27 September 2023 in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
Depression is a serious mental illness. The current best guideline in depression treatment is closely monitoring patients and adjusting treatment as needed. Close monitoring of patients through physician-administered follow-ups or self-administered questionnaires, however, is difficult in clinical settings due to high cost, lack of trained professionals, and burden to the patients. Sensory data collected from mobile devices has been shown to provide a promising direction for long-term monitoring of depression symptoms. Most existing studies in this direction, however, focus on depression detection; the few studies that are on predicting changes in depression are not in clinical settings. In this paper, we investigate using one type of sensory data, sleep data, collected from wearables to predict improvement of depression symptoms over time after a patient initiates a new pharmacological treatment. We apply sleep trend filtering to noisy sleep sensory data to extract high-level sleep characteristics and develop a family of machine learning models that use simple sleep features (mean and variation of sleep duration) to predict symptom improvement. Our results show that using such simple sleep features can already lead to validation F1 score up to 0.68, indicating that using sensory data for predicting depression improvement during treatment is a promising direction.
28 October 2025 in ACM Transactions on Intelligent Systems and Technology
Currently, depression treatment relies on closely monitoring patients’ response to treatment and adjusting the treatment as needed. Using self-reported or physician-administrated questionnaires to monitor treatment response is, however, subjective, costly and suffers from recall bias. In this paper, we explore using location sensory data collected passively on smartphones to predict treatment outcome. To address heterogeneous data collection on Android and iOS phones, the two predominant smartphone platforms, we explore using domain adaptation techniques to map their data to a common feature space, and then use the data jointly to train machine learning models. We further explore integrating contrastive learning with domain adaptation to augment data and learn feature embeddings. These learned embeddings are then used to train machine learning models to predict depression treatment outcomes. Our evaluation shows that using the embeddings learned by jointly integrating contrastive learning and domain adaptation leads to the best prediction accuracy. In addition, our results show that using location features and baseline self-reported questionnaire score can lead to F1 score up to 0.76. This accuracy is comparable to that obtained using periodic self-reported questionnaires, indicating that using location data is a promising direction for predicting depression treatment outcome. Last, when all location and questionnaire data are used together, the F1 score further increases to 0.79.
19 November 2025 in ACM Transactions on Computing for Healthcare
Personalized depression treatment, despite decades of research, is still elusive. The current best practice is closely monitoring patients and adjusting the treatment as needed. Close monitoring of patients is, however, challenging in clinical settings due to high cost, limited resources, and patient burden. In this paper, we explore using daily step counts automatically collected by a fitness wristband for depression treatment status monitoring. We focus on deep learning models that can effectively learn feature representation from a sequence of raw daily step count data, without relying on handcrafted features. One significant challenge is that our dataset is small, while deep learning models typically need large amounts of data to achieve strong generalization. To address this challenge, we investigate self-supervised contrastive pretraining to learn feature representation, followed by utilizing these representations for binary classification. Our results show that this contrastive pretraining approach significantly outperforms several state-of-the-art deep learning models. When using step count data together with a one-shot self-reported baseline questionnaire score, it leads to validation \(F_{1}\) score up to 0.74, comparable to the best \(F_{1}\) score (0.72) when using baseline and burdensome periodic self-reported scores. Last, when further including daily sleep data, the best \(F_{1}\) score is improved to 0.77.
16 August 2024 in ACM Transactions on Multimedia Computing, Communications, and Applications
Smartphones have emerged as ubiquitous platforms for people to consume content in a wide range of consumption contexts (C2) , e.g., over cellular or WiFi, playing back audio and video directly on phone or through peripheral devices such as external screens or speakers. In this article, we argue that a user’s specific C2 is an important factor to consider in Adaptive Bitrate (ABR) streaming. We examine the current practices of using C2 in five popular ABR players, and identify various limitations in existing treatments that have a detrimental impact on network resource usage and user experience. We then formulate C2-cognizant ABR streaming as an optimization problem and develop practical best-practice guidelines to realize it. Instantiating these guidelines, we develop a proof-of-concept implementation in the widely used state-of-the-art ExoPlayer platform and demonstrate that it leads to significantly better tradeoffs in terms of user experience and resource usage. Last, we show that the guidelines also benefit dash.js player that uses an ABR logic significantly different from that of ExoPlayer.