissible policies and further improve the given admissible policy. The convergence of the Hamiltonian ADP to the optimal policy is proven. Implementation of the Hamiltonian-driven ADP by neural networks is discussed based on the assumption that each iterative policy and value function can be updated exactly. Finally, a simulation is conducted to verify the effectiveness of the presented Hamiltonian-driven ADP.

In this paper, two techniques based on Q -learning and action dependent heuristic dynamic programming (ADHDP) are demonstrated for the demand-side management of domestic electric water heaters (DEWHs). The problem is modeled as a dynamic programming problem, with the state space defined by the temperature of output water, the instantaneous hot water consumption rate, and the estimated grid load. According to simulation, Q-learning and ADHDP reduce the cost of energy consumed by DEWHs by approximately 26% and 21%, respectively. The simulation results also indicate that these techniques will minimize the energy consumed during load peak periods. As a result, the customers saved about $466 and $367 annually by using Q-learning and ADHDP techniques to control their DEWHs (100 gallons tank size) operation, which is better than the cost reduction that resulted from using the state-of-the-art ($246) control technique under the same simulation parameters. To the best of the authors' knowledge, this is the first work that uses the approximate dynamic programming techniques to solve the DEWH's load management problem.

Ever-increasing size and complexity of data sets create challenges and potential tradeoffs of accuracy and speed in learning algorithms. This paper offers progress on both fronts. It presents a mechanism to train the unsupervised learning features learned from only one layer to improve performance in both speed and accuracy. The features are learned by an unsupervised feature learning (UFL) algorithm. Then, those features are trained by a fast radial basis function (RBF) extreme learning machine (ELM). By exploiting the massive parallel computing attribute of modern graphics processing unit, a customized compute unified device architecture (CUDA) kernel is developed to further speed up the computing of the RBF kernel in the ELM. Results tested on Canadian Institute for Advanced Research and Mixed National Institute of Standards and Technology data sets confirm the UFL RBF ELM achieves high accuracy, and the CUDA implementation is up to 20 times faster than CPU and the naive parallel approach.

This paper presents a Hamiltonian-driven framework of adaptive dynamic programming (ADP) for continuous time nonlinear systems, which consists of evaluation of an admissible control, comparison between two different admissible policies with respect to the corresponding the performance function, and the performance improvement of an admissible control. It is showed that the Hamiltonian can serve as the temporal difference for continuous-time systems. In the Hamiltonian-driven ADP, the critic network is trained to output the value gradient. Then, the inner product between the critic and the system dynamics produces the value derivative. Under some conditions, the minimization of the Hamiltonian functional is equivalent to the value function approximation. An iterative algorithm starting from an arbitrary admissible control is presented for the optimal control approximation with its convergence proof. The implementation is accomplished by a neural network approximation. Two simulation studies demonstrate the effectiveness of Hamiltonian-driven ADP.

With the exponential growth of data and complexity of systems, fast machine learning/artificial intelligence and computational intelligence techniques are highly required. Many conventional computational intelligence techniques face bottlenecks in learning (e.g., intensive human intervention and convergence time) [item 1) in the Appendix]. However, efficient learning algorithms alternatively offer significant benefits including fast learning speed, ease of implementation, and minimal human intervention. The need for efficient and fast implementation of machine learning techniques in big data and dynamic varying systems poses many research challenges. This special issue highlights some latest development in the related areas.

This paper examines a schema for graph-theoretic clustering using node-based resilience measures. Node-based resilience measures optimize an objective based on a critical set of nodes whose removal causes some severity of disconnection in the network. Beyond presenting a general framework for the usage of node based resilience measures for variations of clustering problems, we emphasize the unique potential of such methods to accomplish the following properties: (i) clustering a graph in one step without knowing the number of clusters a priori, and (ii) removing noise from noisy data. We first present results of clustering experiments using a β-parametrized generalization of vertex attack tolerance, showing high clustering accuracy for both real datasets and equal density synthetic data sets, as well as successful removal of noise nodes. It is shown that arbitrarily increasing β increases the number of noise nodes removed in some cases, and that internal validation measures can be used to determine the correct number of clusters in a class of datasets. Further results are presented using five different resilience measures with a general node-based resilience clustering technique. In a subset of cases a resilience measure, such as integrity, is able to cluster to high accuracy in one step, giving the correct clustering while also determining the correct number of clusters. Integrity is also shown to be promising with respect to noise removal, removing up to 80% of noise on some datasets.

This paper presents a direct heuristic dynamic programming (HDP) based on Dyna planning (Dyna_HDP) for online model learning in a Markov decision process. This novel technique is composed of HDP policy learning to construct the Dyna agent for speeding up the learning time. We evaluate Dyna_HDP on a differential-drive wheeled mobile robot navigation problem in a 2D maze. The simulation is introduced to compare Dyna_HDP with other traditional reinforcement learning algorithms, namely one step Q-learning, Sarsa (λ), and Dyna_Q, under the same benchmark conditions. We demonstrate that Dyna_HDP has a faster near-optimal path than other algorithms, with high stability. In addition, we also confirm that the Dyna_HDP method can be applied in a multi-robot path planning problem. The virtual common environment model is learned from sharing the robots' experiences which significantly reduces the learning time.

Collaborative filtering provides recommendations based on the behavior of each user combined with behavior of users with similar interests. Recommender systems are becoming widespread, helping people choose movies, books, and things to buy. In this study, we examine the use of Biclustering ARTMAP to build a collaborative filtering recommendation system. We introduce a novel modification to how the Biclustering ARTMAP algorithm computes the item-cluster similarity, and a way to adapt it for the prediction of user ratings. We apply the algorithm to the MovieLens 100k dataset, and find that it achieves promising performance compared to other collaborative filtering techniques.

In this paper, an information-theoretic-based adaptive resonance theory (IT-ART) neural network architecture is presented. Each IT-ART category is defined by the first and second order statistics (mean and covariance matrix) of the cluster or class it represents. This information is used to estimate probability density functions (multivariate Gaussians) and compute the activation functions. The match function of the vigilance check is based on Renyi's quadratic cross-entropy: it is the cross information potential. Experiments involving several real world and synthetic data sets were carried out to assess the performance of IT-ART, which was measured in terms of external validity indices. IT-ART expanded the range of successful vigilance parameter values in these tests.

The large scale and complex nature of social media data raises the need to scale clustering techniques to big data and make them capable of automatically identifying data clusters with few empirical settings. In this paper, we present our investigation and three algorithms based on the fuzzy adaptive resonance theory (Fuzzy ART) that have linear computational complexity, use a single parameter, i.e., the vigilance parameter to identify data clusters, and are robust to modest parameter settings. The contribution of this paper lies in two aspects. First, we theoretically demonstrate how complement coding, commonly known as a normalization method, changes the clustering mechanism of Fuzzy ART, and discover the vigilance region (VR) that essentially determines how a cluster in the Fuzzy ART system recognizes similar patterns in the feature space. The VR gives an intrinsic interpretation of the clustering mechanism and limitations of Fuzzy ART. Second, we introduce the idea of allowing different clusters in the Fuzzy ART system to have different vigilance levels in order to meet the diverse nature of the pattern distribution of social media data. To this end, we propose three vigilance adaptation methods, namely, the activation maximization (AM) rule, the confliction minimization (CM) rule, and the hybrid integration (HI) rule. With an initial vigilance value, the resulting clustering algorithms, namely, the AM-ART, CM-ART, and HI-ART, can automatically adapt the vigilance values of all clusters during the learning epochs in order to produce better cluster boundaries. Experiments on four social media data sets show that AM-ART, CM-ART, and HI-ART are more robust than Fuzzy ART to the initial vigilance value, and they usually achieve better or comparable performance and much faster speed than the state-of-the-art clustering algorithms that also do not require a predefined number of clusters.

Autism spectrum disorder (ASD) is characterized by notable phenotypic heterogeneity, which is often viewed as an obstacle to the study of its etiology, diagnosis, treatment, and prognosis. Heterogeneity in ASD is multidimensional and complex including variability in phenotype as well as clinical, physiologic, and pathologic parameters. We apply a hierarchical clustering model suited to dealing with datasets of mixed data types to stratify children with ASD into more homogeneous subgroups in line with the Diagnostic and Statistical Manual of Mental Disorders (DSM)-5 model. The results of this cluster analysis will provide a better understanding the complex issue of ASD phenotypic heterogeneity and identify subgroups useful for further ASD genetic studies. Our goal is to provide insight into viable phenotypic and genotypic markers that would guide further cluster analysis of ASD genetic data. We suggest that analyzing the clusters in a hierarchical structure is a well-suited and meaningful model to unravel the complex heterogeneity of this disorder.

