AI-powered eVolution towards opEn and secuRe edGe architEctures

Verge将从三个角度来应对边缘计算的演进：“Edge for AI”、“AI for Edge”以及AI for Edge的安全、隐私和可信性。“用于AI的边缘”定义了一个灵活、模块化和融合的边缘平台，该平台已准备好在边缘支持分布式AI。这是通过统一云本地应用程序、MEC和网络服务的生命周期管理和闭环自动化实现的，同时充分利用多核和多加速器功能实现超高计算性能。“AI for Edge”通过管理和协调底层物理、网络和计算资源，实现动态功能布局。考虑到Edge资源限制，将以高效且无冲突的方式确保特定于应用的网络和计算KPI。讨论了AI for Edge的安全性、隐私性和可信性，通过为模型决策提供解释以提高对模型的信任度，从而确保基于AI的模型免受对手攻击的安全性、数据和模型的隐私以及训练和执行的透明度。Verge将通过在两个用例中提供7个演示来验证这三个观点-土耳其两个独立的Arçelik站点上的XR驱动的支持Edge的工业B5G应用程序，以及佛罗伦萨的Edge辅助自动有轨电车运营。Verge将通过与相关标准化机构和开源机构的联系和贡献，通过TRL展示进展的一系列演示，并通过创建一个开放的数据空间，使公众能够获得项目产生的数据集，向学术界、工业界和更广泛的利益攸关方传播成果。

VERGE will tackle evolution of edge computing from three perspectives: “Edge for AI”, “AI for Edge” and security, privacy and trustworthiness of AI for Edge. “Edge for AI” defines a flexible, modular and converged Edge platform that isready to support distributed AI at the edge. This is achieved by unifying lifecycle management and closed-loop automation for cloud-native applications, MEC and network services, while fully exploiting multi-core and multi-accelerator capabilities for ultra-high computational performance. “AI for Edge” enables dynamic function placement by managing and orchestrating the underlying physical, network, and compute resources. Application-specific network and computational KPIs will be assured in an efficient and collision-free manner, taking Edge resource constraints in to account. Security, privacy and trustworthiness of AI for Edge are addressed to ensure security of the AI-based models against adversarial attacks, privacy of data and models, and transparency in training and execution by providing explanations for model decisions improving trust in models. VERGE will verify the three perspectives through delivery of 7 demonstrations across two use cases - XR-driven Edge-enabled industrial B5G applications across two separate Arçelik sites in Turkey, and Edge-assisted Autonomous Tram operation in Florence. VERGE will disseminate results to academia, industry and the wider stakeholder community through liaisons and contributions to relevant standardization bodies and open sources, a series of demonstrations showing progression through TRLs and by creating an open dataspace for enabling public access to the datasets generated by the project.