In this paper, we propose a Point Fractal Network (PF-Net), novel learning-based approach for precise and high-fidelity point cloud completion. Unlike existing completion networks, which generate the overall shape of from incomplete always change points encounter noise geometrical loss, PF-Net preserves spatial arrangements can figure out detailed structure missing region(s) in prediction. To succeed at task, estimates hierarchically by utilizing feature-points-based multi-scale generating...

The crux of semi-supervised semantic segmentation is to assign adequate pseudo-labels the pixels unlabeled images. A common practice select highly confident predictions as pseudo ground-truth, but it leads a problem that most may be left unused due their unreliability. We argue every pixel matters model training, even its prediction ambiguous. Intuitively, an unreliable get confused among top classes (i.e., those with highest probabilities), however, should about not belonging remaining...

This work studies the problem of few-shot object counting, which counts number exemplar objects (i.e., described by one or several support images) occurring in query image. The major challenge lies that target can be densely packed image, making it hard to recognize every single one. To tackle obstacle, we propose a novel learning block, equipped with similarity comparison module and feature enhancement module. Concretely, given image first derive score map comparing their projected features...

This paper introduces a perceptual model for determining 3D printing orientations. Additive manufacturing methods involving low-cost printers often require robust branching support structures to prevent material collapse at overhangs. Although the designed shape can successfully be made by adding supports, residual remains contact points after supports have been removed, resulting in unsightly surface artifacts. Moreover, fine details on fabricated easily damaged while removing supports. To...

In this paper, a distributed neurodynamic approach is proposed for constrained convex optimization. The objective function sum of local subproblems, whereas the constraints these subproblems are coupled. Each minimized individually with optimization approach. Through information exchange between connected neighbors only, all nodes can reach consensus on Lagrange multipliers global equality and inequality constraints, decision variables converge to optimum in manner. Simulation results two...

Despite the rapid advance of unsupervised anomaly detection, existing methods require to train separate models for different objects. In this work, we present UniAD that accomplishes detection multiple classes with a unified framework. Under such challenging setting, popular reconstruction networks may fall into an "identical shortcut", where both normal and anomalous samples can be well recovered, hence fail spot outliers. To tackle obstacle, make three improvements. First, revisit...

This paper presents a neurodynamic optimization approach to bilevel quadratic programming (BQP). Based on the Karush-Kuhn-Tucker (KKT) theorem, BQP problem is reduced one-level mathematical program subject complementarity constraints (MPCC). It proved that global solution of MPCC minimal one optimal solutions multiple convex subproblems. A recurrent neural network developed for solving these From any initial state, state proposed convergent an equilibrium point network, which just...

Welding is an important joining technology but the defects in welds wreck quality of product evidently. Due to variety weld defects' characteristics, defect detection a complex task industry. In this paper, we try explore possible solution for and novel image-based approach proposed using small X-ray image data sets. An image-processing based augmentation WGAN are applied deal with imbalanced Then train two deep convolutional neural networks (CNNs) on augmented sets feature-extraction...

In this article, a discrete-time distributed optimization algorithm is proposed for solving the economic dispatch (ED) problem with some groups of generator units to communicate over connected graph, which independent power system. The ED converted an objective sum individual convex functions and constraints local generators. Based on optimal conditions, class algorithms designed find solution problem. can be realized as multiagent system whose convergence proved using dynamic analysis...

This paper presents a neurodynamic optimization approach to robust pole assignment for synthesizing linear control systems via state and output feedback. The problem is formulated as pseudoconvex with robustness measure: i.e., the spectral condition number objective function matrix equality constraints exact assignment. Two coupled recurrent neural networks are applied solving in real time. In contrast existing approaches, exponential convergence of proposed neurodynamics global optimal...

This paper presents a two-time-scale neurodynamic approach to constrained minimax optimization using two coupled neural networks. One of the recurrent networks is used for minimizing objective function and another maximization. It shown that systems operating in different time scales work well optimization. The effectiveness characteristics proposed are illustrated several examples. Furthermore, applied H∞ model predictive control.

This paper presents a collective neurodynamic approach to robust model predictive control (MPC) of discrete-time nonlinear systems affected by bounded uncertainties. The proposed law is combination an MPC within invariant tube for nominal system and ancillary state feedback control. first transformed linear parameter-varying (LPV) system, then its signal computed solving convex optimization problem sequentially in real time using two-layer recurrent neural network (RNN). obtained means gain...

A neurodynamics-based algorithm is developed in this paper for solving multicoupled distributed optimization problems. In formulation, each agent solves a local problem with regard to its own cost function. The objective function sum of convex subproblems, which are not necessarily strict and smooth. With communication between neighbors only, decision variables lie manner succeed converge the global optimum. proposed method suitable large-scale problems energy Internet management thanks...

Abstract To deploy deep neural networks to edge devices with limited computation and storage costs, model compression is necessary for the application of learning. Pruning, as a traditional way compression, seeks reduce parameters weights. However, when network pruned, accuracy will significantly decrease. The decrease loss fine-tuning. When over many are pruned network’s capacity reduced heavily cannot recover high accuracy. In this paper, we apply knowledge distillation strategy abate...

Document Information Extraction (DIE) aims to extract structured information from Visually Rich Documents (VRDs). Previous full-training approaches have demonstrated strong performance but may struggle with generalization unseen data. In contrast, training-free methods leverage powerful pre-trained models like Large Language Models (LLMs) address various downstream tasks only a few examples. Nonetheless, for DIE encounter two primary challenges: (1) understanding the complex relationship...

A collective neurodynamic system is presented to distributed convex optimization subject linear equality and box constraints in framework of an autonomous multiagent network. The overall objective be minimized takes additive form multiple local functions. Agents the system, each which modeled by a recurrent neural network, cooperatively, autonomously develop their dynamic behaviors based on realtime interactions with neighbors. In specific, network has knowledge only no access function. It...

Increasing penetration of distributed generators and flexible loads in a distribution power system triggers the need for transactive mechanisms. This paper proposes mechanism model that allow noncontrollable resources to trade their deviations from schedules with elastic ones. The proposed can reveal cost uncertainty scarcity flexibility. is optimized fully manner which welfare each prosumer maximized individually through information exchange between neighbors. A neurodynamic algorithm...

Distributed algorithms are gaining increasing research interests in the area of power system optimization and dispatch. Existing distributed dispatch (DPDAs) usually assume that suppliers/consumers bid truthfully. However, this article shows need for DPDAs to consider strategic players take account their behavior deviation from what expect. To address this, we propose a strategy update algorithm (DSUA) on top DPDA. The DSUA considers suppliers who optimize bids DPDA, using only information...