Two-step Optimization Scheme Research Articles

Self-Paced Co-Training of Graph Neural Networks for Semi-Supervised Node Classification.

Graph neural networks (GNNs) have demonstrated great success in many graph data-based applications. The impressive behavior of GNNs typically relies on the availability of a sufficient amount of labeled data for model training. However, in practice, obtaining a large number of annotations is prohibitively labor-intensive and even impossible. Co-training is a popular semi-supervised learning (SSL) paradigm, which trains multiple models based on a common training set while augmenting the limited amount of labeled data used for training each model via the pseudolabeled data generated from the prediction results of other models. Most of the existing co-training works do not control the quality of pseudolabeled data when using them. Therefore, the inaccurate pseudolabels generated by immature models in the early stage of the training process are likely to cause noticeable errors when they are used for augmenting the training data for other models. To address this issue, we propose a self-paced co-training for the GNN (SPC-GNN) framework for semi-supervised node classification. This framework trains multiple GNNs with the same or different structures on different representations of the same training data. Each GNN carries out SSL by using both the originally available labeled data and the augmented pseudolabeled data generated from other GNNs. To control the quality of pseudolabels, a self-paced label augmentation strategy is designed to make the pseudolabels generated at a higher confidence level to be utilized earlier during training such that the negative impact of inaccurate pseudolabels on training data augmentation, and accordingly, the subsequent training process can be mitigated. Finally, each of the trained GNN is evaluated on a validation set, and the best-performing one is chosen as the output. To improve the training effectiveness of the framework, we devise a pretraining followed by a two-step optimization scheme to train GNNs. Experimental results on the node classification task demonstrate that the proposed framework achieves significant improvement over the state-of-the-art SSL methods.

Crystal Plasticity simulations of in situ tensile tests: A two-step inverse method for identification of CP parameters, and assessment of CPFEM capabilities

As the computational capability of modern computers increases, the Crystal Plasticity Finite Element Method (CPFEM) becomes more and more popular in materials science to model the mechanical behaviour of polycrystals. Indeed, such analysis provides extensive information about local mechanical fields (such as plastic strain and stress), which can be useful for understanding the behaviour of bulk materials. However, estimating the parameters of the CP constitutive laws is still challenging because they are not directly related to the macroscopic behaviour of the polycrystalline aggregates. Thus, one way to identify such parameters is by inverse analysis from CPFEM simulations. However, such approach is usually extremely time consuming. This paper proposes a two-step optimization scheme to determine these coefficients. The first step is based on a simple model, similar to that proposed by Sachs back in 1928. The second step is based on CPFEM simulations, to be compared with experimental data acquired by an in situ tensile test and full-field measurements made by High-Resolution Digital Image Correlation (HRDIC). The uniqueness of the solution found by inverse analysis is studied and ways to solve the local minima issues are provided. Finally, the ability of CPFEM to replicate an in situ tensile test is assessed.

Predictive Torque Control for Open-End Winding PMSM With Common DC Bus Based on Weighting Factorless and Finite Control Set Optimization

The conventional predictive torque control (PTC) for three-phase open-end winding permanent magnet synchronous motor (OEW-PMSM) has the following disadvantages: tedious tuning of weighting factors, large ripples of zero-sequence current, electromagnetic torque, and stator flux. To overcome the above problems, a PTC based on weighting factorless and finite control set (FCS) optimization is proposed in this article. First, a deadbeat control strategy is proposed to convert the control of zero-sequence current, electromagnetic torque, and stator flux into the control of stator voltage vector. Since the equivalent control objects have the same unit, the weighting factors are not required. Second, a two-step optimization scheme for FCS is proposed to reduce the ripples of zero-sequence current, electromagnetic torque, and stator flux. Finally, an optimized pulsewidth modulation (PWM) strategy, which can help to reduce the switching frequency, is proposed. The feasibility and effectiveness of the proposed PTC strategy is verified by experiments on a 1-kW OEW-PMSM.

