- Luttinen Jaakko. BayesPy: Variational Bayesian Inference in Python. Accepted for publication in Journal of Machine Learning Research, ?, 2015. [BibTeX]
@article{
Luttinen15jmlr,
author = "Luttinen, Jaakko",
volume = "?",
juforank = "",
doi = "",
language = "eng",
title = "{BayesPy:} Variational {Bayesian} Inference in {Python}",
pdf = "",
country = "",
journal = "Accepted for publication in Journal of Machine Learning Research",
issn = "",
publisher = "",
number = "",
abstract = "",
responsibleauthor = "Luttinen, Jaakko",
corerank = "",
flags = "public COIN",
il = "",
year = "2015",
unitcode = "T306-100",
kay = "NA",
impactfactor = "",
pages = "?"
}
- Luttinen Jaakko. Bayesian Latent Gaussian Spatio-Temporal Models. PhD thesis, 2015. Aalto University. [BibTeX]
@phdthesis{
Luttinen15phd,
author = "Luttinen, Jaakko",
pdf = "",
school = "Aalto University",
year = "2015",
flags = "public COIN",
title = "Bayesian Latent {Gaussian} Spatio-Temporal Models"
}
- Luttinen Jaakko, Raiko Tapani, Ilin Alexander. Linear State-Space Model with Time-Varying Dynamics. In Machine Learning and Knowledge Discovery in Databases, ECML/PKDD'2014, volume 8725 of Lecture Notes in Computer Science, pages 338-353, 2014. Springer [More info] [Pdf] [BibTeX]
@inproceedings{
Luttinen14ecml,
author = "Luttinen, Jaakko and Raiko, Tapani and Ilin, Alexander",
isbn = "978-3-662-44851-9",
series = "Lecture Notes in Computer Science",
abstract = "This paper introduces a linear state-space model with time-varying dynamics. The time dependency is obtained by forming the state dynamics matrix as a time-varying linear combination of a set of matrices. The time dependency of the weights in the linear combination is modelled by another linear Gaussian dynamical model allowing the model to learn how the dynamics of the process changes. Previous approaches have used switching models which have a small set of possible state dynamics matrices and the model selects one of those matrices at each time, thus jumping between them. Our model forms the dynamics as a linear combination and the changes can be smooth and more continuous. The model is motivated by physical processes which are described by linear partial differential equations whose parameters vary in time. An example of such a process could be a temperature field whose evolution is driven by a varying wind direction. The posterior inference is performed using variational Bayesian approximation. The experiments on stochastic advection-diffusion processes and real-world weather processes show that the model with time-varying dynamics can outperform previously introduced approaches.",
il = "no",
eventdetails = "EVENT DETAILS",
year = "2014",
title = "Linear State-Space Model with Time-Varying Dynamics",
booktitle = "Machine Learning and Knowledge Discovery in Databases, ECML/PKDD'2014",
juforank = "NA",
unitcode = "T306-100",
impactfactor = "A4",
responsibleauthor = "Luttinen, Jaakko",
volume = "8725",
address = "Berlin Heidelberg",
kay = "NA",
pages = "338--353",
publisher = "Springer",
doi = "10.1007/978-3-662-44851-9_22",
language = "eng",
url = "http://dx.doi.org/10.1007/978-3-662-44851-9_22",
corerank = "",
flags = "public COIN",
pdf = "luttinen_ecml14.pdf"
}
- Luttinen Jaakko. Fast Variational Bayesian Linear State-Space Model. In Machine Learning and Knowledge Discovery in Databases, ECML/PKDD'2013, volume 8188 of Lecture Notes in Computer Science, pages 305-320, 2013. Springer [More info] [Pdf] [BibTeX]
@inproceedings{
Luttinen13ecml,
author = "Luttinen, Jaakko",
isbn = "978-3-642-40987-5",
series = "Lecture Notes in Computer Science",
abstract = "This paper presents a fast variational Bayesian method for linear state-space models. The standard variational Bayesian expectation-maximization (VB-EM) algorithm is improved by a parameter expansion which optimizes the rotation of the latent space. With this approach, the inference is orders of magnitude faster than the standard method. The speed of the proposed method is demonstrated on an artificial dataset and a large real-world dataset, which shows that the standard VB-EM algorithm is not suitable for large datasets because it converges extremely slowly. In addition, the paper estimates the temporal state variables using a smoothing algorithm based on the block LDL decomposition. This smoothing algorithm reduces the number of required matrix inversions and avoids a model augmentation compared to previous approaches.",
