Last week (09–13 Sept.), I attended Ubicomp Conference and ISWC (collocated) — the premier conference for ubiquitous and pervasive computing research — at the QEII Centre in London. Here, a glimpse of what happened during the conference and a summary of a few selected works. The full program is accessible online.
Workshops — A summary of WellComp and Mental Health: Sensing & Intervention (MHSI):
During the first day, the WellComp workshop attracted works under the themes of wellbeing metrics and interventions. The best paper awarded to Heli Koskimäki (Oura Health) et al. for their work titled “Following the Heart — What Does Variation of Resting Heart Rate Tell about Us as Individuals and as Population”. Their analysis on resting heart rate and heart rate variability data obtained from Oura ring users revealed interesting insights about lifestyle, society, and the seasonality effect at population level (n=57K). In another work — building on the Attention Restoration Theory, Zann Andreson, presented their vision on future technologies that support well-being in outdoor activities. Using mixed-methods approaches, they presented early insights on how computing is being used in this set of activities, and they nicely concluded that…“In outdoor activities, human-nature interaction holds priority over human-computer interaction”. During the second day, I attended the MHSI workshop. Martin Maritsch presented their early work on correcting heart rate variability measurements by taking into account motion information (e.g., accelerometer data). Another interesting work examined the effect of temporal factors of listening to music on stress reduction. In a lab experiment, they showed that listening to music before a stressor would lead to better stress reduction than listening after the stressor (HRV-derived stress from E4 devices). This illustrates the need of designing ‘preventive’ technologies that deliver ‘just-in-time’ interventions.
Two fantastic keynotes were delivered by Prof. Marta Kwiatkowska (Univ. of Oxford) and Lama Nachman (Intel Labs). Marta’s talk on Wednesday, focused on safety and reliability guarantees of ML/DL in ubiquitous systems, illustrated through several examples from security concerns in fitness trackers to self-driving cars to recognition of road-signs. Lana’s talk on Thursday, focused on the need of going in-the-wild and build technologies that are deployable in the real world through examples from smart homes, smart manufacturing, adaptive learning, and autonomous vehicles. The highlight of her talk was the work ‘Assistive Context Aware Toolkit (ACAT)’ — the technology designed to help Stephen Hawking. This work perfectly illustrates the need of aligning technology to the specific tasks people are trying to do, and the context in which they do it so.
The main program included more than 200 papers (IMWUT, ISWC long/brief/notes) that spanned over a wide range of topics (from IoT and health to social dynamics to wearable sensing, and many others). Here, I summarize a few interesting ones that are relevant to our Social Dynamics team, Cambridge, UK:
- Differentiating Higher and Lower Job Performers in the Workplace Using Mobile Sensing’ — Using a multi-modal dataset (beacons, smartphones, smartwatches) they extracted features such as mobility, phone usage, physical activity, heart rate, sleep, stress, behavior at work and classified higher and lower job performers using ML algorithms. Interesting implications of this work — how passive objective data could be used to assess and provide feedback to workers.
- Prediction of Mood Instability with Passive Sensing — Analysis of the publicly available StudentLife dataset to predict mood instability with passive data — three weeks of data found to be the minimum amount of data that yielded to reliable mood predictions.
- Detecting Conversing Groups Using Social Dynamics from Wearable Acceleration: Group Size Awareness — Using accelerometer data, from the MingleNMatch dataset, they proposed a method for automatic detection of conversing groups. Based on the F-formation theory, proposed by the social psychologist Adam Kendon, which defines how participants spatially and orientally organize themselves into a group to facilitate conversation, they proposed a method that captures interaction dynamics than proxemics.
- Using Unobtrusive Wearable Sensors to Measure the Physiological Synchrony Between Presenters and Audience Members — An interesting approach to quantify physiological synchrony (PS) — PS occurs when the “physiological activity between two or more people” becomes “associated or interdependent”. Using a Dynamic Time Warping algorithm on EDA time series data collected between presenters and audience members during a 2-days conferences, they defined metrics for physiological synchrony quantification. Natural extensions of this work could be the exploration of other physiological indicators such as HRV-derived stress, and its applicability in other scenarios (e.g., work meetings).
- Animo: Sharing Biosignals on a Smartwatch for Lightweight Social Connection — Animo is an application that runs on Fitbits and allows users to share their biosignals. Results on a two-week study showed how users developed new communication patterns by sharing their biosignal, and how ‘animos’ facilitate the ‘sense of being with another’.
- Drinks & Crowds: Characterizing Alcohol Consumption through Crowdsensing and Social Media — Interesting analysis using a combination of crowdsensing data (Youth@Night) and social media (Instagram posts ~30K) data to uncover temporal, spatial, social, and contextual patterns on alcohol consumption. They showed that the context alone has a basic discriminative power, whereas image content performs better than context; thus learning on crowdsensing+social media brings no improvement in alcohol type recognition. Interesting to explore whether this holds in another domains.
- Multi-target Affect Detection in the Wild: An Exploratory Study illustrated the challenges of Affective Computing (AC) in-the-wild. In a two-week study, they collected physiological data (ACC, PPG, EDA, and TEMP) from 11 healthy individuals using E4 devices, and self-reported data for Valence/Arousal, STAI, and stress. They compared different classification approaches (using classical evaluations — featured-based classifiers and DL approaches — CNN) and reported that even with more sophisticated approaches using DL the improvement compared to a random guess was minor. This study depicted learnings, pitfalls, and challenges of AC in-the-wild, but at the same time illustrates how little this space is explored outside the lab.
- BoostMeUp: Improving Cognitive Performance in the Moment by Unobtrusively Regulating Emotions with a Smartwatch — Another interesting work presented by Jean Costa who showed how people’s perceptions of their own bodily signals can influence their emotional experience. He presented BoostMeUp, a smartwatch application, which uses haptic feedback for emotion regulation. He showed that people exposed in slow haptic feedback in stress-induced tasks decreased their anxiety (increased HRV) compared to fast haptic feedback which yielded the opposite effects.
Overall, it was an amazing experience and a great opportunity to connect with researchers who work in the same space. Looking forward to next year’s conference in Cancun, Mexico…viva!