Collaborative activity recognition with heterogeneous activity sets and privacy preferences

One of the major challenges in Human Activity Recognition (HAR) based on machine learning is scarcity labeled data. Indeed, collecting a sufficient amount training data to build reliable recognition problem often prohibitive. Among many solutions literature mitigate this issue, collaborative emerging as promising direction distribute annotation burden over multiple users that cooperate shared model. issues existing methods they assume static activity model with fixed set target activities. In paper, we propose novel approach Growing When Required (GWR) neural networks. A GWR network continuously adapts itself according input data, and hence it particularly suited when share heterogeneous sets Like federated learning, for sake privacy preservation, each user contributes global classifier by sharing personal parameters, not directly order further threats, implement strategy avoid releasing parameters may indirectly reveal information about activities specifically marked private. Our results two well-known publicly available datasets show effectiveness flexibility our approach.