"Deep Variational Bayes Filters: Unsupervised Learning of State Space Models from Raw Data",
Maximilian Karl, Maximilian Soelch, Justin Bayer, Patrick van der Smagt, ICLR2017.
[Link] https://arxiv.org/abs/1605.06432
1. Terry Taewoong Um (terry.t.um@gmail.com)
University of Waterloo
Department of Electrical & Computer Engineering
Terry Taewoong Um
DEEP VARIATIONAL BAYES FILTERS
: UNSUPERVISED LEARNING OF STATE
SPACE MODELS FROM RAW DATA
1
2. 2
WHAT’S WRONG WITH RNN (LSTM)?
Terry Taewoong Um (terry.t.um@gmail.com)
WHAT’S WRONG WITH AUTOENCODER?
3. 3
LIN. REGRE : BAYES. LIN. REGRE.
= AE : VAE
= RNN : VARIATIONAL RNN
Terry Taewoong Um (terry.t.um@gmail.com)
From “PR12: Variational Autoencoder” by Cha
4. 4
VAE / HMM / KF / RNN
Terry Taewoong Um (terry.t.um@gmail.com)
Latent space
(q_0,…,q_N)
Observation
space (x,y,z)
Latent
t t+1 t+2emission
transition
Hidden Markov Model : discrete states, stochastic transition/emission
Kalman filter : continuous states, stochastic linear transition/emission with Gaussians
Recurrent Neural Networks : deterministic (not good for learning probabilistic densities)
t+2
You should define the model a priori!
X
Z
5. 5
KALMAN FILTER
Terry Taewoong Um (terry.t.um@gmail.com)
Latent space
(q_0,…,q_N)
Observation
space (x,y,z)
Latent
t t+1 t+2
emission
transition
t+2
X
Z
U
[Limitations] (1) Its assumptions are restrictive
(2) The model (F, B, H) has to be known
7. 7
VAE AND VARIATIONAL RNN
Terry Taewoong Um (terry.t.um@gmail.com)
• “Structured Inference Networks for Nonlinear State Space Models”,
R. Krishnan, U. Shalit, and D. Sontag, AAAI2017
8. 8
VAE AND VARIATIONAL RNN
Terry Taewoong Um (terry.t.um@gmail.com)
• “Deep Kalman Filter”, R. Krishnan, U. Shalit, and D. Sontag,
NIPS2016
9. 9
REFERENCES
Terry Taewoong Um (terry.t.um@gmail.com)
• “Deep Kalman Filter”, R. Krishnan, U. Shalit, D. Sontag, NIPS2016
• “Structured Inference Networks for Nonlinear State Space Models”,
R. Krishnan, U. Shalit, and D. Sontag, AAAI2017
• “Learning Stochastic Recurrent Networks”, J. Bayer, C. Osendorfer, ICLR2015
• “Recurrent Latent Variable Model for Sequential Data”, J. Chung, K.
Kastner, L. Dinh, K. Goel, A. Courville, Y. Bengio, NIPS2015
• “Variational Bayes Filters”, M. Karl, M. Soelch, J. Bayer, P. Smagt,
ICLR2017
• “A Disentangled Recognition and Nonlinear Dynamics Model for
Unsupervised Learning”, M. Fraccaroy, S. Kamronn, U. Paquetz, O. Winthery,
NIPS2017
19. 19
EXPERIMENTS
Terry Taewoong Um (terry.t.um@gmail.com)
http://blog.fastforwardlabs.com/2016/08/12/introdu
cing-variational-autoencoders-in-prose-and.html
22. 22
DEEP HMM
Terry Taewoong Um (terry.t.um@gmail.com)
• “Structured Inference Networks for
Nonlinear State Space Models”, AAAI2017
23. 23
KALMAN VAE
“Deep Variational Bayes Filters”, M. Karl et al., ICLR2017
• “A Disentangled Recognition and
Nonlinear Dynamics Model for
Unsupervised Learning”, M. Fraccaroy, S.
Kamronn, U. Paquetz, O. Winthery,
NIPS2017
24. 24
REFERENCES
Terry Taewoong Um (terry.t.um@gmail.com)
• “Deep Kalman Filter”, R. Krishnan, U. Shalit, D. Sontag, NIPS2016
• “Structured Inference Networks for Nonlinear State Space Models”,
R. Krishnan, U. Shalit, and D. Sontag, AAAI2017
• “Learning Stochastic Recurrent Networks”, J. Bayer, C. Osendorfer, ICLR2015
• “Recurrent Latent Variable Model for Sequential Data”, J. Chung, K.
Kastner, L. Dinh, K. Goel, A. Courville, Y. Bengio, NIPS2015
• “Variational Bayes Filters”, M. Karl, M. Soelch, J. Bayer, P. Smagt,
ICLR2017
• “A Disentangled Recognition and Nonlinear Dynamics Model for
Unsupervised Learning”, M. Fraccaroy, S. Kamronn, U. Paquetz, O. Winthery,
NIPS2017