A hybrid approach of deep learning and hidden Markov models (DL-HMM) for sequential activity recognition.

In this talk, I first will highlight how our team during my fellowship at the Data Incubator developed a hybrid approach to deep learning and hidden Markov models (DL-HMM) for bio-sensor data coming from a triaxial accelerometer to classify the data and extract intrinsic features out of it. I’ll explain how our model provided better recognition accuracy of human activities and avoided the expensive design of handcrafted features in existing systems by utilizing the massive unlabeled acceleration samples for unsupervised feature extraction.