ASM2TV: An Adaptive Semi-supervised Multi-Task Multi-View Learning Framework for Human Activity Recognition
چکیده
Many real-world scenarios, such as human activity recognition (HAR) in IoT, can be formalized a multi-task multi-view learning problem. Each specific task consists of multiple shared feature views collected from sources, either homogeneous or heterogeneous. Common among recent approaches is to employ typical hard/soft sharing strategy at the initial phase separately for each view across tasks uncover common knowledge, underlying assumption that all are conditionally independent. On one hand, possibly relate other under practical situations. supervised methods might insufficient when labeled data scarce. To tackle these challenges, we introduce novel framework ASM2TV semi-supervised learning. We present new perspective named gating control policy, learnable task-view-interacted policy adaptively selects most desirable candidate block any task, which uncovers more fine-grained relatedness and improves inference efficiency. Significantly, our proposed gathering consistency adaption procedure takes full advantage large amounts unlabeled fragmented time-series, making it general accommodates wide range applications. Experiments on two diverse HAR benchmark datasets various subjects sources demonstrate framework's superiority over state-of-the-arts. Anonymous codes available https://github.com/zachstarkk/ASM2TV.