This paper studies the novel learning scenarios of Distributed Online Multi-tasks (DOM), where the learning individuals with continuously arriving data are distributed separately and meanwhile they need to learn individual models collaboratively. It has three characteristics: distributed learning, online learning and multi-task learning. It is motivated by the emerging applications of wearable devices, which aim to provide intelligent monitoring services, such as health emergency alarming and movement recognition.

To the best of our knowledge, no previous work has been done for this kind of problems. Thus, in this paper a collaborative learning scheme is proposed for this problem. Specifically, it performs local learning and global learning alternately. First, each client performs online learning using the increasing data locally. Then, DOM switches to global learning on the server side when some condition is triggered by clients. Here, an asynchronous online multi-task learning method is proposed for global learning. In this step, only this client's model, which triggers the global learning, is updated with the support of the difficult local data instances and the other clients' models. The experiments from 4 applications show that the proposed method of global learning can improve local learning significantly. DOM framework is effective, since it can share knowledge among distributed tasks and obtain better models than learning them separately. It is also communication efficient, which only requires the clients send a small portion of raw data to the server.