2018/11/14 by Karan Aggarwal, Shafiq Joty, Aggarwal, Karan +5 · 1 citation
Computer Science · #Anomaly Detection Techniques and Applications #Context-Aware Activity Recognition Systems #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Time Series Analysis and Forecasting
paper · pdf · doi:10.48550/arxiv.1811.06847
openalex publication_date 2018/11/14 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Sufficient physical activity and restful sleep play a major role in the\nprevention and cure of many chronic conditions. Being able to proactively\nscreen and monitor such chronic conditions would be a big step forward for\noverall health. The rapid increase in the popularity of wearable devices\nprovides a significant new source, making it possible to track the user's\nlifestyle real-time. In this paper, we propose a novel unsupervised\nrepresentation learning technique called activity2vec that learns and\n"summarizes" the discrete-valued activity time-series. It learns the\nrepresentations with three components: (i) the co-occurrence and magnitude of\nthe activity levels in a time-segment, (ii) neighboring context of the\ntime-segment, and (iii) promoting subject-invariance with adversarial training.\nWe evaluate our method on four disorder prediction tasks using linear\nclassifiers. Empirical evaluation demonstrates that our proposed method scales\nand performs better than many strong baselines. The adversarial regime helps\nimprove the generalizability of our representations by promoting subject\ninvariant features. We also show that using the representations at the level of\na day works the best since human activity is structured in terms of daily\nroutines\n