483 research outputs found
Robust Camera Pose Estimation via Consensus on Ray Bundle and Vector Field
No full textEstimating the camera pose requires point correspondences. However, in practice, correspondences are inevitably corrupted by outliers, which affects the pose estimation. We propose a general and accurate outlier removal strategy for robust camera pose estimation. The proposed strategy can detect outliers by leveraging the fact that only inliers comply with two effective consensuses, i.e., 3D ray bundle consensus and 2D vector field consensus. Our strategy has a nested structure. First, the outer module utilizes the 3D ray bundle consensus. We define the likelihood based on the probabilistic mixture model and maximize it by the expectation-maximization (EM) algorithm. The inlier probability of each correspondence and the camera pose are determined alternately. Second, the inner module exploits the 2D vector field consensus to refine the probabilities obtained by the outer module. The refinement based on the Bayesian rule facilitates the convergence of the outer module and improves the accuracy of the entire framework. Our strategy can be integrated into various existing camera pose estimation methods which are originally vulnerable to outliers. Experiments on both synthesized data and real images have shown that our approach outperforms state-of-the-art outlier rejection methods in terms of accuracy and robustness
Wu xian wang luo zhong de xing wei jing ji xue fen xi
No full textPh.D.The global mobile data traffic has been growing tremendously in the past few years. In order to alleviate the tension between the mobile data demand and the network capacity, the operator can either utilize the spectrum more efficiently, or flatten the demand curve through pricing. Motivated by some recent practices, this thesis focuses on two specific problems. First, we study the cognitive mobile virtual network operator’s spectrum investment problem. Second, we study the mobile network operator’s optimization problems and users’ behaviors in a mobile data trading market. To understand the decisionmakers’ realistic behaviors, our solutions to these two problems focus on not only classical economic perspectives, where decision makers are assumed to be fully-rational, but also behavioral economic perspectives, which captures the realistic human decision process on uncertainty.In the first part of the thesis, we study a cognitive mobile virtual network operator’s spectrum investment problem under spectrum supply uncertainty using prospect theory. We formulate a hybrid spectrum investment problem including spectrum sensing and leasing as a two-stage optimization problem, and compute the optimal sensing and leasing decisions. Our results show that compared with a fully-rational operator, both the risk-averse and risk-seeking operators achieve a smaller expected profit. On the other hand, a risk-averse operator can guarantee a larger minimum possible profit, while a risk-seeking operator can achieve a larger maximum possible profit.In the second part of the thesis, we first analyze the optimal trading strategy for a single user in the mobile data trading market under prospect theory, to understand the users’ realistic trading behaviors with risk preferences. Building upon our analysis, we design an algorithm to help estimate the user’s risk preference and provide trading recommendations dynamically, considering the latest market and usage information. We also study the operator’s optimal operation fee, and the scenarios under which it is beneficial for an operator to propose a mobile data trading market. Our results suggest an operator with a low initial market share to propose a data trading market.全球移動數據流量在過去幾年中大幅增長。為了緩解移動數據需求與網絡容量之間的緊張關係，運營商可以通過更有效地利用頻譜的方式，或者可以通過由定價來平坦化需求曲線的方式。本論文著重於研究兩類具體問題。一，認知移動網絡虛擬運營商的頻譜投資問題。二，移動網絡運營商在移動數據交易市場中的優化問題和用戶在移動數據交易市場中的行為。為了瞭解決策者的現實行為，我們的方法不僅側重於經典經濟學理論觀點，即決策者被認為是完全理性的，而且還包括行為經濟觀點，從而捕捉現實決策過程中面對不確定性時決策者的行為特徵。在論文的第一部分，我們使用前景理論研究了認知移動網絡虛擬運營商面對頻譜供應不確定性時的頻譜投資問題。我們建模了一個包括頻譜感知和頻譜租賃的兩階段混合頻譜投資問題，並計算出最優的頻譜感知和頻譜租賃的決策。我們的研究結果表明，與完全理性的運營商相比，風險厭惡和風險尋求型的運營商都會得到較小的預期利潤。另一方面，風險厭惡型的運營商可以保證更大的最小可能利潤，而風險尋求型的運營商可以獲得更大的最大可能利潤。在論文的第二部分，我們首先分析移動數據交易市場中單一用戶在前景理論下的最優交易策略，從而瞭解現實交易中，風險偏好對用戶行為的影響。基於我們的分析，我們設計了一種交易算法來幫助估計用戶的風險偏好，並基於最新的市場信息和用戶數據用量信息，提供動態交易建議。我們同時還研究運營商的最優運營費用的決策，以及使得運營商可以依靠推出移動數據交易市場而獲益的場景。研究結果表明，初始市場份額較低的運營商可以靠推出數據交易市場而獲益。Yu, Junlin.Thesis Ph.D. Chinese University of Hong Kong 2017.Includes bibliographical references (leaves 154-162).Abstracts also in Chinese.Title from PDF title page (viewed on 22, August, 2019).Yu, Junlin
Trustworthy Autonomy Through Robust Control and Alignment