In this paper, novel hierarchical clustering algorithms, Growing Fuzzy ART (GFA) and Self-Resonant Growing Fuzzy ART (SRGFA), based on connecting prototypes, are presented. The prototypes are generated by vector quantization algorithms: K-means, Self-Organizing Maps, and Fuzzy ART. The Euclidean distance is used to train the first two algorithms in order to allocate the centroids and neurons, respectively. The latter uses fuzzy set operations to check resonance and learn the categories. For each method, a subset of the data set is associated with each prototype; this subset consists of all patterns that, according to a similarity measure, are within a certain threshold from a given prototype. In the case of K-means and Self-Organizing Map, the region is a hypersphere, and in the case of Fuzzy ART, it is a hyperbox. In order to relax the similarity constraint and create larger subsets of data for each prototype, the values of the Euclidean norm and the vigilance parameter are continuously increased and decreased, respectively, according to a step size. Prototypes that have patterns in common are linked together in the process. The data set's final partition is selected as the clustering state in which the algorithm spent most of its time. Synthetic and real world data sets are used to depict the experimental results. External validity indices are used as figures of merit to evaluate the quality of the final partitions.

A Particle Swarm Optimization (PSO) technique, in conjunction with Fuzzy Adaptive Resonance Theory (ART), was implemented to adapt vigilance values to appropriately compensate for a disparity in data sparsity. Gaining the ability to optimize a vigilance threshold over each cluster as it is created is useful because not all conceivable clusters have the same sparsity from the cluster centroid. Instead of selecting a single vigilance threshold, a metric must be selected for the PSO to optimize on. This trades one design decision for another. The performance gain, however, motivates the tradeoff in certain applications.

Linear and nonlinear models for time series analysis and prediction are well-established. Clustering methods have also been applied to this area. This paper explores a framework that can be used to cluster time series data. The range of values of a time series is clustered. Then the time series is clustered by data windows that flow into the initial set of value clusters. This allows predictive temporal patterns to be discovered across the whole range of values.

This paper investigates how to train a recurrent neural network (RNN) using the Levenberg-Marquardt (LM) algorithm as well as how to implement optimal control of a grid-connected converter (GCC) using an RNN. To successfully and efficiently train an RNN using the LM algorithm, a new forward accumulation through time (FATT) algorithm is proposed to calculate the Jacobian matrix required by the LM algorithm. This paper explores how to incorporate FATT into the LM algorithm. The results show that the combination of the LM and FATT algorithms trains RNNs better than the conventional backpropagation through time algorithm. This paper presents an analytical study on the optimal control of GCCs, including theoretically ideal optimal and suboptimal controllers. To overcome the inapplicability of the optimal GCC controller under practical conditions, a new RNN controller with an improved input structure is proposed to approximate the ideal optimal controller. The performance of an ideal optimal controller and a well-trained RNN controller was compared in close to real-life power converter switching environments, demonstrating that the proposed RNN controller can achieve close to ideal optimal control performance even under low sampling rate conditions. The excellent performance of the proposed RNN controller under challenging and distorted system conditions further indicates the feasibility of using an RNN to approximate optimal control in practical applications.

This research demonstrates the use of genetic programming to derive the objective function that ranks the candidate concepts and selects the set of best matching concepts for a sentence within medical text. A short set of example primitive and linguistic variables was input into the GP process, and a set of manually tagged sentences extracted from the literature was used to derive different objective functions potentially suitable for tagging. This proof-of-concept demonstrates the potential of this approach to simplify automated semantic tagging and to identify some of the likely challenges of applying the GP approach to complex linguistics problems of this nature.

The EigenAnt Ant Colony System (EAAS) model is an Ant Colony Optimization (ACO) model based on the EigenAnt algorithm. In previous work, EAAS was found to perform competitively with the Enhanced Ant Colony System (EACS) algorithm, a state-of-the-art method for the Sequential Ordering Problem (SOP). In this paper, we extend EAAS by increasing the amount of stochasticity in its solution construction procedure. In experimental results on the SOPLIB instance library, we find that our proposed method, called Probabilistic EAAS (PEAAS), performs better than both EAAS and EACS. The non-parametric Friedman test is applied to determine statistical significance.

Three-phase grid-connected converters are widely used in renewable and electric power system applications. Traditionally, grid-connected converters are controlled with standard decoupled d-q vector control mechanisms. However, recent studies indicate that such mechanisms show limitations in their applicability to dynamic systems. This paper investigates how to mitigate such restrictions using a neural network to control a grid-connected rectifier/inverter. The neural network implements a dynamic programming algorithm and is trained by using back-propagation through time. To enhance performance and stability under disturbance, additional strategies are adopted, including the use of integrals of error signals to the network inputs and the introduction of grid disturbance voltage to the outputs of a well-trained network. The performance of the neural-network controller is studied under typical vector control conditions and compared against conventional vector control methods, which demonstrates that the neural vector control strategy proposed in this paper is effective. Even in dynamic and power converter switching environments, the neural vector controller shows strong ability to trace rapidly changing reference commands, tolerate system disturbances, and satisfy control requirements for a faulted power system.

Intelligent argumentation system facilitates stakeholders to exchange dialogue over issues and provides decision support by capturing rationale of the stakeholders through arguments. In argumentation process, stakeholders tend to polarize on their opinions and form polarization groups. A method [1] was developed earlier to identify polarization groups, however, polarization groups tend to overlap to a certain degree and each stakeholder may be a member of multiple polarization groups to varied degrees. Quantifying stakeholders' membership in multiple polarization groups in argumentation for collaborative decision making is not addressed earlier. We present an approach using fuzzy clustering algorithm to address this issue and evaluate the approach using an argumentation tree built by twenty four stakeholders.

The HONEST network is a high order neural network that uses product units and adaptable exponential weights. In this paper, we explore the use of several learning methods with the HONEST network: Levenberg-Marquardt (LM), Conjugate Gradient (CG), Scaled Conjugate Gradient (a technique that combines LM and CG), and resilient propagation (RP). Using a benchmark of 19 datasets, we find that the first three methods mentioned produce lower average test set errors than RP to a statistically significant extent.

This paper presents a new approach, which we call UFL-ELM, to classification using both unsupervised and supervised learning. Unlike traditional approaches in which features are extracted, hand-crafted, and then trained using time-consuming, iterated optimization, this proposed method leverages unsupervised feature learning to learn features from the data themselves and then train the classifier using an extreme learning machine to reach the analytic solution. The result is therefore widely and quickly applied to universal data. Experiments on a large dataset of images confirm the ease of use and speed of training of this unsupervised feature learning approach. Furthermore, the paper discusses how to speed up training, using massively parallel programming.

With the improvement of battery technology over the past two decades and automotive technology advances, more and more vehicle manufacturers have joined in the race to produce new generation of affordable, high-performance Electric Drive Vehicles (EDVs). Permanent Magnet Synchronous Motors (PMSMs) are at the top of AC motors in high performance drive systems for EDVs. Traditionally, a PMSM is controlled with standard decoupled d-q vector control mechanisms. However, recent studies indicate that such mechanisms show serious limitations. This paper investigates how to mitigate such problems using a nested-loop neural network architecture to control a PMSM. The neural network implements a dynamic programming algorithm and is trained using backpropagation through time. The performance of the neural controller is studied for typical vector control conditions and compared with conventional vector control methods, which demonstrates the neural vector control strategy proposed in this paper is effective. Even in a highly dynamic switching environment, the neural vector controller shows strong ability to track rapidly changing reference commands, tolerate system disturbances, and satisfy control requirements for complex EDV drive needs.

Despite the advantages of fast and stable learning, Adaptive Resonance Theory (ART) still relies on an empirically fixed vigilance parameter value to determine the vigilance regions of all of the clusters in the category field (F2), causing its performance to depend on the vigilance value. It would be desirable to use different values of vigilance for different category field nodes, in order to fit the data with a smaller number of categories. We therefore introduce two methods, the Activation Maximization Rule (AMR) and the Confliction Minimization Rule (CMR). Despite their differences, both ART with AMR (AM-ART) and with CMR (CM-ART) allow different vigilance levels for different clusters, which are incrementally adapted during the clustering process. Specifically, AMR works by increasing the vigilance value of the winner cluster when a resonance occurs and decreasing it when a reset occurs, which aims to maximize the participation of clusters for activation. On the other hand, after receiving an input pattern, CMR first identifies all of the winner candidates that satisfy the vigilance criteria and then tunes their vigilance values to minimize conflicts in the vigilance regions. In this paper, we chose Fuzzy ART to demonstrate these concepts, but they will clearly carry over to other ART architectures. Our comparative experiments show that both AM-ART and CM-ART improve the robust performance of Fuzzy ART to the vigilance parameter and usually produce better cluster quality.

A new compressive video sampling method is investigated. As opposed to other video sampling methods for which processing is conducted on a single-frame basis, this method is applied on multiple frames of the video stream. By exploiting the extension of Wavelet to 3-D with the support of Noiselet 3-D and combining it with fast reconstruction algorithms, this framework produces successful results quickly while maintaining the quality of the video stream. Despite its simplicty, this new approach outperforms other sophisticated methods.

