BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:An Empirical Study of Lifelong Learning on Mobile Applications  - 
 Young D. Kwon\, Hong Kong University of Science and Technology
DTSTART:20191203T131500Z
DTEND:20191203T134500Z
UID:TALK135718@talks.cam.ac.uk
CONTACT:Seyyed Ahmad Javadi
DESCRIPTION:Deep neural networks achieve state-of-the-art performances on 
 sensor data generated by a wide variety of mobile applications. However\, 
 the capability of deep learning models to perform lifelong learning (conti
 nual learning)\, that is\, to learn new inputs (new tasks or classes) cont
 inuously\, is often impaired by catastrophic forgetting\, i.e.\, a model f
 orgets previously learned knowledge when learning new tasks. This can be v
 ery detrimental in ubiquitous computing\, where a deployed model needs to 
 accommodate new sensor inputs and changing environments continuously. Seve
 ral techniques have been proposed to solve catastrophic forgetting\, but t
 heir performance has not been fully examined in mobile sensing application
 s. In this talk\, for the first time\, we systematically study the perform
 ance of three predominant lifelong learning schemes (i.e.\, regularization
 \, replay and replay with examples) on mobile sensing applications of huma
 n activity recognition and gesture recognition. With scenarios consisting 
 of different learning complexity\, as encountered in practice\, we investi
 gate the generalizability\, trade-offs between the performance\, storage\,
  and latency of different lifelong learning methods on mobile sensing appl
 ications. Finally\, we summarize our results into a series of lessons that
  can guide practitioners in their use of lifelong deep learning for mobile
  applications.
LOCATION:Computer Laboratory\, William Gates Building\, Room FW11
END:VEVENT
END:VCALENDAR