Federated Mobile Sensing for Activity Recognition


The tutorial’s requirements from the audience will be kept minimal to enhance participation.

Hands-on section

In the hands-on section of our tutorial, we will be experimenting on a public cloud over live terminals or notebooks. However, users will be free to deploy the code locally to their devices, with fewer clients running simultaneously. As such, we would require the participants to bring their own laptop, equipped with a fairly recent CPU and at least 8GB of RAM (even less with smaller batch sizes). Our tutorial does not explicitly require the existence of a GPU, although it would speed up training. Finally, we plan to share package and data requirements in advance so as not to overload the venue's local network with simultaneous packages downloads.

The dataset and code will become publicly available later on, closer to the date of the tutorial. Please stay tuned.

Prior knowledge

In terms of audience background, we expect participants to have familiarity with Machine Learning and some passing knowledge on how Recurrent Neural Networks (RNNs) work. Moreover, we expect them to be fluent in Python, and ideally have some experience with PyTorch, although not strictly needed.