| Sumario: | Distributed ML is a rapidly evolving, interdisciplinary field bringing together techniques from distributed systems, networks and machine learning. The advent of deep learning and the pursuit of artificial intelligence has created previously unfathomable deployments both in the datacenter and in the wild, enabling use-cases ranging from virtual reality and intelligent assistants to robots and self-driving cars. However, be it in the form of training or inference, these workloads pose several challenges pushing the compute, memory, sensing and networking capabilities of devices. As such, ML deployments become innately distributed, as a means to scale out local capabilities by leveraging remote or ambient resources in a distributed fashion in a collaborative manner. Simultaneously, the strive for user-privacy and sustainability gives birth to new training solutions, such as Federated Learning. However, adversarial actors may pose challenges to the robustness of the such deployments. Be it deployed in the cloud, edge or consumer end, Distributed ML challenges and pushes the limits of today's compute, networking and algorithmic frontier. To this end, the DistributedML workshop comes in a timely manner to provide a forum for ideas coming from different disciplines to be aired so as to solve current challenges and shape the technology of the future.
|
|---|
