Workshop on Systems Challenges in Reliable and Secure Federated Learning

Workshop Deadlines

Camera ready instructions Camera-ready papers can have 2 pages of technical content, including all text, figures, tables, etc. Bibliographic references are not included in the 2-page limit. You must use the US letter paper size, with all text and figures fitting inside a 178 x 229 mm (7 x 9 in) block centered on the page, using two columns separated by 8 mm (0.33) of whitespace. Use 10-point font (typeface Times Roman, Linux Libertine, etc.) on 12-point (single-spaced) leading. Most of these rules are automatically applied when using the official LaTeX or MS Word templates from the ACM, available here. For the documentclass, use the command "\documentclass[sigconf]{acmart}".

Federated learning is a distributed machine learning approach that allows a global model to be trained across multiple decentralized client devices. This offers privacy, accuracy, and scalability advantages by utilizing multiple clients for training and using their local data. Since its inception in 2017, the area has seen great strides on the theoretical side and to a lesser extent, on the practical deployment side.

The natural progression of technology is that once some technology becomes usable and then useful at scale, attention starts to get focused on the reliability and the security aspects of the technology. Distributed learning is no exception. In some sense, the bar for this technology is high as this is meant to be used in our private spaces, listening and learning to model our behavior and carry out our wishes with lightning speed. So this workshop invites papers focusing on the systems challenges and initial solutions to the problems of reliability and security in federated and more generally distributed learning.

The short papers of 2 pages length in standard ACM two-column format (+ unlimited references) are meant to present fresh design ideas, with some initial results that show feasibility of the ideas. Any topic within the scope defined above is of interest. These will be presented at a poster session in the workshop and awards will be given for the best poster(s). The submissions should be anonymous (author names should not appear on the submission).

Some sample topics are:

• How to create a trusted execution environment on client devices, e.g., through novel TCBs?
• How can we certify the behavior of clients or the server, either continually or upon critical events?
• How can we guarantee the accuracy of the execution in the face of model or data poisoning attacks at the clients?
• How can we deal with network issues connecting clients to the federated server?
• How can we develop asynchronous algorithms for this setting?
• How can FL solutions leverage heterogeneity in the system, data, models and or clients?
• What is a compelling application scenario?