Full View Optical Flow Estimation Leveraged From Light Field Superpixel

In this paper, we present a full view optical flow estimation method for plenoptic imaging. Our method employs the structure delivered by the four-dimensional light field over multiple views making use of superpixels. These superpixels are four dimensional in nature and can be used to represent the objects in the scene as a set of slanted-planes in three-dimensional space so as to recover a piecewise rigid depth estimate. Taking advantage of these superpixels and the corresponding slanted planes, we recover the optical flow and depth maps by using a two-step optimization scheme where the flow is propagated from the central view to the other views in the imagery. We illustrate the utility of our method for depth and flow estimation making use of a dataset of synthetically generated image sequences and real-world imagery captured using a Lytro Illum camera. We also compare our results with those yielded by a number of alternatives elsewhere in the literature.

A two-step optimization scheme based on equivalent stiffness parameters for forcing convexity of fiber winding angle in composite frames

For stiffness design optimization of composite frame structures, one of the major problems when using fiber winding angles as design variables directly is the lack of convexity of the objective function, which may lead to different local optima depending on initial designs when a traditional gradient-based optimization algorithm is applied. Therefore, the present paper adopts a gradient-based two-step optimization scheme to cope with the difficulty and search for a better optimal design of composite frames in which the fiber winding angles are taken as design variables. To realize the two-step optimization scheme, the equivalent stiffness parameters of a composite beam with circular cross-section are derived in explicit expressions and used to force the convexity of the design optimization of the composite frame. The stiffness matrices are linearly expressed in terms of the stiffness parameters, which guarantee the convexity of the design variable feasible region in the stiffness parameter space. The equivalent stiffness parameters are adopted to keep invariance of physical quantities between fiber winding angle and equivalent stiffness parameter spaces. In the two-step optimization scheme, the minimum identification problem with the constraint that the objective function at the new starting point is less than or equal to the previous objective function at the optimum point in fiber winding angle space is established. Then, the two-step optimization scheme can be implemented in the fiber winding angle and structural equivalent stiffness parameter spaces, respectively, until the minimum identification problem is not possible to identify a new starting point. The proposed two-step optimization scheme for composite frames fully takes advantage of the stiffness parameters in convexity and fiber winding angles as practically physical quantities, respectively. The sensitivity information of the objective function with respect to fiber winding angles and equivalent stiffness parameters is derived by the analytical sensitivity analysis method. Numerical examples show that the two-step optimization scheme can effectively force convexity of the optimization model and help to eliminate the initial design dependency. The effectiveness of the proposed two-step scheme is further verified through the particle swarm optimization (PSO) algorithm which is an evolutionary algorithm with global optimization capability.

Energy cost saving strategies in distributed power networks

In this paper we study energy cost saving strategies in power networks in presence of prosumers. Three tips are considered: (i) distributed power network architecture, (ii) peak energy shaving with the integration of prosumers’ contribution, (iii) Electric vehicles self-charging by means of prosumers’ production. The proposed distributed power network architecture reduces significantly the transmission costs and can reduce significantly the global energy cost up to 42 percent. Different types of prosumer who use self-charging photovoltaic systems, are able to intelligently buy energy from, or sell it, to the power grid. Therein, prosumers interact in a distributed environment during the purchase or sale of electric power using a double auction with negotiation mechanism. Using a two-step combined learning and optimization scheme, each prosumer can learn its optimal bidding strategy and forecast its energy production, consumption and storage. Our simulation results, conducted for the region of Sicily in Italy, show that the integration of prosumers can reduce peak hour costs up to 19 percent and 6 percent for eligible prosumers with electric vehicles.

