Banner
Home      Log In      Contacts      FAQs      INSTICC Portal
 

Keynote Lecture

 

Event Detection and Classification in Big IoT Networks from Scarce and Imprecise Labels

Zoran Obradovic
Temple University
United States
 

Brief Bio
Zoran Obradovic is a Distinguished Professor and a Center director at Temple University, an Academician at the Academia Europaea (the Academy of Europe) and a Foreign Academician at the Serbian Academy of Sciences and Arts. He mentored 45 postdoctoral fellows and Ph.D. students, many of whom have independent research careers at academic institutions (e.g. Northeastern Univ., Ohio State Univ,) and industrial research labs (e.g. Amazon, Facebook, Hitachi Big Data, IBM T.J.Watson, Microsoft, Yahoo Labs, Uber, Verizon Big Data, Spotify). Zoran is the editor-in-chief at the Big Data journal and the steering committee chair for the SIAM Data Mining conference. He is also an editorial board member at 13 journals and was the general chair, program chair, or track chair for 11 international conferences. His research interests include data science and complex networks in decision support systems addressing challenges related to big, heterogeneous, spatial and temporal data analytics motivated by applications in healthcare management, power systems, earth and social sciences. His studies were funded by AFRL, DARPA, DOE, KAUST, NIH, NSF, ONR, and the PA Department of Health and industry. For more details see http://www.dabi.temple.edu/zoran-obradovic


Abstract
Accurate predictions at multiple temporal and spatial scales from many IoT devices can potentially enable innovations across various industries. For example, moving from corrective to predictive maintenance of complex infrastructure based on knowledge extracted by many IoT instruments could be more cost effective since this can facilitate early and interpretable risk predictions with uncertainty estimates and allow optimization of damage mitigation and prevention strategies. Similarly, in proactive emergency monitoring, IoT network could estimate operating conditions before they occur, which can direct deployment of control measures for avoiding undesirable outcomes. In this talk an overview of our recently developed methods to facilitate such end-to-end solutions will be discussed within the context of our ongoing DOE funded project aimed at predictive analytics in a large electricity grid from multiple phasor measurement units.  Challenges and the proposed solutions will be discussed related to (1) deep-learning based detection and classification of local and system-wide events using rapidly refined, partially inspected event labels; (2) digital-twin based data enhancement for events insufficiently represented in field-recordings over the training period; and (3) transfer learning to leverage relevant labeled events from a different network to minimize additional labeling effort. 



footer