No full textAs artificial intelligence systems are increasingly applied in safety critical domains such as robotics, autonomous driving, and decision making under uncertainty, ensuring their trustworthiness has become a central challenge. This dissertation addresses two major facets of trustworthy AI: reliable control through formal guarantees and alignment against adversarial manipulation. The first part of the dissertation focuses on provably stable, robust, and safe control for nonlinear systems using learning based methods. We introduce the first general framework for synthesizing neural Lyapunov controllers in discrete time systems. This method combines a sound verifier based on mixed integer linear programming with gradient based counterexample generation to efficiently learn control policies that satisfy formal stability conditions. We extend this framework to adversarial settings where state observations are perturbed, developing verification and training techniques that produce controllers robust to both persistent and intermittent attacks. We then propose a method for verified safe reinforcement learning in neural dynamical systems using finite horizon reachability and curriculum learning. Our approach achieves strong safety guarantees across multiple benchmarks while preserving task performance. To improve robustness in environments with high dimensional perceptual inputs, we develop a novel curriculum based adversarial training framework that significantly enhances deep reinforcement learning against large input perturbations. Finally, we introduce a partially supervised reinforcement learning framework that enables safety certification in partially observable environments by leveraging access to interpretable low dimensional states during training. The second part of the dissertation investigates how AI systems that learn from human preferences can be manipulated. We model election control through voter perception manipulation using spatial voting theory and characterize the computational complexity of such attacks under various assumptions. Building on this insight, we explore preference poisoning attacks on reward models, a core component of value aligned AI systems including reinforcement learning from human feedback. We develop and evaluate both gradient based and heuristic attacks that show high success even with minimal data poisoning across domains such as autonomous control and large language model alignment. Together, these contributions offer principled methods for building AI systems that are stable, robust, safe, and aligned with human values, laying a foundation for future progress in trustworthy autonomy
Non-equilibrium wall-bounded turbulence and associated noise generation
No full textAbstract : The present study investigates the response of turbulence in a non-equilibrium flows such as transient periodic channel flows and spatially developing boundary layers subjected to pressure gradients. Such a fundamental study is important to understand noise generation in complex wall-bounded turbulent flows. First, to understand the flow dynamics in transient accelerating flows, direct numerical simulations (DNS) of periodic channel flows responding to an impulse acceleration are carried out. The turbulent flow undergoes reverse transition toward a quasi-laminar state, followed by a retransition phase to the new equilibrium state. To reduced simulation cost, the minimal-span methodology is applied and evaluated for simulations of transient flows. Detailed comparisons with a full-span case show that the small-span test case captures the essential dynamics during the transition process despite small, quantitative differences attributed to a slower streak transient growth. A small span is used to characterize accelerating channels with riblets. Results indicate that riblets delay the transition to high Reynolds number state, as it reduces streak meandering. Next, to study non-equilibrium boundary layer flows in the presence of convex wall curvature, DNS simulations over an airfoil (suction side) and a flat plate are compared. Both cases are characterized by matching adverse pressure gradient (APG) along the streamwise direction. For the airfoil boundary layer, existing DNS data obtained by \cite{wu2019effects} of flow around a controlled-diffusion (CD) airfoil is used. For the flat-plate boundary layer, a DNS simulation is carried out, with prescribed pressure gradient distribution that matches that of the airfoil flows in the APG region. Comparison between the two cases shows how the wall curvature affects turbulence in an APG boundary layer, important in industrial applications such as fan flows. Overall, the comparison shows that the boundary layer developments are very similar. This indicates that a flat-plate boundary layer can serve as a low-cost surrogate of an airfoil boundary layer in numerical studies of important features of an airfoil flow. The difference between the two cases represents the effect of a mild convex wall curvature. Specifically, in the region of weak APG, the curvature effect dominates that of the pressure gradient and yields a lower friction coefficient. In high-APG regions (near the trailing edge of the airfoil) the effects of wall curvature and APG appear to interact. Lastly, various existing analytical models are evaluated on their predictions of wall pressure fluctuations, which are essential for noise prediction in non-equilibrium boundary layer turbulent flows that develop on fan