Modeling Time-Series
with Deep Networks
MARTIN LÄNGKVIST
Technology
Örebro Studies in Technology 63 I
ÖREBRO 20142014
M
AR
TIN
L
ÄN
GK
VIS
T
M
od
eli
ng T
im
e-S
eri
es w
ith Dee
p N
etw
ork
s
martin längkvist received his MSc (2009) in Applied Physics and Elec-trical Engineering from The Institute of Technology, Linköping University, Linköping, Sweden. In 2009-2014 he has been a graduate student at the Center for Applied Autonomous Sensor Systems, Örebro University, Örebro, Sweden. During winters 2012-2014 he visited SASTRA University, Thanjavur, Tamil Nadu, India for research collaborations. His research interests include machine learning with particular focus on developing deep learning methods for time-series data applied to medical applications.
Deep learning is a relatively new field that has shown promise in a number of applications and is currently outperforming other algorithms on a variety of commonly used benchmark data sets. Much focus in deep learning research has been on applications and methods for static data and not so much on time-series data. Learning models of complex high-dimensional time-series data introduces a number of challenges that either require modifications to the learning algorithms or special pre-processing of the data. The contribu-tions presented in this thesis focus around the design of algorithms inspired from deep learning for learning feature representations from non-filtered multivariate time-series data.
issn 1650-8580 isbn 978-91-7529-054-6
