Την Παρασκευή 02-02-2024 και ώρα 13:00, η ακαδημαϊκή κοινότητα του Τμήματος Ηλεκτρονικών Μηχανικών καλείται να παρακολουθήσει την πολύ ενδιαφέρουσα διάλεξη με τίτλο “Federated Learning: Trade-off between Utility and Privacy“ της Prof. Rebeca Pilar Redondo από το Department of Telematics Engineering του University of Vigo της Ισπανίας, στο πλαίσιο των ATHENA European University Talks 2023-24.
About the talk
There is a growing common view that the transition from centralized ML to distributed ML at the network edge is necessary but largely complex. While it is generally agreed that the intelligence of ML should be moved closer to the devices (data producers located at the network edge) and benefit from plentiful computing nodes, the emerging design of efficient distributed ML algorithms has to deal explicitly with the heterogeneity of the computing and communication equipment (e.g, from IoT sensors to cloud servers; from wireless channels and strong interference to local data; from privacy concerns to public data). The first breakthrough for employing multiple nodes for training and guaranteeing privacy is federated learning, which enables model synthesis from a large corpus of decentralized data. However, some aspects must be analyzed in detail since substantial gains can potentially be realized in terms of reduced delay, training speed and handling of massive data sets if (i) a system-level design for collaborative sampling, splitting of the computations, training and prediction among the nodes is analyzed and optimized for performance (convergence speed, accuracy, and a carefully chosen problem-dependent utility score); (ii) a set of communication-level techniques, such as coding, massive random access and differential privacy are used to fulfill privacy requirements and network-level performance as for delay, energy, and bandwidth.
About the speaker
Rebeca P. Díaz-Redondo is a full professor at the Telematics Engineering Department (Universidade de Vigo – Uvigo, Spain), and she leads the Information & Computing Lab (IC Lab), a research group affiliated with the Atlantic Research Center. In the last decade, her research area has focused on applying artificial intelligence (IA) solutions in distributed and coordinated networks of devices. Her current work faces the definition of efficient information-sharing protocols among peer devices to avoid overhead in communications and computation in the field of (distributed) federated learning ((D)FL). This approach must be aligned with privacy & security-aware designs to balance efficiency, privacy, security and utility. She is especially interested in applying these approaches in IoT, taking into account specific protocols and distributed edge computing solutions that combine capabilities of Cloud/Edge (Fog/Mist) computing as a whole, i.e., cloud to-edge continuum. Her contributions have been published in more than 80 indexed scientific journals (32 Q1, 16 in the last 5 years, and 28 Q2), as a result of the research work developed within more than 10 research projects in the last 10 years (supported by National and European funds), having been PI in 7. Since 2015, she has co-advised 13 PhD theses (4 of them funded by EU grants and 2 of them funded by FPI and FPU programmes) and she is currently advising 7 doctoral students (4 of them in industry and 1 of them funded by an FPU grant) with the DocTIC doctoral programme at UVigo.
Η ομιλία θα μεταδοθεί ζωντανά μέσω της πλατφόρμας Zoom, ακολουθώντας το σύνδεσμο https://us02web.zoom.us/j/4039520498?pwd=ZVRzUm16T2gwTENwMlh0NnZ3T2FBUT09&omn=85962775320.
