Two new papers on Federated Domain Generalization and Knowledge Graphs
The BIFOLD fellows, Prof. Dr. Manfred Hauswirth and Dr. Danh Le Phuoc, are leading the agility project "Unified Processing Model for Distributed Stream Reasoning", which began in October 2022 and is expected to conclude in December 2024. Recently, researchers from this project released two preprints, which have been accepted for the CVPR 2024 and ESWC 2024 conferences. Both preprints will be published in the conference proceedings.
The paper "Efficiently Assemble Normalization Layers and Regularization for Federated Domain Generalization" will be presented at the IEEE/CVF Conference on Computer Vision and Pattern Recognition 2024 (CVPR 2024), which is regarded as one of the most important conferences in its field. According to Google Scholar Metrics (2022), it is the highest impact computing venue. The paper aims to solve the problem of domain shift, a formidable challenge in machine learning, in the context of a federated learning setting. To better generalize unseen data provided by participant clients, the authors proposed a novel method called gPerXAN, which relies on a normalization scheme working in concert with a regularizer. It does this by making sure the domain-specific data at each client is selectively filtered (normalization) while giving the learning process necessary hints on what's important to focus on (guiding regularizer). This helps the learning algorithm to ignore irrelevant details while still being good at recognizing general patterns. The paper's comprehensive tests show that gPerXAN works better than other methods, especially when dealing with different kinds of data, like medical information.
Publication Details: Le Huy Khiem, Long Tuan Ho, Cuong Do, Danh Le-Phuoc, Kok-Seng Wong. Efficiently Assemble Normalization Layers and Regularization for Federated Domain Generalization. https://arxiv.org/abs/2403.15605
The paper "VisionKG: Unleashing the Power of Visual Datasets via Knowledge Graph" will be presented at the Extended Semantic Web Conference (ESWC 2024). The researchers introduce a novel tool called "VisionKG." It leverages knowledge graphs and Semantic Web technologies to interlink, organize, and manage visual datasets in a unified manner. In simpler terms, it helps organize and manage lots of visual data, even if it is in different formats or categories. Unlike other methods, VisionKG focuses on understanding the meaning of the data itself rather than just its descriptions. This makes it better at finding specific pictures or details within the data. VisionKG is assessable online via https://vision.semkg.org/.
Publication Details: Jicheng Yuan, Anh Le-Tuan, Manh Nguyen-Duc, Trung-Kien Tran, Manfred Hauswirth, Danh Le-Phuoc. VisionKG: Unleashing the Power of Visual Datasets via Knowledge Graph. https://arxiv.org/abs/2309.13610
The BIFOLD agility project "Unified Processing Model for Distributed Stream Reasoning" aims to develop a framework that integrates complex data from various distributed stream data sources, like cameras and lidars, using semantic stream reasoning linked to knowledge graphs. The project's goals include building a runtime compiler for queries and a coordinator for distributed computing resources. It also focuses on optimizing reasoning operations, such as learning and probabilistic reasoning, for devices with limited resources. Additionally, the project is exploring the design of dynamic optimization algorithms for the effective management of distributed reasoning tasks. These efforts are directed toward answering key research questions about deployment, operation, and coordination within distributed systems.