|Distributed Bayesian-based Multi-Label Classification Framework: Application to Social Robots, a seminar by Luís Santos|
In this work we propose a distributed Bayesian-based multi-label classifier to perform inference on a person daily behaviour (routine) model, based on multi- modal input data. The model fuses local perceived information such as facial expression or recognized speech with remotely classified information coming, in this specific implementation, from a cloud server, such as recognized objects or the person’s personal agenda or preferences. The benefits of the proposed distributed classification framework are, for once that the proposed multi-label classification scheme is applied to infer a comprehensive description of a person daily behaviour, addressing the questions Who, Why, How, When and What simultaneously, and secondly that the computational effort can be distributed between a robot and a cloud. We propose to solve our problem in two steps based on the principles of Binary Relevance and Label Power-set: (1) a label classification is used to filter input instances into independent labels; (2) the algorithm will map the labels into an hyper-label space, where each hyper-label represents the behaviour which maximizes input instance correlations. This method enables a robot to capture user typical behaviour and identify abnormal situations that might demand a range of care-giver interventions.