A cellular simultaneous recurrent network (CSRN) [1–11] is a neural network architecture that uses conventional simultaneous recurrent networks (SRNs), or cells in a cellular structure. The cellular structure adds complexity, so the training of CSRNs is far more challenging than that of conventional SRNs. Computer Go serves as an excellent test bed for CSRNs because of its clear-cut objective. For the training data, we developed an accurate theoretical foundation and game tree for the 2×2 game board. The conventional CSRN architecture suffers from the multi-valued function problem; our modified CSRN architecture overcomes the problem by employing ternary coding of the Go board's representation and a normalized input dimension reduction. We demonstrate a 2×2 game tree trained with the proposed CSRN architecture and the proposed cellular particle swarm optimization.

Three-phase grid-connected converters are widely used in renewable and electric power system applications. Traditionally, grid-connected converters are controlled with standard decoupled d-q vector control mechanisms. However, recent studies indicate that such mechanisms show limitations. This paper investigates how to mitigate such problems using a neural network to control a grid-connected rectifier/inverter. The neural network implements a dynamic programming (DP) algorithm and is trained using backpropagation through time. The performance of the DP-based neural controller is studied for typical vector control conditions and compared with conventional vector control methods. The paper also investigates how varying grid and power converter system parameters may affect the performance and stability of the neural control system. Future research issues regarding the control of grid-connected converters using DP-based neural networks are analyzed.

Swarm intelligence has emerged as a worthwhile class of clustering methods due to its convenient implementation, parallel capability, ability to avoid local minima, and other advantages. In such applications, clustering validity indices usually operate as fitness functions to evaluate the qualities of the obtained clusters. However, as the validity indices are usually data dependent and are designed to address certain types of data, the selection of different indices as the fitness functions may critically affect cluster quality. Here, we compare the performances of eight well-known and widely used clustering validity indices, namely, the Caliński-Harabasz index, the CS index, the Davies-Bouldin index, the Dunn index with two of its generalized versions, the I index, and the silhouette statistic index, on both synthetic and real data sets in the framework of differential-evolution-particle-swarm-optimization (DEPSO)-based clustering. DEPSO is a hybrid evolutionary algorithm of the stochastic optimization approach (differential evolution) and the swarm intelligence method (particle swarm optimization) that further increases the search capability and achieves higher flexibility in exploring the problem space. According to the experimental results, we find that the silhouette statistic index stands out in most of the data sets that we examined. Meanwhile, we suggest that users reach their conclusions not just based on only one index, but after considering the results of several indices to achieve reliable clustering structures.

Hierarchical clustering is an important and powerful but computationally extensive operation. Its complexity motivates the exploration of highly parallel approaches such as Adaptive Resonance Theory (ART). Although ART has been implemented on GPU processors, this paper presents the first hierarchical ART GPU implementation we are aware of. Each ART layer is distributed in the GPU's multiprocessors and is trained simultaneously. The experimental results show that for deep trees, the GPU's performance advantage is significant.

The applications of recently developed meta-heuristics in cluster analysis, such as particle swarm optimization (PSO) and differential evolution (DE), have increasingly attracted attention and popularity in a wide variety of communities owing to their effectiveness in solving complicated combinatorial optimization problems. Here, we propose to use a hybrid of PSO and DE, known as differential evolution particle swarm optimization (DEPSO), in order to further improve search capability and achieve higher flexibility in exploring the natural while hidden data structures of data of interest. Empirical results show that the DEPSO-based clustering algorithm achieves better performance in terms of the number of epochs required to reach a pre-specified cutoff value of the fitness function than either of the other approaches used. Further experimental studies on both synthetic and real data sets demonstrate the effectiveness of the proposed method in finding meaningful clustering solutions.

Video sensor capabilities and sophistication has improved to the point that they are being utilized in vast and diverse applications. Many such applications are now on the verge of providing too much video information reducing the ability to review, categorize, and process the immense amounts of video. Advancement in other technology areas such as Global Positioning System (GPS) processors and single board computers have paved the way for a new development of smart video sensors. A need exists to be able to identify stationary objects, such as buildings, and register their location back to the GIS database. Furthermore, transmitting large image streams from remote locations would quickly use available band width (BW) precipitating the need for processing to occur at the sensor location. This paper addresses the problem of automatic target recognition. Utilizing an Adaptive Resonance Theory approach to cluster templates of target buildings processing and memory requirements can be significantly reduced allowing for processing at the sensor. The results show that the network successfully classifies targets and their location in a virtual test bed environment eventually leading to autonomous and passive information processing.

Backpropagation is the most widely used neural network learning technique. It is based on the mathematical notion of an ordered derivative. In this paper, we present a formulation of ordered derivatives and the backpropagation training algorithm using the important emerging area of mathematics known as the time scales calculus. This calculus, with its potential for application to a wide variety of inter-disciplinary problems, is becoming a key area of mathematics. It is capable of unifying continuous and discrete analysis within one coherent theoretical framework. Using this calculus, we present here a generalization of backpropagation which is appropriate for cases beyond the specifically continuous or discrete. We develop a new multivariate chain rule of this calculus, define ordered derivatives on time scales, prove a key theorem about them, and derive the backpropagation weight update equations for a feedforward multilayer neural network architecture. By drawing together the time scales calculus and the area of neural network learning, we present the first connection of two major fields of research.

This paper shows packet delivery rate can be improved by adopting learning-based hybrid routing strategies when a wired network suffers from severe link disruption. The dynamics of the link disruptions complicate the routing problem; successful and stable routing operations of conventional routing approaches are hindered as the level of disruption increases. The target is to develop a robust and efficient routing approach in a single structure. A robust routing approach means a packet should be delivered to a destination even under severe disruptions. Efficient routing should deliver a packet with the shortest path at no disruption. These goals should be achieved with the maximum utilization of preexisting network components and with the minimal human intervention once installed. Therefore, we chose a popular conventional routing scheme, link state, and add-ons that can learn changing network environment. Our approach is to add a learning agent and a simple routing scheme to link state in order to automatically select a better routing scheme at an arbitrary level of disruption. Markov decision process is employed to model this problem. The simulation results show robustness and packet delivery rate are increased up to 35% at acceptable cost of computational and architectural complexity even when link state approach is close to be collapsed.

We propose a new research organization management paradigm to increase throughput of projects by allowing researchers to choose their own projects through self-organization. Our methods draw upon the field of Agent-Based computational social science where Artificial Life and simulated societies have been used to study complex systems including economies and financial markets. Modeling the researchers as individual agents, we simulate our new management structure against a more traditional organization where the researchers are broken into departments based on their skills and assigned projects by management. Our results, measuring the amount of time it takes a research organization to serve a given number of contracts, show promise in the less hierarchical approach.

The study of strategic interaction among a society of agents is often handled using the machinery of game theory. This research examines how a Markov decision process (MDP) model may be applied to an important element of repeated game theory: the iterated prisoner's dilemma. Our study uses a Markovian approach to the game to represent the problem of in a computer simulation environment. A pure Markov approach is used on a simplified version of the iterated game and then we formulate the general game as a partially observable Markov decision process (POMDP). Finally, we use a cellular structure as an environment for players to compete and adapt. We apply both a simple replacement strategy and a cellular neural network to the environment.

This paper briefly summarizes some valuable properties of ART architectures that are advantageous in engineering applications, and outlines some areas of likely future progress, together with their motivations. Some of ART's advantages, such as its stability, biological plausibility, and responsiveness to the stability-plasticity dilemma, are well-described in the literature. This paper's focus will be on the advantages of scalability, speed, configurability, potential for parallelization, and ability to interpret the results. A valuable new area of innovation will be the application of ART to more generalized data structures such as trees and grammars. Continued progress on distributed representations would be valuable because of increased data representation capability, both in terms of system capacity and template complexity. Another valuable area of progress would be removal of the dichotomy between match-based and error-based learning.

The presence of large amounts of data in hyperspectral images makes it very difficult to perform further tractable analyses. Here, we present a method of analyzing real hyperspectral data by dimensionality reduction using diffusion maps. Diffusion maps interpret the eigenfunctions of Markov matrices as a system of coordinates on the original data set in order to obtain an efficient representation of data geometric descriptions. A neural network clustering theory, Fuzzy ART, is further applied to the reduced data to form clusters of the potential minerals. Experimental results on a subset of hyperspectral core imager data show that the proposed methods are promising in addressing the complicated hyperspectral data and identifying the minerals in core samples.

High-throughput messenger RNA (mRNA) expression profiling with microarray has been demonstrated as a more effective method of cancer diagnosis and treatment than the traditional morphology or clinical parameter-based methods. Recently, the discovery of a class of small non-coding RNAs, named microRNAs (miRNAs), provides another promising method of cancer classification. MIRNAs play a critical role in the tumorigenic process by functioning either as oncogenes or as tumor suppressors. Here, we apply a neural-based classifier, default ARTMAP, to classify different types of cancers based on their miRNA expression fingerprints. Experimental results on the multiple human cancers show that default ARTMAP performs consistently well on all the data, and the classification accuracy is better than or comparable to that of the other popular classifiers.

