Deep Learning for Sensing in Complex and Dynamic Environments

The deep learning techniques have revolutionized the sensing field over the past few years and have outperformed the human experts on a wide range of tasks. However, training and deploying deep learning models for complex sensing tasks in dynamic scenarios remains a challenge, especially when labeled data are scarce.

In this talk, we focus on the problem of complex event detection over heterogeneous sensory data. We first propose DeepCEP, a neural-symbolic framework that combines neural network models with Complex Event Processing (CEP) engines. DeepCEP encodes prior knowledge provided by the users to help make effective inferences over the long-term, complex events. To enable the learning of this neural-symbolic system, we introduce Neuroplex, which trains the neural network models efficiently using high-level, complex event labels, with prior knowledge guidance. Compared with mainstream deep learning models, Neuroplex reduces the data annotation requirement by 100 times and speeds up the learning process by four times. Furthermore, we propose the generalized framework DeepSQA for flexible inferencing over heterogeneous sensory data. We extensively evaluate our proposed models on different real-life datasets, and the empirical results show the effectiveness, reliability, and robustness of the proposed models.

Biography:
Tianwei Xing is a Ph.D. candidate in Electrical & Computer Engineering at the University of California, Los Angeles, under the supervision of Prof. Mani Srivastava. Before this, he obtained his Master’s degree at UCLA and a Bachelor’s degree at Zhejiang University. Tianwei held multiple internship positions at Samsung Research America and IBM Research. His research interests include deep cross-modal knowledge transfer, time-series modeling, and Neural-symbolic AI.