Instilling social to physical: co-regularized heterogeneous transfer learning

摘要: Ubiquitous computing tasks, such as human activity recognition (HAR), are enabling a wide spectrum of applications, ranging from healthcare to environment monitoring. The success ubiquitous task relies on sufficient physical sensor data with groundtruth labels, which always scarce due the expensive annotating process. Meanwhile, social media platforms provide lot or semantic context information. People share what they doing and where frequently in messages post. This rich set socially shared activities motivates us transfer knowledge address sparsity issue labelled data. In order context, we propose Co-Regularized Heterogeneous Transfer Learning (CoHTL) model, builds common space derived two heterogeneous domains. Our proposed method outperforms state-of-the-art methods namely region function discovery.