LabRat™is an autonomous, self-contained mobile robot kit with batteries, motors, two bumper whisker sensors, and three infrared proximity sensors that double as channels for “Rat-to-Rat” communication. The vehicle determines its position with an optical sensor that detects movement in both lateral directions. The LabRat™design is completely open source, including software examples and libraries. LabRat™is designed to fit inside the body of a computer mouse and has applications in the classroom, the lab and the home. The device has been successfully used in an undergraduate robotics class.

The advent of computers and the World Wide Web diversified the way in which the game of Go is played. While traditional human-to-human play still remains an important form of game play, amateur players, along with some professional players, have shifted the play domain from “off-line” club houses to “on-line” Go servers. Computer Go is an important field of study to develop a software to play Go or a Go engine. In addition to human-to-human play, a Go engine or computer intelligence to play Go adds another axis to play configuration: human-to-computer play and computer-to-computer play. These revolutions in the game of Go happened in an extremely short period of time compared to the history of the game, which is more than 4,000 years. We summarize this unavoidable change for the first time in the literature, to our knowledge, and propose a novel way to interact with the current technological advances. We present the new Human-Machine-Computer-Network Interface concept and our implementation of the machine interface with a robot arm. This Lynxmotion robotic arm named Cheonsoo-I successfully places stones on a board under the proposed architecture.

The problem of robotic area coverage is applicable to many domains, such as search, agriculture, cleaning, and machine tooling. The robotic area coverage task is concerned with moving a vehicle with an effector, or sensor, through the task space such that the sensor passes over every point in the space. For covering complex areas, back and forth paths are inadequate. This paper presents a real-time path planning architecture consisting of layers of a clustering method to divide and conquer the problem combined with a two layered, global and local optimization method. This architecture is able to optimize the execution of a series of waypoints for a restricted mobility vehicle, a fixed wing airplane.

The NetFlix Prize is a research contest that will award $1 Million to the first group to improve NetFlix’s movie recommendation system by 10%. Contestants are given a dataset containing the movie rating histories of customers for movies. From this data, a processing scheme must be developed that can predict how a customer will rate a given movie on a scale of 1 to 5. An architecture is presented that utilizes the Fuzzy-Adaptive Resonance Theory clustering method to create an interesting set of data attributes that are input to a neural network for mapping to a classification.

Much recent research activity has focused on the theory and application of quantum calculus. This branch of mathematics continues to find new and useful applications and there is much promise left for investigation into this field. We present a formulation of dynamic programming grounded in the quantum calculus. Our results include the standard dynamic programming induction algorithm which can be interpreted as the Hamilton-Jacobi-Bellman equation in the quantum calculus. Furthermore, we show that approximate dynamic programming in quantum calculus is tenable by laying the groundwork for the backpropagation algorithm common in neural network training. In particular, we prove that the chain rule for ordered derivatives, fundamental to backpropagation, is valid in quantum calculus. In doing this we have connected two major fields of research.

Early detection of a tumorpsilas site of origin is particularly important for cancer diagnosis and treatment. The employment of gene expression profiles for different cancer types or subtypes has already shown significant advantages over traditional cancer classification methods. Here, we apply a neural network clustering theory, Fuzzy ART, to generate the division of cancer samples, which is useful in investigating unknown cancer types or subtypes. On the other hand, we use diffusion maps, which interpret the eigenfunctions of Markov matrices as a system of coordinates on the original data set in order to obtain efficient representation of data geometric descriptions, for dimensionality reduction. The curse of dimensionality is a major problem in cancer type recognition-oriented gene expression data analysis due to the overwhelming number of measures of gene expression levels versus the small number of samples. Experimental results on the small round blue-cell tumor (SRBCT) data set, compared with other widely used clustering algorithms, demonstrate the effectiveness of our proposed method in addressing multidimensional gene expression data.

The time scales calculus, which includes the study of the Nabla derivatives, is an emerging key topic due to many multidisciplinary applications. We extend this calculus to approximate dynamic programming. In particular, we investigate application of the Nabla derivative, one of the fundamental dynamic derivatives of time scales. We present a Nabla-derivative based derivation and proof of the Hamilton-Jacobi-Bellman equation, the solution of which is the fundamental problem in the field of dynamic programming. By drawing together the calculus of time scales and the applied area of stochastic control via approximate dynamic programming, we connect two major fields of research.

The time scales calculus is a key emerging area of mathematics due to its potential use in a wide variety of multidisciplinary applications. We extend this calculus to approximate dynamic programming (ADP). The core backward induction algorithm of dynamic programming is extended from its traditional discrete case to all isolated time scales. Hamilton-Jacobi-Bellman equations, the solution of which is the fundamental problem in the field of dynamic programming, are motivated and proven on time scales. By drawing together the calculus of time scales and the applied area of stochastic control via ADP, we have connected two major fields of research.

Algorithm Development Environment for Permutation-based problems (ADEP) is a software environment for configuring meta-heuristics for solving combinatorial optimization problems. This paper describes the key features of ADEP and how the environment was used to generate a Memetic Algorithm (MA) solution for Hamiltonian Cycle Problems (HCP). The effectiveness of the MA algorithm is demonstrated through computer simulations and its performance is compared with backtracking and other heuristic techniques such as Simulated Annealing, Tabu Search, and Ant Colony Optimization.

In standard particle swarm optimization (PSO), the best particle in each neighborhood exerts its influence over other particles in the neighborhood. Fuzzy PSO is a generalization which differs from standard PSO in the following respect: charisma (influence over others) is defined to be a fuzzy variable, and more than one particle in each neighborhood can have a non-zero degree of charisma, and, consequently, is allowed to influence others to a degree that depends on its charisma. In this paper, we compare between the use of the Gaussian and Cauchy membership functions (MF) as the MF of the charisma fuzzy variable. We evaluate the performance of the two MFs using the weighted max-sat problem.

Modern graphics processing units (GPU) and game consoles are used for much more than simply 3D graphics applications and video games. From machine vision to finite element analysis, GPU's are being used in diverse applications, collectively called General Purpose computation onf graphics processor units (GPGPU). Additionally, game consoles are entering the market of high performance computing as inexpensive nodes in computing clusters. This paper explores the capabilities and limitations of modern GPU's and game consoles, surveying the ADP and neural network technologies that can be applied to these devices.

It is crucial for cancer diagnosis and treatment to accurately identify the site of origin of a tumor. With the emergence and rapid advancement of DNA microarray technologies, constructing gene expression profiles for different cancer types has already become a promising means for cancer classification. In addition to research on binary classification such as normal versus tumor samples, which attracts numerous efforts from a variety of disciplines, the discrimination of multiple tumor types is also important. Meanwhile, the selection of genes which are relevant to a certain cancer type not only improves the performance of the classifiers, but also provides molecular insights for treatment and drug development. Here, we use Semisupervised Ellipsoid ARTMAP (ssEAM) for multiclass cancer discrimination and particle swarm optimization for informative gene selection. ssEAM is a neural network architecture rooted in Adaptive Resonance Theory and suitable for classification tasks. ssEAM features fast, stable, and finite learning and creates hyperellipsoidal clusters, inducing complex nonlinear decision boundaries. PSO is an evolutionary algorithm-based technique for global optimization. A discrete binary version of PSO is employed to indicate whether genes are chosen or not. The effectiveness of ssEAM/PSO for multiclass cancer diagnosis is demonstrated by testing it on three publicly available multiple-class cancer data sets. ssEAM/PSO achieves competitive performance on all these data sets, with results comparable to or better than those obtained by other classifiers.

Genetic regulatory network inference is critically important for revealing fundamental cellular processes, investigating gene functions, and understanding their relations. The availability of time series gene expression data makes it possible to investigate the gene activities of whole genomes, rather than those of only a pair of genes or among several genes. However, current computational methods do not sufficiently consider the temporal behavior of this type of data and lack the capability to capture the complex nonlinear system dynamics. We propose a recurrent neural network (RNN) and particle swarm optimization (PSO) approach to infer genetic regulatory networks from time series gene expression data. Under this framework, gene interaction is explained through a connection weight matrix. Based on the fact that the measured time points are limited and the assumption that the genetic networks are usually sparsely connected, we present a PSO-based search algorithm to unveil potential genetic network constructions that fit well with the time series data and explore possible gene interactions. Furthermore, PSO is used to train the RNN and determine the network parameters. Our approach has been applied to both synthetic and real data sets. The results demonstrate that the RNN/PSO can provide meaningful insights in understanding the nonlinear dynamics of the gene expression time series and revealing potential regulatory interactions between genes.

