The last decade has seen a paradigm shift in data storage and exploitation: centralized data centers have been largely replaced by cloud storage which connects to multiple remote heterogeneous endpoints, including virtual machines, containers and various Internet of Things devices. This is a key factor in the continuous growth of the complexity of software applications, and has posed new challenges for applications that require the orchestration of multiple non-standardized cloud solutions and extra layers of administration for virtual machines. In response to this challenge, LINAGORA Labs has developed deployment frameworks that are open-source, extensible, scalable, and IaaS and PaaS-agnostic for multi-cloud applications, guaranteeing interoperability.
Privacy-aware AI applications, and in particular those for speech processing and natural language understanding, must confront the challenge of how to maintain acceptable performance while keeping voice data as close as possible to its source. Edge computing offers a solution that combines local and distant computing devices to produce systems that address both needs. In our LinTO platform, we use edge computing to embed wake word detection and feature extraction locally to an IoT device (Raspberry, smartphone, etc.) while maintaining large vocabulary decoding on a server potentially deployed in the cloud.
Federated Learning is a recent fast-growing topic in AI that investigates methods to collaboratively train machine learning models on non-uniformly dispersed datasets without moving the data from their original locations. This is preferable in maintaining data privacy and avoiding the digital cost of transferring large amounts of data to a central server. It is also an integral part of implementing sophisticated machine learning techniques on edge devices.
Pro-VE 2013, IFIP Working Conference on VIRTUAL ENTERPRISES
Roland Stühmer, Yiannis Verginadis, Iyad Alshabani, Thomas Morsellino, Antonio Aversa