Keynotes

The conference will include keynote talks by distinguished speakers from industry and academia.

Edge computing is more and more used for running machine learning algorithms to
assist IoT (Internet of Things) tasks such as video analytics, compression,
anomaly detections, privacy preservation, and others. On the other hand, edge
computing is often resource constrained, challenging training and inference
phases of machine learning algorithms.
In this talk, we will discuss Federated Learning (FL), a machine learning
paradigm that enables a cluster of decentralized edge devices to collaboratively
train a shared machine learning model without exposing users’ raw data and hence
providing privacy to clients. However, the intensive model training computation
is energy-demanding and poses severe challenges to end devices’ battery life. We
will discuss few approaches to develop a training pace controller deployed on
the edge devices that actuates the hardware operational frequencies over
multiple configurations to achieve energy-efficient federated learning; and to
tackle the straggler problem in FL via the decentralized selection of coresets,
representative subsets of a dataset, where our approach creates coresets
directly on edges and optimizes the coreset clusters to reduce FL training time
and hence energy and other resource usage.
We will conclude with discussing further challenges of machine learning on edge
devices.

In the past decade, there has been significant progress in private space development, commonly referred to as NewSpace. One of the most notable advancements is the deployment of satellite constellations, with numerous small satellites being placed in low Earth orbit (LEO) to support communication and remote sensing applications. Looking ahead, it is anticipated that CPS (Cyber-Physical Systems) will be extended into space by leveraging artificial satellites as IoT devices or NTN communication platforms. To achieve this vision, we are conducting fundamental research on high-bandwidth communication satellites utilizing formation flying techniques. This presentation will introduce these emerging trends and highlight the related research activities.

Large-scale distributed machine learning training networks are increasingly facing scaling problems with respect to FLOPS per deployed compute node. Communication bottlenecks can inhibit the effective utilization of expensive GPU resources. The root cause of these performance problems is not insufficient transmission speed or slow servers; it is the structure of the distributed computing and the communication characteristics it incurs. Large machine learning workloads typically provide relatively asymmetric, and sometimes
centralized, communication structures, such as gradient aggregation and model update distribution.  Even when training networks are less centralized, the amount of data that needs to be sent to aggregate several thousand input values through collective communication functions such as AllReduce can lead to Incast problems that overload network resources and servers. This talk discusses opportunities and challenges for a systems approach towards making distributed machine learner faster, more energy-efficient, and scalable.