Automated Design of Common-Mode Suppressed Balanced Wideband Bandpass Filters by Means of Aggressive Space Mapping

The automated and unattended design of balanced microstrip wideband bandpass filters by means of aggressive space mapping (ASM) optimization is reported in this paper. The proposed filters are based on multisection mirrored stepped impedance resonators (SIRs) coupled through quarter-wavelength transmission lines, acting as admittance inverters. Such resonant elements provide transmission zeros useful for the suppression of the common mode in the region of interest (differential filter pass band) and for the improvement of the differential-mode stopband (rejection level and selectivity). Due to the limited functionality of the inverters, related to the wide fractional bandwidths, the automated filter design requires a two-step process. With the first ASM, the filter schematic satisfying the required specifications (optimum filter schematic) is determined. Then, the layout is synthesized by means of a second ASM algorithm. Both algorithms are explained in detail and are applied to the synthesis of two filters, as illustrative (and representative) examples. With this paper, it is demonstrated that the two-step ASM optimization scheme (first providing the optimum schematic and then the layout), previously applied by the authors to wideband single-ended filters, can be extended (conveniently modified) to common-mode suppressed differential-mode bandpass filters. Thus, the value of this two-step ASM approach is enhanced by demonstrating its potential for the unattended design of complex filters, as those considered in this paper.

Tree-Based Visualization and Optimization for Image Collection.

The visualization of an image collection is the process of displaying a collection of images on a screen under some specific layout requirements. This paper focuses on an important problem that is not well addressed by the previous methods: visualizing image collections into arbitrary layout shapes while arranging images according to user-defined semantic or visual correlations (e.g., color or object category). To this end, we first propose a property-based tree construction scheme to organize images of a collection into a tree structure according to user-defined properties. In this way, images can be adaptively placed with the desired semantic or visual correlations in the final visualization layout. Then, we design a two-step visualization optimization scheme to further optimize image layouts. As a result, multiple layout effects including layout shape and image overlap ratio can be effectively controlled to guarantee a satisfactory visualization. Finally, we also propose a tree-transfer scheme such that visualization layouts can be adaptively changed when users select different "images of interest." We demonstrate the effectiveness of our proposed approach through the comparisons with state-of-the-art visualization techniques.

A two-step optimization scheme for maximum stiffness design of laminated plates based on lamination parameters

The purpose of this paper is to propose an effective solution scheme of simultaneous optimization design of layup configuration and fiber distribution for maximum stiffness design of laminated plates. Firstly, a numerical analysis of the lamination parameters feasible region for a laminated plate consisting of various given number of ply groups (each ply group may have different thickness and all the fibers in one ply group are orientated in an identical direction) is carried out, and it is found that the feasible region based on only a few ply groups is very close to the overall one determined by infinite plies. Therefore, it is suggested that the feasible region of lamination parameters of a laminated plate could be approximately determined by the layup configuration of least ply groups. Secondly, a two-step simultaneous optimization scheme of layup configuration and fiber distribution for maximum stiffness design of laminated plates is proposed. Accordingly, by using ply thickness, fiber orientation angle and fiber volume fraction in a laminated plate of least ply groups as design variables, the optimal lamination parameters for maximum stiffness is obtained. Then, taking the optimal lamination parameters as the design objective, a detailed layup design optimization is implemented by considering some limitations on manufacturing, such as preset ply thickness, and specific fiber orientation angle and a limited maximum number of consecutive plies in the same fiber orientation. Numerical examples are also presented to validate the proposed two-step optimization scheme.

MMSE relay precoding schemes based on quantization error compensation for multi-relay system

In order to deal with the performance degradation resulting from the quantization error, this paper introduces the statistics of Grassmannian codebook’s quantization error into relay precoding design for multirelay system. Based on the two-step MMSE optimization algorithm for relay precoding and receiver equalizing with full feedback under QoS requirements, we first propose a limited feedback two-step MMSE optimization scheme for relay precoding and receiver equalizing with quantization error compensation, and then we propose a limited feedback joint MMSE optimization scheme for relay precoding and receiver equalizing with quantization error compensation subject to a global power constraint at the output of relays. Simulation results show that the proposed joint MMSE optimization scheme with quantization error compensation has a much lower bit-error rate than the two-step MMSE optimization scheme with quantization error compensation and the joint MMSE optimization scheme without quantization error compensation.