blades. Limitations of the existing models are evaluated; new parameters that do not involve the ill-defined wall friction in a boundary layer under strong adverse pressure gradients are proposed. The primary role of the mean velocity logarithmic layer in affecting the overlap range of the wall pressure spectrum is also demonstrated. A new wall pressure spectrum model is proposed and tested in a wide range of boundary layer flows under different Reynolds numbers and zero, adverse and favorable pressure gradients. The test database includes existing experimental data and various DNS flat-plate simulations. The new wall pressure spectrum model is the first generalized model designed for boundary layer flows with a wide range of pressure gradients and Reynolds numbers.Ce mémoire étudie la réponse de la turbulence dans des écoulements hors équilibre, tels que les écoulements transitoires dans un canal périodique et les couches limites se développant spatiallement soumises à des gradients de pression. Une telle étude fondamentale est importante pour comprendre la génération du bruit dans des écoulements complexes turbulents. Premièrement, pour comprendre la dynamique d’écoulements transitoires soumis à une accélération, des simulations directes d’écoulements instationnaires dans un canal périodique soumis à une accélération impulsionnelle ont été réalisées. L’écoulement turbulent subit une transition inversée vers un état quasi-laminaire, suivi par une nouvelle phase de transition vers un nouvel équilibre. Pour réduire le coût de calcul, la méthode de l’envergure minimale du domaine de calcul est appliquée et validée pour de telles simulations instationnaires. Des comparaisons détaillées avec un cas d’envergure complète montrent que la simulation avec une envergure minimale capture l’essentiel de la dynamique de l’écoulement durant la phase de transition et ce malgré quelques petites différences attribuées à la croissance plus lente des tourbillons longitudinaux le long de la paroi (“streaks”). Une envergure réduite est ensuite appliquée à l’étude d’un écoulement accéléré dans un canal avec de micro-sillons ou “riblets”. Les résultats montrent que les riblets retardent la transition du fait qu’ils stabilisent la turbulence de proche paroi. Deuxièmement, pour étudier les couches limites hors équilibre sur une paroi convexe, des simulations directes sur l’extrados d’un profil aérodynamique et d’une plaque plane sont comparées. Les deux cas sont caractérisés par le même gradient de pression adverse dans la direction de l’écoulement. Pour la couche limite sur le profil, on utilise les données existantes de la simulation directe de Wu et al. (2019) autour du profil à diffusion controllée (CD). Pour la couche limite sur la plaque plane, une nouvelle simulation directe a été réalisée avec le même gradient de pression adverse que sur le profil. La comparaison des deux cas montre que la courbure de la paroi convexe peut modifier la turbulence dans une couche limite soumise à un gradient de pression adverse qui est important dans les applications industrielles comme les écoulements dans des ventilateurs. Cependant les modifications restent mineures et la comparaison montre que le développement des couches limite turbulentes dans les deux cas est semblable. Ceci implique que la couche limite sur une plaque plaque sur un domaine réduit peut servir de substitut à celle sur un profil aérodynamique qui requiert un domaine plus grand et des ressources de calcul plus importante. La différence observée entre les deux cas permet d’évaluer l’effet d’une paroi faiblement convexe. Spécifiquement, dans la région de faible gradient de pression adverse, les effets de courbure dominent ceux du gradient de pression et réduisent le coefficient de frottement pariétal. Dans les zones de fort gradient de pression adverse, près du bord de fuite, les effets de gradient de pression et de courbure interagissent. Finalement, la dernière étape a été d’évaluer les différents modèles analytiques de fluctuations de pression pariétale qui sont au centre des prédictions de bruit dans les couches limites turbulentes hors équilibre qui se développent sur les pales de ventilateurs. Les limites des modèles précédents sont évaluées et de nouveaux paramètres ne faisant pas intervenir le frottement pariétal mal défini dans une couche limite à fort gradient de pression adverse sont proposés. Le rôle primordial de la zone logarithmique dans la couche limite turbulente sur le gabarit spectral des spectres de pression pariétale est aussi mis en évidence. Le nou veau modèle de spectre de pression pariétale est ensuite testé sur plusieurs couches limites attachées avec des gradients de pression favorables, adverses, et des écoulements décollés à divers nombres de Reynolds basés sur l’épaisseur de quantité de mouvement. Les données proviennent de bases de données expérimentales et numériques existantes. Des simulations directes supplémentaires ont également été réalisées pour étendre les résultats numériques (notamment sur le profil CD) à des nombres de Reynolds plus élevés. Pour la première fois, un modèle est capable de reproduire les spectres de pression pariétale pour tous ces types d’écoulement
