76 research outputs found
Majorization-Minimization Techniques and Applications in Optimization and Statistical Learning.
No full textAn Integrated Graph Model for Spatial–Temporal Urban Crime Prediction Based on Attention Mechanism
No full textCrime issues have been attracting widespread attention from citizens and managers of cities due to their unexpected and massive consequences. As an effective technique to prevent and control urban crimes, the data-driven spatial–temporal crime prediction can provide reasonable estimations associated with the crime hotspot. It thus contributes to the decision making of relevant departments under limited resources, as well as promotes civilized urban development. However, the deficient performance in the aspect of the daily spatial–temporal crime prediction at the urban-district-scale needs to be further resolved, which serves as a critical role in police resource allocation. In order to establish a practical and effective daily crime prediction framework at an urban police-district-scale, an “online” integrated graph model is proposed. A residual neural network (ResNet), graph convolutional network (GCN), and long short-term memory (LSTM) are integrated with an attention mechanism in the proposed model to extract and fuse the spatial–temporal features, topological graphs, and external features. Then, the “online” integrated graph model is validated by daily theft and assault data within 22 police districts in the city of Chicago, US from 1 January 2015 to 7 January 2020. Additionally, several widely used baseline models, including autoregressive integrated moving average (ARIMA), ridge regression, support vector regression (SVR), random forest, extreme gradient boosting (XGBoost), LSTM, convolutional neural network (CNN), and Conv-LSTM models, are compared with the proposed model from a quantitative point of view by using the same dataset. The results show that the predicted spatial–temporal patterns by the proposed model are close to the observations. Moreover, the integrated graph model performs more accurately since it has lower average values of the mean absolute error (MAE) and root mean square error (RMSE) than the other eight models. Therefore, the proposed model has great potential in supporting the decision making for the police in the fields of patrolling and investigation, as well as resource allocation.Safety and Security Scienc
Disruptive Technologies: Legal and Insurance Implications in Shipping
No full textDisruptive technologies could, potentially, have an immense impact on shipping in the near future, especially when Maritime Autonomous Surface Ships (MASS) are introduced into commercial shipping. However, such technologies are also changing the way conventional ships and ports operate. These changes in shipping are a concern for insurers and there is a debate as to how insurance law and practice need to change to ensure that risks are appropriately assessed, and insurance policies are adjusted. This thesis intends to elaborate on i) the impact of such technologies on traditional marine insurance policies; and ii) new solutions that need to be implemented (how traditional legal doctrines and practices need to be amended). The author is of the opinion that such risks are still insurable but some fundamental changes in insurance law and practice will be needed in the years to come. The primary purpose of this thesis is to analyse parts of standard insurance clauses that need to be amended in particular to be able to offer effective insurance for ships that utilize disruptive technologies. The thesis will also consider the changes in risk allocation that might follow and how such risks could be reallocated in light of traditional insurance principles and doctrines. The thesis will also consider how the use of disruptive technologies will affect port operations and the liabilities that might emerge as a result with specific reference to the insurance position
Robust and Holistic Perception for Autonomous Vehicles in the Urban Scene
No full textAutonomous vehicle is one of the most promising direction and key application areas of artificial intelligence. The success of autonomous vehicles in urban scene heavily relies on the ability to handle the complex environments, where the accurate and robust perception is the foundation.To achieve the holistic and accurate perception, autonomous vehicles are equipped with various sensors, including camera, radar and LiDAR, in which LiDAR is considered as the most critical one as it can provide the accurate depth information. How to effectively and efficiently cope with LiDAR point cloud remains an open problem. On the other hand, to maintain the robustness of perception when facing various weather and environment, large-scale labeled data with appropriate variance is required, where Computer Graphics offer an alternative solution to address the data issue. However, how to handle the domain gap between real-world data and synthetic data is also challenging. In this thesis, we aim to establish the holistic and robust perception for autonomous vehicles from two perspectives, \ie, effective and efficient 3D perception from LiDAR point cloud and robust scene understanding based on the domain adaptation.Specifically, we first investigate the natural properties of current LiDAR sensors on the autonomous vehicle, \ie, sparsity and varying density, where regions that are far away from the origin have much sparse points. Based on this finding, we propose a new framework, which maintain the 3D geometric information and handle these issues from partition and networks, respectively. We