The winner-take-all (WTA) network is useful in database management, very large scale integration (VLSI) design, and digital processing. The synthesis procedure of WTA on single-layer fully connected architecture with sigmoid transfer function is still not fully explored. We discuss the use of simultaneous recurrent networks (SRNs) trained by Kalman filter algorithms for the task of finding the maximum among N numbers. The simulation demonstrates the effectiveness of our training approach under conditions of a shared-weight SRN architecture. A more general SRN also succeeds in solving a real classification application on car engine data.

The twenty-two papers in this special issue are devoted to neural networks for feedback control systems. Covers some of the following topics: reinforcement learning; applications; neurocontrol systems; discrete time systems; and network architectures and training methods.

For applications such as force protection, an effective decision maker needs to maintain an unambiguous grasp of the environment. Opportunities exist to leverage computational mechanisms for the adaptive fusion of diverse information sources. The current research involves the use of neural networks and Markov chains to process information from sources including sensors, weather data, and law enforcement. Furthermore, the system operator's input is used as a point of reference for the machine learning algorithms. More detailed features of the approach are provided along with an example scenario.

Two image recognition systems based on random local descriptors are described. Random local descriptors play the role of features that have to be extracted from the image. The advantage of this type of features is a possibility to create sufficiently general description of the image. This approach was tested in different image recognition tasks: handwritten digit recognition, face recognition, metal surface texture recognition and micro work piece shape recognition. The best result for handwritten digit recognition on the MNIST database is the error rate of 0.37% and for face recognition on the ORL database is the error rate of 0.1%. The results for texture and micro work piece shape recognition are also promising.

This paper presents a neural network (NN) based decentralized excitation controller design for large scale power systems. The proposed controller design considers not only the dynamics of generators but also the algebraic constraints of the power flow equations. The control signals are calculated using only local signals. The transient stability and the coordination of the subsystem controllers can be guaranteed. NNs are used to approximate the unknown/imprecise dynamics of the local power system and the interconnections. All signals in the closed loop system are guaranteed to be uniformly ultimately bounded (UUB). Simulation results with a 3-machine power system demonstrate the effectiveness of the proposed controller design.

Technical and economic factors have caused the field of physical design automation to receive increasing attention and commercialization. The steady down-scaling of complementary metal oxide semiconductor (CMOS) device dimensions has been the main stimulus to the growth of microelectronics and computer-aided very large scale integration (VLSI) design. The more an Integrated Circuit (IC) is scaled, the higher its packing density becomes. For example, in 2006 Intel's 65-nm process technology for high performance microprocessor has a reduced gate length of 35 nanometers. In their 70-Mbit SRAM chip, there are up to 0.5 billion transistors in a 110 mm2chip size with 3.4 GHz clock speed. New technology generations come out every two years and provide an approximate 0.7 times transistor size reduction as predicted by Moore's Law. For the ultimate scaled MOSFET beyond 2015 or so, the transistor gate length is projected to be 10 nm and below. The continually increasing size of chips, measured in either area or number of transistors, and the wasted investment involving fabricating and testing faulty circuits, make layout analysis an important part of physical design automation. Layout-versus-schematic (LVS) is one of three kinds of layout analysis tools. Subcircuit extraction is the key problem to be solved in LVS. In LVS, two factors are important. One is run time, the other is identification correctness. This has created a need for computational intelligence. Fuzzy attributed graph is not only widely used in the fields of image understanding and pattern recognition, it is also useful to the fuzzy graph matching problem. Since the subcircuit extraction problem is a special case of a general-interest problem known as subgraph isomorphism, fuzzy attributed graphs are first effectively applied to the subgraph isomorphism problem. Then we provide an efficient fuzzy attributed graph algorithm based on the solution to subgraph isomorphism for the subcircuit extraction problem. Similarity measurement makes a significant contribution to evaluate the equivalence of two circuit graphs. To evaluate its performance, we compare fuzzy attributed graph approach with the commercial software called SubGemini, and two of the fastest approaches called DECIDE and SubHDP. We are able to achieve up to 12 times faster performance than alternatives, without loss of accuracy

A feature extractor and neural classifier for image recognition systems are proposed. The proposed feature extractor is based on the concept of random local descriptors (RLDs). It is followed by the encoder that is based on the permutation coding technique that allows to take into account not only detected features but also the position of each feature on the image and to make the recognition process invariant to small displacements. The combination of RLDs and permutation coding permits us to obtain a sufficiently general description of the image to be recognized. The code generated by the encoder is used as an input data for the neural classifier. Different types of images were used to test the proposed image recognition system. It was tested in the handwritten digit recognition problem, the face recognition problem, and the microobject shape recognition problem. The results of testing are very promising. The error rate for the Modified National Institute of Standards and Technology (MNIST) database is 0.44% and for the Olivetti Research Laboratory (ORL) database it is 0.1%.

A reactive fuzzy logic based control strategy was developed for mobile robot navigation. To decrease the number of fuzzy rules and related processing, a RAM-based neural network was combined with the fuzzy logic strategy. The fuzzy rules are used to interpret sensor information. The neural network uses results from the fuzzy logic as well as environmental information to make navigation decisions. The feasibility of this neuro-fuzzy approach was demonstrated on a mobile robot using a simple, 8-bit microcontroller. Experiments show the approach works well, as the robot was able to successfully avoid objects while seeking a goal in real-time. The neuro-fuzzy approach is code-efficient, fast, and easy to relate to the physical world

To overcome the shortcomings of fully analog and fully digital implementation of artificial neural networks (ANNs), we adopted mixed analog/digital technique. We proposed a switched-resistor (SR) element as a programmable synapse. The switched-resistor implementation of synapse captures both the advantages of analog implementation and the programmability of digital implementation. We also designed a CMOS analog neuron that performs a near-tanh nonlinearity function. We evaluated the performance of the neural networks using Pspice. The results showed that our approach can successfully implement the neural network, and exhibit a very high modularity

In standard particle swarm optimization (PSO), the best particle in each neighborhood exerts its influence over other particles in the neighborhood. In this paper, we propose fuzzy PSO, a generalization which differs from standard PSO in the following respect: charisma is defined to be a fuzzy variable, and more than one particle in each neighborhood can have a non-zero degree of charisma, and, consequently, is allowed to influence others to a degree that depends on its charisma. We evaluate our model on the weighted maximum satisfiability (maxsat) problem, comparing performance to standard PSO and to Walk-Sat.

Abduction is the process of proceeding from data describing a set of observations or events, to a set of hypotheses which best explains or accounts for the data. Cost-based abduction (CKA) is an AI formalism in which evidence to be explained is treated as a goal to be proven, proofs have costs based on how much needs to be assumed to complete the proof, and the set of assumptions needed to complete the least-cost proof are taken as the best explanation for the given evidence. In this paper, we introduce two techniques for improving the performance of high order recurrent networks (HORN) applied to cost-based abduction. In the backtrack-points technique, we use heuristics to recognize early that the network trajectory is moving in the wrong direction; we then restore the network state to a previously-stored point, and apply heuristic perturbations to nudge the network trajectory in a different direction. In the negative reinforcement technique, we add hyperedges to the network to reduce the attractiveness of local-minima. We apply these techniques on a 300-hypothesis, 900-rule particularly-difficult instance of CBA.

We applied an architecture which automates the design of simultaneous recurrent network (SRN) using a new evolutionary learning algorithm. This new evolutionary learning algorithm is based on a hybrid of particle swarm optimization (PSO) and evolutionary algorithm (EA). By combining the searching abilities of these two global optimization methods, the evolution of individuals is no longer restricted to be in the same generation, and better performed individuals may produce offspring to replace those with poor performance. The novel algorithm is then applied to the simultaneous recurrent network for the engine data classification. The experimental results show that our approach gives solid performance in categorizing the nonlinear car engine data.

This paper describes research to investigate an artificial neural network (ANN) approach to minimize aircraft cabin noise in flight. The ANN approach is shown to be able to accurately model the non-linear relationships between engine unbalance, airframe vibration, and cabin noise to overcome limitations associated with traditional linear influence coefficient methods. ANN system inverse models are developed using engine test-stand vibration data and on-airplane vibration and noise data supplemented with influence coefficient empirical data. The inverse models are able to determine balance solutions that satisfy cabin noise specifications. The accuracy of the ANN model with respect to the real system is determined by the quantity and quality of test stand and operational aircraft data. This data-driven approach is particularly appealing for implementation on future systems that include continuous monitoring processes able to capture data while in operation.

A feature extractor and neural classifier for image recognition system are proposed. They are based on the permutative coding technique which continues our investigations on neural networks. It permits us to obtain sufficiently general description of the image to be recognized. Different types of images were used to test the proposed image recognition system. It was tested on the handwritten digit recognition problem, the face recognition problem and the shape of microobjects recognition problem. The results of testing are very promising. The error rate for the MNIST database is 0.44% and for the ORL database is 0.1%.