Feature Preserving Mesh Simplification Using Feature Sensitive Metric

We present a new method for feature preserving mesh simplification based on feature sensitive (FS) metric. Previous quadric error based approach is extended to a high-dimensional FS space so as to measure the geometric distance together with normal deviation. As the normal direction of a surface point is uniquely determined by the position in Euclidian space, we employ a two-step linear optimization scheme to efficiently derive the constrained optimal target point. We demonstrate that our algorithm can preserve features more precisely under the global geometric properties, and can naturally retain more triangular patches on the feature regions without special feature detection procedure during the simplification process. Taking the advantage of the blow-up phenomenon in FS space, we design an error weight that can produce more suitable results. We also show that Hausdorff distance is markedly reduced during FS simplification

Intra-retinal layer segmentation in optical coherence tomography images

Retinal layer thickness, evaluated as a function of spatial position from optical coherence tomography (OCT) images is an important diagnostics marker for many retinal diseases. However, due to factors such as speckle noise, low image contrast, irregularly shaped morphological features such as retinal detachments, macular holes, and drusen, accurate segmentation of individual retinal layers is difficult. To address this issue, a computer method for retinal layer segmentation from OCT images is presented. An efficient two-step kernel-based optimization scheme is employed to first identify the approximate locations of the individual layers, which are then refined to obtain accurate segmentation results for the individual layers. The performance of the algorithm was tested on a set of retinal images acquired in-vivo from healthy and diseased rodent models with a high speed, high resolution OCT system. Experimental results show that the proposed approach provides accurate segmentation for OCT images affected by speckle noise, even in sub-optimal conditions of low image contrast and presence of irregularly shaped structural features in the OCT images.

Defect reduction of GaAs/Si epitaxy by aspect ratio trapping

We report on the metallorganic chemical vapor deposition growth of GaAs on patterned Si (001) substrates, which utilizes the aspect ratio trapping method. It was found that when growing GaAs above the SiO2 trenched region, coalescence-induced threading dislocations and stacking faults originated on top of the GaAs/SiO2 interfaces. These defects were found to be indirectly related to the initial defect-trapping process during trenched GaAs growth. Causes of coalescence defect formation and its reduction were experimentally investigated by employing a two-step growth optimization scheme. Improvement of material quality has been characterized by cross-sectional and plan-view transmission electron microscopy and x-ray diffraction.

Three-dimensional shape reconstruction of abdominal aortic aneurysm

This paper presents an effective computational technique for reconstructing a three-dimensional shape of an abdominal aortic aneurysm (AAA), from a limited number of computed tomography (CT) images. The three-dimensional template geometry of a healthy abdominal aorta is used as a priori knowledge, and the template geometry is deformed by extended free-form deformation (EFFD), to generate a patient-specific AAA geometry. A two-step optimization scheme is devised to find an optimal set of EFFD parameters that match the cross-section of a deformed template with an AAA contour shown in a CT image. The geometric continuity of a deformed model is maintained by raising the order of the polynomial function used in EFFD. Experimental results show that the proposed method creates the three-dimensional shape of AAA suitable for structural finite element analysis and computational fluid dynamics for medical diagnosis.

Optimizing the Release Conditions for a Free Throw in Wheelchair Basketball

The primary purpose of this study was to determine the optimal release conditions and corresponding arm movement pattern for the free throw for players classified as 3 to 4.5 on the international player classification system in wheelchair basketball. A 2-D, three-segment simulation model was used to investigate this problem. The computational process involved a two-step optimization scheme in which an outer computational loop was used to optimize the magnitude and timing of the muscle torques that generate the arm's motion, and an inner computational loop was used to determine the optimal angle and speed of the ball at the moment of release. The inner optimization loop revealed that Brancazio's (1981) and Hay's (1993) approaches to determining the optimal release angle produced identical results. The lowered seated height of the wheelchair basketball player required that the ball be released at a steeper angle with greater vertical velocity, and hence the need for greater shoulder torque. For the wheelchair player, the peak shoulder flexion torque generated by the model was reduced by approximately 43% when the upper arm was initially positioned at an angle approximately 40° below the horizontal, as compared to being positioned at an angle of 10° above the horizontal. For the wheelchair player, the optimal release angle and speed for a ball released at a horizontal distance of 4.09 m from the center of the basket, and 1.30 m below the rim, was computed to be 53.8° and 7.4 m/s, respectively.