Determination of rare earths and other trace elements in samples of Antarctica by neutron activation analysis
No full textThermal Error Modelling and Compensation of The Voice Coil Motor Force for Hydraulic Valves
No full textDeep and Shallow Feature Fusion in Feature Score Level for Palmprint Recognition
No full textContactless palmprint recognition offers friendly customer experience due to its ability to operate without touching the recognition device under rigid constrained conditions. Recent palmprint recognition methods have shown promising accuracy; however, there still exist some issues that need to be further studied such as the limited discrimination of the single feature and how to effectively fuse deep features and shallow features. In this paper, deep features and shallow features are integrated into a unified framework using feature-level and score-level fusion methods. Specifically, deep feature is extracted by residual neural network (ResNet), and shallow features are extracted by principal component analysis (PCA), linear discriminant analysis (LDA), and competitive coding (CompCode). In feature-level fusion stage, ResNet feature and PCA feature are dimensionally reduced and fused by canonical correlation analysis technique to achieve the fused feature for the next stage. In score-level fusion stage, score information is embedded in the fused feature, LDA feature, and CompCode feature to obtain a more reliable and robust recognition performance. The proposed method achieves competitive performance on Tongji dataset and demonstrates more satisfying generalization capabilities on IITD and CASIA datasets. Comprehensive validation across three palmprint datasets confirms the effectiveness of our proposed deep and shallow feature fusion approach
Production trade-offs in data envelopment analysis models with ratio inputs and outputs
No full textData envelopment analysis (DEA) is a well-established method to evaluate the performance of a homogeneous set of organisations for benchmarking purposes. The standard constant and variable returns-to-scale (CRS and VRS) models (Charnes et al., 1978; Banker et al., 1984) were developed based on volume measures (i.e., absolute numbers) and the underlying production axioms were assumed on volumes too. Ratio measures (e.g., percentages, rates, and averages) are often used in DEA applications to represent quality indicators; however, they can violate some of the underlying production axioms that the standard CRS and VRS models assume, e.g., the convexity axiom, which presumes the production set is convex (Emrouznejad and Amin, 2009). Such violations can result in incorrect efficiency results. Olesen et al. (2015) developed ratio-VRS (R-VRS) and ratio-CRS (R-CRS) models to allow ratio measures (as their native types of data) to be used with volume measures simultaneously in the DEA models. However, these models have not been used in real-data applications. In this dissertation, we use an application to a large sample of secondary schools and a Monte Carlo simulation to examine the differences between using the ratio DEA (i.e., R-VRS and R-CRS) and the standard DEA models. Our results show that although the ratio DEA models correct the issues of using ratio measures, the cost is to obtain lower efficiency discrimination. To improve the discriminating power of the ratio DEA models, we extend the specification of production trade-offs to them and develop new models. Subsequently, we use the foregoing school application and incorporate seven production trade-offs identified between volume measures and ratio measures into the efficiency computation. The application results show that the incorporation of production trade-offs significantly improves efficiency discrimination.</p
Transit Operations with Deterministic Optimal Fare and Frequency Control
No full textAbstractFare and frequency determination is essential in transit operations. In this paper, passenger demand is analysed and embodied under distributed passengers’ willingness to pay. The transit operator's objective is defined as a weighted combination of operator's profit and consumer surplus. The closed-form optimal solutions of transit fare and frequency are obtained in the aim of operator's objective. Different cost structures of the transit system are taken into consideration concerning the marginal effect of passenger demand on transit operating cost. The optimal fare and frequency control strategies are then derived in response to a deterministic changing environment where demand fluctuations are caused by the changes of exogenous factors. For stable or gradually varying situations of exogenous factors, they are determined by implicit equations based on optimal solutions; while for abrupt changing situations, they are determined by the gradient field of the transit operator's objective. A case study for daily optimal operation control of a rapid transit service is demonstrated
- …