Presents the welcome message from the conference proceedings.

Large-scale time series gene expression data generated from DNA microarray experiments provide us a new means to reveal fundamental cellular processes, investigate functions of genes, and understand their relations and interactions. To infer gene regulatory networks from these data with effective computational tools has attracted intensive efforts from artificial intelligence and machine learning. Here, we use a recurrent neural network (RNN), trained with particle swarm optimization (PSO), to investigate the behaviors of regulatory networks. The experimental results, on a synthetic data set and a real data set, show that the proposed model and algorithm can effectively capture the dynamics of the gene expression time series and are capable of revealing regulatory interactions between genes.

Data analysis plays an indispensable role for understanding various phenomena. Cluster analysis, primitive exploration with little or no prior knowledge, consists of research developed across a wide variety of communities. The diversity, on one hand, equips us with many tools. On the other hand, the profusion of options causes confusion. We survey clustering algorithms for data sets appearing in statistics, computer science, and machine learning, and illustrate their applications in some benchmark data sets, the traveling salesman problem, and bioinformatics, a new field attracting intensive efforts. Several tightly related topics, proximity measure, and cluster validation, are also discussed.

To predict the 100 missing values from the time series consisting of 5000 data given for the IJCNN 2004 time series prediction competition, we applied an architecture which automates the design of recurrent neural networks using a new evolutionary learning algorithm. This new evolutionary learning algorithm is based on a hybrid of particle swarm optimization (PSO) and evolutionary algorithm (EA). By combining the searching abilities of these two global optimization methods, the evolution of individuals is no longer restricted to be in the same generation, and better performed individuals may produce offspring to replace those with poor performance. The novel algorithm is then applied to the recurrent neural network for the time series prediction. The experimental results show that our approach gives good performance in predicting the missing values from the time series.

This paper describes the use of a multi-stream extended Kalman filter (EKF) to tackle the IJCNN 2004 challenge problem - time series prediction on CATS benchmark. A weighted bidirectional approach was adopted in the experiments to incorporate the forward and backward predictions of the time series. EKF is a practical, general approach to neural networks training. It consists of the following: 1) gradient calculation by backpropagation through time (BPTT); 2) weight updates based on the extended Kalman filter; and 3) data presentation using multi-stream mechanics.

Power system stabilizers are widely used to generate supplementary control signals for the excitation system in order to damp out the low frequency oscillations. This paper proposes a stable neural network (NN) controller for the stabilization of a single machine infinite bus power system. In the power system control literature, simplified-analytical models are used to represent the power system and the controller designs are not based on rigorous stability analysis. This work overcomes the two major problems by using an accurate analytical model for controller development and presents the closed-loop stability analysis. The NN is used to approximate the complex nonlinear power system online and the weights of which can be set to zero to avoid the time consuming offline training process. Magnitude constraint of the activators is modeled as saturation nonlinearities and is included in the Lyapunov stability analysis. Simulation results demonstrate that the proposed design can successfully damp out oscillations. The control algorithms of This work can also be applied to other similar control problems.

Large-scale gene expression data coming from microarray

The purpose of the paper is an experimental study of the formation of class descriptions, taking place during learning, in assembly neural networks. The assembly neural network is artificially partitioned into several sub-networks according to the number of classes that the network has to recognize. The features extracted from input data are represented in neural column structures of the sub-networks. Hebbian neural assemblies are formed in the column structure of the sub-networks by weight adaptation. A specific class description is formed in each sub-network of the assembly neural network due to intersections between the neural assemblies. The process of formation of class descriptions in the sub-networks is interpreted as feature generalization. A set of special experiments is performed to study this process, on a task of character recognition using the MNIST database.

Fuzzy attributed graph (FAG) is not only widely used in the fields of image understanding and pattern recognition, but is useful to fuzzy graph matching problem. One of the applications of fuzzy graph matching is the subcircuit extraction problem. Subcircuit extraction problem is very important for VLSI testing, layout versus schematic (LVS) check, and circuit partition, etc. In this paper, fuzzy attributed graph (FAG) is first effectively applied to the subgraph isomorphism problem. And then we provide an efficient fuzzy attributed graph algorithm based on the solution to subgraph isomorphism for the subcircuit extraction problem. Similarity measurement makes a significant contribution to both the subgraph isomorphism problem and the subcircuit extraction problem.

Optimizing the membership functions of a fuzzy system can be viewed as a system identification problem for a nonlinear dynamic system. Basically, we can view the optimization of fuzzy membership functions as a weighted least-squares minimization problem, where the error vector is the difference between the fuzzy system outputs and the target values for those outputs. The extended Kalman filter algorithm is a good choice to solve this system identification problem, not only because it is a derivative-based algorithm that is suitable to solve the weighted least-squares minimization problem, but also because of its appealing predictor-corrector feature for nonlinear system model. In this paper, we present an extended Kalman filter approach to optimize the membership functions of the inputs and outputs of the fuzzy controller. The effect of the measurement noise covariance R on the convergence of the fuzzy controller is also investigated. Experimental results show that the optimized fuzzy controller achieves significant improvement on performance. In addition, the smaller the measurement noise covariance R is, the faster the optimized fuzzy controller would converge.

One important application of mobile robots is searching a geographical region to locate the origin of a specific sensible phenomenon. We first propose a fuzzy logic approach using a decision table. A novel fuzzy rule based was designed. And then a fuzzy search strategy is adopted by utilizing the three tier centers of mass coordination. Experimental results show that fuzzy logic algorithm is an efficient approach for the collective robots to locate the target source. In addition, noise and the position of the target affect the searching result.

Strain prediction at various locations on a smart composite wing can provide useful information on its aerodynamic condition. The smart wing consisted of a glass/epoxy composite beam with three extrinsic Fabry-Perot interferometric (EFPI) sensors mounted at three different locations near the wing root. Strain acting on the three sensors at different air speeds and angles-of-attack were experimentally obtained in a closed circuit wind tunnel under normal conditions of operation. A function mapping the angle of attack and air speed to the strains on the three sensors was simulated using feedforward neural networks trained using a backpropagation training algorithm. This mapping provides a method to predict the stall condition by comparing the strain available in real time and the predicted strain by the trained neural network.

Method of elastic maps allows to construct efficiently 1D, 2D and 3D nonlinear approximations to the principal manifolds with different topology (piece of plane, sphere, torus etc.) and to project data onto it. We describe the idea of the method and demonstrate its applications in analysis of genetic sequences.

This paper describes the use of artificial neural networks (ANNs) with the vibration data from real flight tests for detecting engine health condition - mass imbalance herein. Order-tracking data, calculated from time series is used as the input to the neural networks to determine the amount and location of mass imbalance on aircraft engines. Several neural network methods, including multilayer perceptron (MLP), extended Kalman filter (EKF) and support vector machines (SVMs) are used in the neural network inverse model for the performance comparison. The promising performances are presented at the end.

Power system stabilizers (PSS) are used to generate supplementary control signals for the excitation system in order to damp the low frequency power system oscillations. To overcome the drawbacks of conventional PSS (CPSS), numerous techniques have been proposed in the literature. Based on the analysis of existing techniques, this paper presents an indirect adaptive neural network based power system stabilizer (IDNC) design. The proposed IDNC consists of a neuro-controller, which is used to generate a supplementary control signal to the excitation system, and a neuro-identifier, which is used to model the dynamics of the power system and to adapt the neuro-controller parameters. The proposed method has the features of a simple structure, adaptivity and fast response. The proposed IDNC is evaluated on a single machine infinite bus power system under different operating conditions and disturbances to demonstrate its effectiveness and robustness.

The traveling salesman problem (TSP) is a very hard optimization problem in the field of operations research. It has been shown to be NP-complete, and is an often-used benchmark for new optimization techniques. One of the main challenges with this problem is that standard, non-AI heuristic approaches such as the Lin-Kernighan algorithm (LK) and the chained LK variant are currently very effective and in wide use for the common fully connected, Euclidean variant that is considered here. This paper presents an algorithm that uses adaptive resonance theory (ART) in combination with a variation of the Lin-Kernighan local optimization algorithm to solve very large instances of the TSP. The primary advantage of this algorithm over traditional LK and chained-LK approaches is the increased scalability and parallelism allowed by the divide-and-conquer clustering paradigm. Tours obtained by the algorithm are lower quality, but scaling is much better and there is a high potential for increasing performance using parallel hardware.

Over the past few years, security has been an increasing concern, with the growth of network and technological development. An intrusion detection system is a critical component for secure information management. Unfortunately, present IDS's falls short of providing protection required for growing concern. Creation of an IDS to detect anomaly intrusions, in a timely and accurate manner, has been an elusive goal for researchers. This paper describes a host-based IDS model, utilizing a Radial Basis Function neural network. It functions as a combined anomaly/misuse detector that helps to overcome most of the limitations in existing models. Rather than creating user profiles or behavioral characteristics, we trained our network using session data in the identification and tested experimentally on different attack/normal sessions. These results suggest that training the IDS on session data is not only effective in detecting intrusions, but also accurate and timely.