then propose a LiDAR-based 3D detection method, where it is first time to introduce the shape information into the multi-class LiDAR detection and a well-designed shape signature is proposed to extract the shape embedding. Since the sequential point cloud is a real-world data form, we further extend these single-scan perception methods to multi-scan perception, where a novel method is introduced to explore the motion information. For the robust perception against domain shift, including different locations and weathers, we first give a deep analysis for the domain adaptation for object detection and reveal a crucial aspect to the success of object detector adaptation, namely, the focus to local regions when bridging domain gaps. For the domain adaptation for semantic segmentation, we propose a Conservative Loss to learn the domain invariant features.隨著人工智能技術的發展，越來越多的產業都在走向智能化，其中自動駕駛汽車是人工智能技術應用中最重要也是最有前景的部分之一。自動駕駛汽車的成功離不開汽車對於周圍複雜環境的準確高效且魯棒的感知。為了實現全面且準確的感知，自動駕駛汽車往往準備了多種不同的傳感器，包括攝像頭，毫米波雷達和激光雷達，其中激光雷達因為能夠提供準確的深度信息，被認為是最重要的傳感器。但是如何更高效的利用激光雷達的點雲數據依然是一個沒有被解決的問題。另一方面，自動駕駛汽車需要面臨各種各樣不同的場景和天氣條件，這就對於數據有著更高的要求，需要數據能夠盡可能的滿足不同場景和天氣要求，才能訓練出魯棒通用的感知模型。計算機圖形學中的渲染技術提供了這方面的技術可行性，但是本身渲染技術產生的數據與真實環境下的數據存在的域差異，如何克服這種域差異就成了實現魯棒模型的一個關鍵點。在本文中，我們從兩個方向來探索實現高效全面且魯棒的感知算法，一是通過使用精準的激光雷達來實現三維感知，二是運用域遷移技術來實現渲染數據到真實數據的轉換，進而實現魯棒通用的感知。首先，我們仔細觀察了室外激光雷達的分佈屬性，存在稀疏性和近密遠疏的特性，基於這一發現，我們提出了一個基於圓柱體坐標系的劃分方法，在保持了三維結構的前提下，運用不同大小的扇面來劃分整個空間，進而滿足近密遠疏的特性。我們將上述方法進一步的擴展到了點雲全景分割，點雲檢測等任務上。然後，我們提出了一種基於激光雷達點雲的三維檢測算法，第一次將物體的形狀信息引入到三維物體檢測模型中，提出了一個形狀描述子來提取不同物體的形狀信息。因為在真實場景下，我們的數據都是以連續幀的形式存在的，所以我們進一步提出了一個建模連續幀的方法，通過連續幀之間的位置關係來建模運動信息。另一方面，對於在不同天氣，不同環境的情況下的魯棒通用感知算法，我們分別針對物體檢測和場景分割兩個任務，提出了兩種解決方案，對於物體檢測，我們利用感興趣區域對齊的方式來實現檢測的域遷移，而針對場景分割，則是提出了一個居中損失函數來學習具有域不變性的特征表示。ZHU, Xinge.Ph.D. Chinese University of Hong Kong 2021.Includes bibliographical references (leaves )Abstracts in English and Chinese.Title from PDF title page (viewed on ...
Application of Blockchain Technology in Online Education
No full textBlockchain is a data structure of data blocks arranged in chronological order. It is featured by decentralization, trustworthiness, data sharing, security, etc. It has been widely used in digital currency, smart contract, credit encryption and other fields. With the development of the Internet technology, online education, a novel education mode, has been greatly popularized. However, this education mode still faces many problems in course credibility, credit and certificate certification, stu-dent privacy, and course sharing. Through literature review and case analysis, this paper discusses the basic technical principles and application features of blockchain technology, and proposes a solution to the problems of online educa-tion based on blockchain technology. The blockchain technology can store learn-ing records in a trusted, distributed manner, provide credible digital certificates, realize learning resource sharing with smart contract, and protect intellectual property through data encryption. The research shows that the integration of blockchain technology is a promising trend in the development of online educa-tion
Feature-aligned Surface Parameterization Using Secondary Laplace Operator and Loop Subdivision
No full textAbstractSurface parameterization is of great importance for many applications such as quadrangulation, texture mapping and surface fitting. An important issue for surface parameterization is how to align parametric lines with feature directions. To address this issue, in this paper we first utilize Loop subdivision basis functions and isogeometric analysis (IGA) to calculate eigenfunctions of the secondary Laplace operator (SLO) on triangle meshes. Eigenfunctions are then used for centroidal Voronoi tessellation (CVT) based surface segmentation, and boundaries of the segmented regions are extracted as feature lines which contain concave creases and convex ridges. Along each feature line, adjacent triangles are defined as guidance triangles to parameterize the surface using a constrained cross field method, where feature lines are preserved and aligned to parametric lines. Several examples are presented in the end to verify the robustness of our algorithm
Discrepancies in performance for heterojunction organic field-effect transistors with different channel lengths
No full textForecasting Human Core and Skin Temperatures: A Long-Term Series Approach
No full textHuman core and skin temperature (Tcr and Tsk) are crucial indicators of human health and are commonly utilized in diagnosing various types of diseases. This study presents a deep learning model that combines a long-term series forecasting method with transfer learning techniques, capable of making precise, personalized predictions of Tcr and Tsk in high-temperature environments with only a small corpus of actual training data. To practically validate the model, field experiments were conducted in complex environments, and a thorough analysis of the effects of three diverse training strategies on the overall performance of the model was performed. The comparative analysis revealed that the optimized training method significantly improved prediction accuracy for forecasts extending up to 10 min into the future. Specifically, the approach of pretraining the model on in-distribution samples followed by fine-tuning markedly outperformed other methods in terms of prediction accuracy, with a prediction error for Tcr within ±0.14 °C and Tsk, mean within ±0.46 °C. This study provides a viable approach for the precise, real-time prediction of Tcr and Tsk, offering substantial support for advancing early warning research of human thermal health
- …
