Search

With the massive amount of data generated from mobile devices and the increase of computing power of edge devices, the paradigm of Federated Learning has attracted great momentum. In federated learning, distributed and heterogeneous nodes collaborate to learn model parameters. However, while providing benefits such as privacy by design and reduced latency, the heterogeneous network present challenges to the synchronisation methods, or barrier control methods, used in training, regarding system progress and model convergence etc. The design of these barrier mechanisms is critical for the performance and scalability of federated learning systems. We propose a new barrier control technique called Probabilistic Synchronous Parallel (PSP). In contrast to existing mechanisms, it introduces a sampling primitive that composes with existing barrier control mechanisms to produce a family of mechanisms with improved convergence speed and scalability. Our proposal is supported with a convergence analysis of PSP-based SGD algorithm. In practice, we also propose heuristic techniques that further improve the efficiency of PSP. We evaluate the performance of proposed methods using the federated learning specific FEMNSIT dataset. The evaluation results show that PSP can effectively achieve good balance between system efficiency and model accuracy, mitigating the challenge of heterogeneity in federated learning.

In this article, a novel adaptive trajectory tracking controller is designed for a payload-carrying spacecraft under full state constraints. The proposed controller can tackle state-dependent uncertainties without a priori knowledge of their structures and upper bounds. The controller ensures time-varying constraints on all states and their time derivatives. The closed-loop stability of the proposed scheme is verified analytically via the Lyapunov method, and real-life experiments using a robotic testbed validated the effectiveness of the proposed adaptive controller over the state-of-the-art. Validerad;2023;Nivå 2;2023-04-17 (hanlid)

Lane keeping control needs to achieve smooth steering operation while ensuring safety-no lane departures and maintain small lateral displacement. In this paper, we solve this challenge by developing a preview lane keeping control supervised by a safeguard controller. The preview control utilizes both tracking errors and future lane curvatures to generate the optimal steering commands. The safeguard controller is then designed to guarantee bounded tracking errors. It supervises the preview control and intervenes if and only if the tracking error is approaching the safety boundary. Both algorithms have analytical control laws and thus require little on-line computations. The integrated system is compared with a model predictive control (MPC) design in terms of both tracking performance and computing efficiency. We implemented the proposed controls on a self-driving vehicle platform and tested on open roads and in the Mcity test facility.

Submitted by Ana Maria de Sousa Rebelo (amsr@isep.ipp.pt) on 2021-09-06T11:26:23Z No. of bitstreams: 1 DM_JorgeBeato_2021_MEEC.pdf: 5708089 bytes, checksum: 1c649f1bfbb68090ccbf555e5eba0455 (MD5) Approved for entry into archive by Ana Maria de Sousa Rebelo (amsr@isep.ipp.pt) on 2021-09-06T11:26:55Z (GMT) No. of bitstreams: 1 DM_JorgeBeato_2021_MEEC.pdf: 5708089 bytes, checksum: 1c649f1bfbb68090ccbf555e5eba0455 (MD5) Made available in DSpace on 2021-09-06T11:26:55Z (GMT). No. of bitstreams: 1 DM_JorgeBeato_2021_MEEC.pdf: 5708089 bytes, checksum: 1c649f1bfbb68090ccbf555e5eba0455 (MD5) Previous issue date: 2021

The paper proposes a sensors platform to control a barrier that is installed for vehicles entrance. This platform is automatized by image-based license plate recognition of the vehicle. However, in situations where standardized license plates are not used, such image-based recognition becomes non-trivial and challenging due to the variations in license plate background, fonts and deformations. The proposed method first detects the approaching vehicle via ultrasonic sensors and, at the same time, captures its image via a camera installed along with the barrier. From this image, the license plate is automatically extracted and further processed to segment the license plate characters. Finally, these characters are recognized with the help of a standard optical character recognition (OCR) pipeline. The evaluation of the proposed system shows an accuracy of 98% for license plates extraction, 96% for character segmentation and 93% for character recognition.

Yüksek Lisans Tezi Yapay zeka, insan zekasına özgü olan, algılama, öğrenme, fikir yürütme, sorun çözme gibi yüksek bilişsel fonksiyonları veya otonom davranışları sergilemesi beklenen ve düşüncelerinden tepkiler üretebilen bir sistem olarak tanımlanmaktadır. Plaka tanıma uygulamaları genellikle kontrol ve güvenlik amacıyla kullanılan, ihtiyaç duyulan bir sistemdir. Mevcut teknolojik gelişmelerin ışığında insansız olarak yapay zeka kullanılarak geliştirilecek sistem, trafik denetleme, gişe uygulamaları, otopark giriş çıkış yönetimi vb. alanlarda araçların kontrolünde verimli olarak kullanılabilecektir. Yapılan çalışma Karayolları Trafik Yönetmeliğindeki Türk plaka standartlarına uyan ve tek satır olan, sivil araçlar için geliştirilen bir uygulamadır.Tasarlanan sistemde plaka alanının tespiti için imgeye morfolojik işlemler uygulanmakta ve imgedeki nesnelerin kenar tespitleri yapılmaktadır. Önceden belirlenen kriterlere uyan alanlar kesilerek plaka alanı tespit edilmektedir. Çıkarılan plaka alanındaki gereksiz nesneler temizlendikten sonra karakterler ayrıştırılmakta ve karakterler yapay zeka algoritmalarından biri olan yapay sinir ağları ile eğitilmiş OCR motoru kullanılarak plaka tanıma işlemi tamamlanmaktadır. Tanınan plaka için bariyer açılmaktadır.Çalışmada plaka yerinin tespiti için %100, karakter ayrıştırma için %99,33 ve tanıma için %95,30 doğruluk oranlarına ulaşılmıştır. Abstract Artificial intelligence is defined as a system producing reactions from ideas and showing autonomous behaviours or high cognitive functions such as perception, learning, stating opinion, problem solving which are peculiar to human intelligence. License plate recognition practices are a needed system generally used for control and security. By means of present technological developments, this system, which will be developed with artificial intelligence without human being, will be used in traffic inspection, box office applications, car park entrance and exit management etc. This study is targeted to be an application for civil vehicles with single line that are compatible with Turkish license plate standards in Highway Regulations. In premeditated system, morphological operations are applied for identifying plate area and the edge detection of the images is done. Plate area is established by cutting the areas which are suitable for preset criterions. After cleaning unnecessary objects in the displacement plate area, characters are analyzed and the plate is recognized by processing characters with Optical Character Recognation module. The barrier is opened for recognized plate.In this study, validity percentages are % 100 for identifying plate localization, %99,33 for character segmentation and % 95,30 for recognisation.