With the emergence and rapid advancement of DNA microarray technologies, construction of gene expression profiles for different cancer types has already become a promising means for cancer diagnosis and treatment. Most previous research has focused on binary classification. Here, we use a probabilistic neural network (PNN) for multi-classification of cancer data. The experimental results demonstrate the effectiveness of the PNN in addressing gene expression data.

A RAM-based neural network is being developed for a mobile robot controlled by a simple microprocessor system. Conventional neural networks often require a powerful and sophisticated computer system. Training a multi-layer neural network requires repeated presentation of training data, which often results in very long learning time. The goal for this paper is to demonstrate that RAM-based neural networks are a suitable choice for embedded applications with few computational resources. This functionality is demonstrated in a simple robot powered by an 8051 microcontroller with 512 bytes of RAM. The RAM-based neural network allows the robot to detect and avoid obstacles in real time.

An important application of mobile robots is searching a region to locate the origin of a specific phenomenon. A variety of optimization algorithms can be employed to locate the target source, which has the maximum intensity of the distribution of some detected function. We propose two neural network algorithms: stochastic optimization algorithm and dual heuristic programming (DHP) to solve the collective robotic search problem. Experiments were carried out to investigate the effect of noise and the number of robots on the task performance, as well as the expenses. The experimental results showed that the performance of the dual heuristic programming (DHP) is better than the stochastic optimization method.

Models of real-world applications often include a large number of parameters with a wide dynamic range, which contributes to the difficulties of neural network training. Creating the training data set for such applications becomes costly, if not impossible. In order to overcome the challenge, one can employ an active learning technique known as query-based learning (QBL) to add performance-critical data to the training set during the learning phase, thereby efficiently improving the overall learning/generalization. The performance-critical data can be obtained using an inverse mapping called network inversion (discrete network inversion and continuous network inversion) followed by oracle query. This paper investigates the use of both inversion techniques for QBL learning, and introduces an original heuristic to select the inversion target values for continuous network inversion method. Efficiency and generalization was further enhanced by employing node decoupled extended Kalman filter (NDEKF) training and a causality index (CI) as a means to reduce the input search dimensionality. The benefits of the overall QBL approach are experimentally demonstrated in two aerospace applications: a classification problem with large input space and a control distribution problem.

This paper presents the design and practical hardware implementation of optimal neurocontrollers that replace the conventional automatic voltage regulator (AVR) and the turbine governor of turbogenerators on multimachine power systems. The neurocontroller design uses a powerful technique of the adaptive critic design (ACD) family called dual heuristic programming (DHP). The DHP neurocontrollers' training and testing are implemented on the Innovative Integration M67 card consisting of the TMS320C6701 processor. The measured results show that the DHP neurocontrollers are robust and their performance does not degrade unlike the conventional controllers even when a power system stabilizer (PSS) is included, for changes in system operating conditions and configurations. This paper also shows that it is possible to design and implement optimal neurocontrollers for multiple turbogenerators in real time, without having to do continually online training of the neural networks, thus avoiding risks of instability.

Evolutionary programming was used to attempt to optimize a program written in the pseudo-assembly language Redcode, invented by A.K. Dewdney. Corewars is the game under which Redcode programs compete. Since 1994, the last standardization of Redcode, many complicated, effective Redcode programs have been written by people, but intense study is required to learn the nuances of the language and perfect programs. Since this is such a difficult task, evolutionary techniques may outperform humans. Multiple point, variable length crossover and change, insert, and delete mutations were the operators used. Relative fitnesses were calculated within a subset of the population on remote client computers. A food model was used to select the most fit programs. Current results are preliminary, but already one of the resulting programs wins 38% and ties 29% against a common type of human-written program. The best performance is 151 wins, 49 losses, and 0 ties against a typical human program.

This paper presents the design of an optimal neurocontroller that replaces the conventional automatic voltage regulator (AVR) and the turbine governor for a turbogenerator connected to the power grid. The neurocontroller design uses a novel technique based on the adaptive critic designs (ACDs), specifically on heuristic dynamic programming (HDP) and dual heuristic programming (DHP). Results show that both neurocontrollers are robust, but that DHP outperforms HDP or conventional controllers, especially when the system conditions and configuration change. This paper also shows how to design optimal neurocontrollers for nonlinear systems, such as turbogenerators, without having to do continually online training of the neural networks, thus avoiding risks of instability.

Traditionally, fed-batch biochemical process optimization and control uses complicated off-line optimizers, with no online model adaptation or re-optimization. This study demonstrates the applicability of a class of adaptive critic designs for online re-optimization and control of an aerobic fed-batch fermentor. Specifically, the performance of an entire class of adaptive critic designs, viz., heuristic dynamic programming, dual heuristic programming and generalized dual heuristic programming, was demonstrated to be superior to that of a heuristic random optimizer, on optimization of a fed-batch fermentor operation producing monoclonal antibodies.

We present an approach to compute the efficient frontier for portfolio optimization based on evolutionary programming (EP) technique. Our approach relies on multiple EP runs within a search to create the frontier. Results from simulation, which runs on a personal computer platform, are shown for data set consisting of 24 types of securities. The algorithm converges quickly with consistent performance, making it suitable for creating an efficient frontier for a much larger number of assets. The versatility of the approach makes it viable to accommodate constraints or scenarios, which we perceive as either investors or market imposed conditions. Our technique opens up an avenue to conveniently overcome the symptomatic "unrealizable or unreasonable portfolios" syndrome that plagued methodology that relies on identifying corner portfolios as a basis for creating the frontier.

This paper presents the design of a neurocontroller for a turbogenerator that augments/replaces the conventional automatic voltage regulator and the turbine governor. The neurocontroller uses a novel technique based on the adaptive critic designs with emphasis on heuristic dynamic programming (HDP) and dual heuristic programming (DHP). Results are presented to show that the DHP based neurocontroller is robust and performs better than the HDP based neurocontroller, as well as the conventional controller, especially when the system conditions and configuration changes.

The generalized maze problem has been considered as an interesting testbed by various researchers in AI and neural networks. The most significant results, from a neural networks point of view, were: 1. Simultaneous recurrent networks are necessary if a neural network-based cellular automaton approach to the problem is to be successful. 2. These networks can be designed so that convergence to a correct solution is assured. Here, a simple closed-form solution for the critic is shown, making adaptation unnecessary. Furthermore, it is shown that the design converges to the correct solution in only J steps, and the worst case convergence speed for an N/spl times/N mesh is derived.

The ball-and-beam problem is a benchmark for testing control algorithms. In the World Congress on Neural Networks, 1994, Prof. L. Zadeh proposed a twist to the problem, which, he suggested, would require a fuzzy logic controller. This experiment uses a beam, partially covered with a sticky substance, increasing the difficulty of predicting the ball's motion. We complicated this problem even more by not using any information concerning the ball's velocity. Although it is common to use the first differences of the ball's consecutive positions as a measure of velocity and explicit input to the controller, we preferred to exploit recurrent neural networks, inputting only consecutive positions instead. We have used truncated backpropagation through time with the node-decoupled extended Kalman filter (NDEKF) algorithm to update the weights in the networks. Our best neurocontroller uses a form of approximate dynamic programming called an adaptive critic design. A hierarchy of such designs exists. Our system uses dual heuristic programming (DHP), an upper-level design. To our best knowledge, our results are the first use of DHP to control a physical system. It is also the first system we know of to respond to Zadeh's challenge. We do not claim this neural network control algorithm is the best approach to this problem, nor do we claim it is better than a fuzzy controller. It is instead a contribution to the scientific dialogue about the boundary between the two overlapping disciplines.

Chartists usually rely on technical indicators to predict trends in time series charts. Although the indicators are precise and most practitioners tend to concur to a large extent on the general meaning of the indicators, it is hard to specify precise thresholds as a basis for deciding on a particular course of action. Probabilistic based approaches do offer recourse for handling such kinds of uncertainty. However, a probabilistic mode of managing uncertainty lacks the flair for capturing the essence of subjectivity that unfortunately (or fortunately) is an inherent trait of human chartists. Fuzzy logic therefore offers a better alternative in this sense. It is further suggested that the fuzzy knowledge can be tuned by means of various existing learning algorithms to accommodate the needs of the chartists.

The strength of synapses from granule cells (GrC) to Purkinje cells (PC) changes, as a function of the interval between firing of GrCs and the PC's climbing fiber (CF). In the case of constant and piecewise constant signals, an adequate tuning could be provided by a phase-based memorizing system, based on the observed properties of cerebellar neurons. We now discuss the cerebellar model for general nonstationary signals.

