BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Bayesian Deep Learning to Predict Air Pollution &amp\; Personalize
 d Air Pollution Monitoring and Health Management - Victor O.K. Li &amp\; J
 acqueline CK Lam\, The University of Hong Kong
DTSTART:20180620T131500Z
DTEND:20180620T143000Z
UID:TALK106801@talks.cam.ac.uk
CONTACT:Hatice Gunes
DESCRIPTION:HKU and Cambridge colleagues\, working together in HKU-Cambrid
 ge CEERP\, have recently won a major grand challenge grant on AI and Air P
 ollution Monitoring and Health Management. These two talks will be providi
 ng insight into the research work conducted so far as part of this project
 .\n\nThe project addresses five major challenges. FIRST\, urban air qualit
 y data is sparse\, rendering it difficult to provide timely personalized a
 lert and advice. SECOND\, collected data\, especially those involving huma
 n inputs\, such as health perception\, are often missing and erroneous. TH
 IRD\, data collected are heterogeneous\, and highly complex\, not easily c
 omprehensible to facilitate individual or collective decision-making. FOUR
 TH\, the causal relationships between personal air pollutants exposure (sp
 ecifically PM(2.5\,1.0) and NO2) and personal health conditions\, and heal
 th (well-being) perception\, of young asthmatics and young healthy citizen
 s\, are yet to be established. FIFTH\, one must determine if information a
 nd advice provided can effect behavioral change.  \n\nAir pollution has de
 teriorated rapidly in many metropolitan cities\, such as Beijing. Since po
 or air quality has clear public health impacts\, accurately monitoring and
  predicting the concentration of PM2.5 and other pollutants have become in
 creasingly crucial. This talk presents a hybrid approach where time series
  decomposition and Bayesian Long Short-Term Memory (BLSTM) are combined as
  a framework for air pollution forecast\, based on historical data of air 
 quality\, meteorology and traffic in Beijing. LSTM has been proven to achi
 eve state-of-the-art performance in many time series prediction applicatio
 ns due to its capability of memorizing long term sequential correlations. 
 In addition\, the model uncertainty estimates generated by Bayesian method
 s may reduce overfitting\, improving the accuracy of the prediction. In ou
 r experiment\, deseasonalized features are fed into BLSTM to predict the a
 ir pollution in the next 48 hours of each monitoring station in Beijing. R
 esults show that the BLSTM framework outperforms the baseline models inclu
 ding SVR\, STL\, ARIMA\, and traditional LSTM with dropout regularization.
LOCATION:FW11 Meeting Room\, Computer Laboratory
END:VEVENT
END:VCALENDAR