Recently, the multimoal and high dimensional sensor data are prone to problems such as artificial error and time-consuming acquisition processes, especially in supervised human activity recognition. Therefore, this study proposes an activity recognition framework called compositional Bidir-LSTM-CNN Networks, which automatically extracts features from raw data using the optimized Convolutional Neural Network and further capture dynamic temporal features through the Bidirectional Lone Short Term Memory Network. Finally, this study paves the way for accurate recognition of human activities using the proposed framework with significantly improve 8% recognition accuracy along with additional features such as robustness and generalization.
Previously, a lot of applications on recurrent neural networks with multi-valued neuron (MVNRNNs) in asynchronous update mode, used zero diagonal elements as network stability condition. However, according to our recent finding, only positive diagonal elements can guarantee the whole network to be complete convergent. How to adopt positive diagonal element to amend former research results is still unknown. This paper tries to investigate this question by using MVNRNNs with different positive diagonal elements in associative memory (AM) design. Simulation results illustrate that choosing appropriate positive diagonal elements can promote AM's performance compared with AM employing zero diagonal elements.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.