At the IJCNN'93 in Nagoya we pronounced a challenging goal: to get activity patterns of pairs of Purkinje cells (PC), controlled with the same climbing fiber (CF), a CF PC twins problem. Here, for the first time in cerebellar studies, CF PC twins have been identified and studied. Several important features of the CF PC twins activity are demonstrated: (1) high constancy of conduction time of impulses of cells of inferior olives to the targeted PCs; (2) a relatively high failure rate (0.05-0.18) of impulse propagation into terminal branches of CF; (3) a salient difference in complex spikes (CS)-simple spikes (SS) interaction between the PC twins; and (4) SS cross-correlation between twin cells is zero, thus contradicting a naive prediction of several cerebellar learning theories.

Neural networks with active neurons which self-organize their structure can use inductive sorting-out GMDH algorithms for their neurons. New threshold type GMDH algorithm with polynomial complexity is developed to decrease computing time in case of large input data sample.

Traditionally fed-batch biochemical process optimization and control uses complicated theoretical off-line optimizers, with no online model adaptation or re-optimization. This study demonstrates the applicability, effectiveness, and economic potential of a simple phenomenological model for modeling, and an adaptive critic design, generalized dual heuristic programming, for online re-optimization and control of an aerobic fed-batch fermentor. The results are compared with those obtained using a heuristic random optimizer.

Two coarse coding schemes are considered: the random subspace scheme of the authors, and the modified Kanerva model of Prager et al. (1993). Some properties and characteristics of these schemes are investigated experimentally and by analysing their geometrical interpretation. Both schemes do not require exponential growth of the binary code dimensionality against that of the input space. The random subspace scheme allows the code density to be independent from the maximal dimensionality of hyper-rectangle receptive fields. It is especially important when low-dimensional receptive fields are required, as with classifiers or approximators of real-world data.

We use a multi-layer perceptron (MLP) network to estimate wind turbine power generation. Wind power can be influenced by many factors such as wind speeds, wind directions, terrain, air density, vertical wind profile, time of day, and seasons of the year. It is usually important to train a neural network with multiple influence factors and big training data set. We have parallelized the extended Kalman filter (EKF) training algorithm, which can provide fast training even for large training data sets. The MLP network is then trained with the consideration of various possible factors, which can influence turbine power production. The performance of the trained network is studied from the point of view of information presented to the network through network inputs regarding different affecting factors and large training data set covering all the seasons of a year.

Strategies to efficiently train a neural network for an aerospace problem with a large multidimensional input space are developed and demonstrated. The neural network provides classification for over 100,000,000 data points. A query-based strategy is used that initiates training using a small input set, and then augments the set in multiple stages to include important data around the network decision boundary. Neural network inversion and oracle query are used to generate the additional data, jitter is added to the query data to improve the results, and an extended Kalman filter algorithm is used for training. A causality index is discussed as a means to reduce the dimensionality of the problem based on the relative importance of the inputs.

In virtual reality (VR), head movement is tracked through inertial and optical sensors. Computation and communication times result in delays between measurements and updating of the new frame in the head mounted display(HMD). These delays result in problems, including motion sickness. We use recurrent and time delay neural networks to predict the head location and use it to calculate the new frame. A predictability analysis is used in designing the prediction system.

The temporal difference (TD) method is applied on a committee of neural network experts to learn the board evaluation function for the oriental board game Go. The game has simple rules but requires complex strategies to play well, and the conventional tree search algorithm for computer games makes a poor Go program. Thus, the game Go is an ideal problem domain for exploring machine learning algorithms. Here, the neural networks learned a board evaluation function for Go played on 9/spl times/9 board sizes. Two learning algorithms, e.g., hybrid mixture of experts (HME) and Meta-Pi, are used to train the neural network experts. Both algorithms learned good Go evaluation functions and the neural network based Go engines were able to defeat a public domain rule-based program more than 50% of the times. The performances of the mixture networks are compared with that of a single feedforward network trained similarly.

Fuzzy fault trees provide a powerful and computationally efficient technique for developing fuzzy probabilities based on independent inputs. The probability of any event that can be described in terms of a sequence of independent unions, intersections, and complements may be calculated by a fuzzy fault tree. Unfortunately, fuzzy fault trees do not provide a complete theory: many events of substantial practical interest cannot be described only by independent operations. Thus, the standard fuzzy extension (based on fuzzy fault trees) is not complete since not all events are assigned a fuzzy probability. Other complete extensions have been proposed, but these extensions are not consistent with the calculations from fuzzy fault trees. We propose a new extension of crisp probability theory. Our model is based on n independent inputs, each with a fuzzy probability. The elements of our sample space describe exactly which of the n input events did and did not occur. Our extension is complete since a fuzzy probability is assigned to every subset of the sample space. Our extension is also consistent with all calculations that can be arranged as a fault tree. Our approach allows the reliability analyst to develop complete and consistent fuzzy reliability models from existing crisp reliability models. This allows a comprehensive analysis of the system. Computational algorithms are provided both to extend existing models and develop new models. The technique is demonstrated on a reliability model of a three-stage industrial process.

The archival magnetic typed data of experiments with cat wakefulness and sleep are analyzed. General goals and possible tools for the analysis of non-linear interactions in the nervous system are discussed The interaction between air-flow data, ponto-geniculo-occipital (PGO) waves, brain-stem neurons and electrical activity recorded in the diaphragm is analyzed. The analysis methods are engineered to meet the data processing requirements and include the analysis of individual potential waves and procedures for treating the non-stationary properties of the signals due to respiration. This analysis has definitely confirmed the existence of inhibition, which PGO waves exert on the diaphragm activity, and patterns of interaction of neurons with the diaphragm.

Three networks are compared for low false alarm stock trend predictions. Short-term trends, particularly attractive for neural network analysis, can be used profitably in scenarios such as option trading, but only with significant risk. Therefore, we focus on limiting false alarms, which improves the risk/reward ratio by preventing losses. To predict stock trends, we exploit time delay, recurrent, and probabilistic neural networks (TDNN, RNN, and PNN, respectively), utilizing conjugate gradient and multistream extended Kalman filter training for TDNN and RNN. We also discuss different predictability analysis techniques and perform an analysis of predictability based on a history of daily closing price. Our results indicate that all the networks are feasible, the primary preference being one of convenience.

We discuss a variety of adaptive critic designs (ACDs) for neurocontrol. These are suitable for learning in noisy, nonlinear, and nonstationary environments. They have common roots as generalizations of dynamic programming for neural reinforcement learning approaches. Our discussion of these origins leads to an explanation of three design families: heuristic dynamic programming, dual heuristic programming, and globalized dual heuristic programming (GDHP). The main emphasis is on DHP and GDHP as advanced ACDs. We suggest two new modifications of the original GDHP design that are currently the only working implementations of GDHP. They promise to be useful for many engineering applications in the areas of optimization and optimal control. Based on one of these modifications, we present a unified approach to all ACDs. This leads to a generalized training procedure for ACDs.

In the above paper by Mao-Jain (ibid., vol.7 (1996)), the Mahalanobis distance is used instead of Euclidean distance as the distance measure in order to acquire the hyperellipsoidal clustering. We prove that the clustering cost function is a constant under this condition, so hyperellipsoidal clustering cannot be realized. We also explains why the clustering algorithm developed in the above paper can get some good hyperellipsoidal clustering results. In reply, Mao-Jain state that the Wang-Xia failed to point out that their HEC clustering algorithm used a regularized Mahalanobis distance instead of the standard Mahalanobis distance. It is the regularized Mahalanobis distance which plays an important role in realizing hyperellipsoidal clusters. In conclusion, the comments made by Wang-Xia together with this response provide some new insights into the behavior of their HEC clustering algorithm. It further confirms that the HEC algorithm is a useful tool for understanding the structure of multidimensional data.

Describes a system that predicts significant short-term price movement in a single stock utilizing conservative strategies. We use preprocessing techniques, then train a probabilistic neural network to predict only price gains large enough to create a significant profit opportunity. Our primary objective is to limit false predictions (known in the pattern recognition literature as false alarms). False alarms are more significant than missed opportunities, because false alarms acted upon lead to losses. We can achieve false alarm rates as low as 5.7% with the correct system design and parameterization.

A solution to the problem of implementation of the adaptive resonance theory (ART) of neural networks that uses an optical correlator which allows the large body of correlator research to be leveraged in the implementation of ART is presented. The implementation takes advantage of the fact that one ART-based architecture, known as ART1, can be broken into several parts, some of which are better to implement in parallel. The control structure of ART, often regarded as its most complex part, is actually not very time consuming and can be done in electronics. The bottom-up and top-down gated pathways, however, are very time consuming to simulate and are difficult to implement directly in electronics due to the high number of interconnections. In addition to the design, the authors present experiments with a laboratory prototype to illustrate its feasibility and to discuss implementation details that arise in practice. This device can potentially outperform alternative implementations of ART1 by as much as two to three orders of magnitude in problems requiring especially large input fields.