Explaining black-box models for time series data is critical for the wide-scale adoption of deep learning techniques across domains such as healthcare. Recently, explainability methods for deep time series models have seen significant progress by adopting saliency methods that perturb masked segments of time series to uncover their importance towards the prediction of black-box models. Thus far, such methods have been largely restricted to regular time series. Irregular time series, however, sampled at irregular time intervals and potentially with missing values, are the dominant form of time series in various critical domains (e.g., hospital records). In this paper, we conduct the first evaluation of saliency methods for the interpretation of irregular time series models. We first translate techniques for regular time series into the continuous time realm of irregular time series and show under which circumstances such techniques are still applicable. However, existing perturbation techniques neglect the timing and structure of observed data, e.g., informative missingness when data is not missing at random. Thus, we propose Contimask, a simple framework to also apply non-differentiable perturbations, such as simulating that parts of the data had not been observed using NeuroEvolution. Doing so, we successfully detect how structural differences in the data can bias irregular time series models on a real-world sepsis prediction task where 90% of the data is missing. Source code is available on GitHub.

Today's fitness bands and smartwatches typically track heart rates (HR) using optical sensors. Large behavioral studies such as the U.K. Biobank use activity trackers without such optical sensors and thus lack HR data, which could reveal valuable health trends for the wider population. In this paper, we present the first dataset of wrist-worn accelerometer recordings and electrocardiogram references in uncontrolled at-home settings to investigate the recent promise of IMU-only HR estimation via ballistocardiograms. Our recordings are from 42 patients during the night, totaling 310 hours. We also introduce a frequency-based method to extract HR via curve tracing from IMU recordings while rejecting motion artifacts. Using our dataset, we analyze existing baselines and show that our method achieves a mean absolute error of 0.88 bpm-76% better than previous approaches and the first to surpass established medical standards for heart rate monitors. Our results validate the potential of IMU-only HR estimation as a key indicator of cardiac activity in existing longitudinal studies to discover novel health insights.

Virtual reality (VR) presents immersive opportunities across many applications, yet the inherent risk of developing cybersickness during interaction can severely reduce enjoyment and platform adoption. Cybersickness is marked by symptoms such as dizziness and nausea, which previous work primarily assessed via subjective post-immersion questionnaires and motion-restricted controlled setups. In this paper, we investigate the dynamic nature of cybersickness while users experience and freely interact in VR. We propose a novel method to continuously identify and quantitatively gauge cybersickness levels from users' passively monitored electroencephalography (EEG) and head motion signals. Our method estimates multitaper spectrums from EEG, integrating specialized EEG processing techniques to counter motion artifacts, and, thus, tracks cybersickness levels in real-time. Unlike previous approaches, our method requires no user-specific calibration or personalization for detecting cybersickness. Our work addresses the considerable challenge of reproducibility and subjectivity in cybersickness research. In addition to our method's implementation, we release our dataset of 16 participants and approximately 2 hours of total recordings to spur future work in this domain. Source code: https://github.com/eth-siplab/EEG_Cybersickness_Estimation_VR-Beyond_Subjectivity.

When learning to play an instrument, it is crucial for the learner’s muscles to be in a relaxed state when practicing. Identifying, which parts of a song lead to increased muscle tension requires self-awareness during an already cognitively demanding task. In this work, we investigate unobtrusive pressure sensing for estimating muscle tension while practicing songs with the guitar. First, we collected data from twelve guitarists. Our apparatus consisted of three pressure sensors (one on each side of the guitar pick and one on the guitar neck) to determine the sensor that is most suitable for automatically estimating muscle tension. Second, we extracted features from the pressure time series that are indicative of muscle tension. Third, we present the hardware and software design of our PressurePick prototype, which is directly informed by the data collection and subsequent analysis.

Virtual Reality (VR) systems have traditionally required users to operate the user interface with controllers in mid-air. More recent VR systems, however, integrate cameras to track the headset's position inside the environment as well as the user's hands when possible. This allows users to directly interact with virtual content in mid-air just by reaching out, thus discarding the need for hand-held physical controllers. However, it is unclear which of these two modalities-controller-based or free-hand interaction-is more suitable for efficient input, accurate interaction, and long-term use under reliable tracking conditions. While interacting with hand-held controllers introduces weight, it also requires less finger movement to invoke actions (e.g., pressing a button) and allows users to hold on to a physical object during virtual interaction. In this paper, we investigate the effect of VR input modality (controller vs. free-hand interaction) on physical exertion, agency, task performance, and motor behavior across two mid-air interaction techniques (touch, raycast) and tasks (selection, trajectory-tracing). Participants reported less physical exertion, felt more in control, and were faster and more accurate when using VR controllers compared to free-hand interaction in the raycast setting. Regarding personal preference, participants chose VR controllers for raycast but free-hand interaction for mid-air touch. Our correlation analysis revealed that participants' physical exertion increased with selection speed, quantity of arm motion, variation in motion speed, and bad postures, following ergonomics metrics such as consumed endurance and rapid upper limb assessment. We also found a negative correlation between physical exertion and the participant's sense of agency, and between physical exertion and task accuracy.

ybersickness has been one of the main impediments to the widespread adoption of Virtual Reality for decades. It has been argued that several factors can influence the occurrence of cybersickness, such as technical factors, interaction design, but also users’ demographics and their perceived presence. Yet, previous studies had comparably small sample sizes and demographically homogeneous samples; comparisons across studies (e.g., regarding demographic factors) are challenging due to the large variation in the studied virtual environments. In this paper, we address these limitations and report the results of a lab-in-the-field experiment on cybersickness with a large and heterogeneous sample of $N =837$ participants who navigated and interacted inside a virtual environment (ages 18–80, $M = 29.34, SD = 9.50$, 431 males, 400 females, 6 non-binaries and other). We found that female participants and participants with lower VR experience were more susceptible to experiencing higher levels of cybersickness. Participants’ cybersickness levels increased with the time spent in VR and with the distance traversed in the virtual world up to a point, above which reported levels declined. We also found a link between higher levels of cybersickness and reduced head motion, as well as between lower levels of cybersickness and more head motion, which led them to explore more of the virtual environment. In contrast to past studies, we did not find any evidence suggesting an effect of age on cybersickness, nor a negative correlation between presence and cybersickness. Based on our results, we derived a model that achieves a mean classification accuracy of 67.1% for two levels of cybersickness using demographic, user experience, and behavioral data in VR.