Deep Learning For UbiComp
A UbiComp 2017 tutorial
Overview and Call for Participation
Deep Learning is almost without doubt the biggest breakthroughs in contemporary machine learning research in a generation, carrying extraordinary promises for a wide variety of domains related to automated sensor data analysis. Through Deep Learning methods some of the most challenging machine learning and pattern recognition problems can now (almost) be considered “solved”. Examples of which are automated speech recognition or visual object recognition. Progress of such magnitude gains the attention and hopes of many and as such it is not surprising that the wider ubiquitous computing community of researchers and practitioners has started adopting Deep Learning for various application areas.
Applications of ubiquitous computing impose substantial and specific challenges on machine learning based sensor data analysis techniques. Examples of which include: (1) the inherent noisiness of sensor data including, e.g., faulty or missing data, sparse and typically skewed data / class distributions; (2) the fact that most data analysis problems are essentially time-series analysis tasks with their very own set of challenges; (3) substantial resource limitations and constraints, e.g., on mobile devices; (4) substantial challenges to obtain ground truth information, that is labelled data as it is required for the predominant supervised learning paradigm; and (5)(near) real-time constraints for natural interaction. When employing Deep Learning techniques for UbiComp applications one has to specifically address these (and other) challenges. This, of course, is the case for any analysis method yet it implies that Deep Learning techniques might not be usable “out of the box” and specific considerations need to be taken into account.
In response to the great attraction Deep Learning has to many, and the substantial challenges the UbiComp domain imposes on properly employing such techniques, we will be giving a tutorial on “Deep Learning for Ubiquitous Computing” at the 2017 ACM Joint Conference on Ubiquitous and Pervasive Computing, co-located with the International Symposium on Wearable Computing. We invite applications for participation from researchers and practitioners who are interested in learning about Deep Learning thereby specifically focussing on how to utilize this exciting area of machine learning techniques for real-world applications as they are tackled in the wider ubiquitous, mobile, and wearable computing arena.
The one-day tutorial will be held prior to the main UbiComp conference. It is organized by four expert researchers from the field who all have substantial experience with machine learning for UbiComp in general and with Deep Learning applied to this subject area in particular. This whole day tutorial will contain lectures and discussions on the theoretical foundations of Deep Learning — grounded into the Ubicomp domain. Whilst exploring the relevant background that as it is necessary for understanding the concepts of Deep Learning, we will put great emphasis on practical examples as they are relevant for those interested in realising real-world UbiComp applications. The combination of expert lecturing and hands-on practical engagement in the very UbiComp application domain will provide participants with the unique opportunity to learn about the foundations of Deep Learning whilst at the same time gaining practical experience on how to apply these complex modelling techniques in their core research and application domain.
Attendance and Registration
Practitioners and researchers who are interested in participating on-site are invited to apply for a place at the tutorial. Details regarding attendance, venue and costs are provided on the main UbiComp conference website here. The tutorial will take place on September 11th prior to the main conference. On-site participation will be limited to a maximum of 100 participants. Applications have to be submitted online (follow conference link above) and are limited to 500 words statements that shall provide background information on the candidate with regards to experience in the machine learning as well as in the ubiquitous, mobile, and wearable computing area, their research interests (possibly including links to recent publications), and their expectations on the tutorial. Notifications about selection are sent on a rolling basis by organisers who provide a registration code to those accepted – tutorial organisers will respond to submissions within five days, at the limit. On-site participants will have to register at least for the workshop part of the conference, which will cover the tutorial as well.
Required Equipment and Software
All material shown at and used for the tutorial can be found here (password protected; not for distribution beyond attendees of tutorial).
|Thomas Ploetz||Georgia Tech, USAfirstname.lastname@example.org|
|Yu Guan||Newcastle University, UKemail@example.com|
|Sourav Bhattacharya||Nokia Bell Labs, UKfirstname.lastname@example.org|
|Nic Lane||Nokia Bell Labs & UCL, UKemail@example.com|