il = "no",
eventdetails = "EVENT DETAILS",
year = "2013",
title = "Fast Variational {Bayesian} Linear State-Space Model",
booktitle = "Machine Learning and Knowledge Discovery in Databases, ECML/PKDD'2013",
juforank = "NA",
unitcode = "T306-100",
impactfactor = "A4",
responsibleauthor = "Luttinen, Jaakko",
volume = "8188",
address = "Berlin Heidelberg",
kay = "NA",
pages = "305--320",
publisher = "Springer",
doi = "10.1007/978-3-642-40988-2_20",
language = "eng",
url = "http://dx.doi.org/10.1007/978-3-642-40988-2_20",
corerank = "",
flags = "public COIN",
pdf = "luttinen_ecml13.pdf"
}
- Luttinen Jaakko, Ilin Alexander, Karhunen Juha. Bayesian Robust PCA of Incomplete Data. Neural Processing Letters, 36(2):189-202, 2012. [More info] [Pdf] [BibTeX]
@article{
Luttinen12npl,
author = "Luttinen, Jaakko and Ilin, Alexander and Karhunen, Juha",
volume = "36",
juforank = "1",
doi = "10.1007/s11063-012-9230-4",
language = "eng",
title = "{Bayesian} Robust {PCA} of Incomplete Data",
pdf = "luttinen_npl12.pdf",
country = "",
journal = "Neural Processing Letters",
issn = "1573-773X",
publisher = "Springer US",
number = "2",
abstract = "We present a probabilistic model for robust factor analysis and principal component analysis in which the observation noise is modeled by Student- t distributions in order to reduce the negative effect of outliers. The Student- t distributions are modeled independently for each data dimensions, which is different from previous works using multivariate Student- t distributions. We compare methods using the proposed noise distribution, the multivariate Student- t and the Laplace distribution. Intractability of evaluating the posterior probability density is solved by using variational Bayesian approximation methods. We demonstrate that the assumed noise model can yield accurate reconstructions because corrupted elements of a bad quality sample can be reconstructed using the other elements of the same data vector. Experiments on an artificial dataset and a weather dataset show that the dimensional independency and the flexibility of the proposed Student- t noise model can make it superior in some applications.",
responsibleauthor = "Luttinen, Jaakko",
corerank = "B",
flags = "public COIN",
il = "no",
year = "2012",
unitcode = "T306-100",
impactfactor = "A1",
pages = "189--202"
}
- Luttinen Jaakko, Ilin Alexander. Efficient Gaussian Process Inference for Short-Scale Spatio-Temporal Modeling. In JMLR Workshop and Conference Proceedings, AISTATS'2012, pages 741-750, 2012. [More info] [Pdf] [BibTeX]
@inproceedings{
Luttinen12aistats,
author = "Luttinen, Jaakko and Ilin, Alexander",
volume = "22",
juforank = "NA",
eventtime = "April 21-23",
language = "eng",
title = "Efficient {Gaussian} Process Inference for Short-Scale Spatio-Temporal Modeling",
eventlocation = "La Palma, Canary Islands",
country = "",
booktitle = "JMLR Workshop and Conference Proceedings, AISTATS'2012",
corerank = "B",
pdf = "luttinen_aistats12.pdf",
abstract = "This paper presents an efficient Gaussian process inference scheme for modeling shortscale phenomena in spatio-temporal datasets. Our model uses a sum of separable, compactly supported covariance functions, which yields a full covariance matrix represented in terms of small sparse matrices operating either on the spatial or temporal domain. The proposed inference procedure is based on Gibbs sampling, in which samples from the conditional distribution of the latent function values are obtained by applying a simple linear transformation to samples drawn from the joint distribution of the function values and the observations. We make use of the proposed model structure and the conjugate gradient method to compute the required transformation. In the experimental part, the proposed algorithm is compared to the standard approach using the sparse Cholesky decomposition and it is shown to be much faster and computationally feasible for 100-1000 times larger datasets. We demonstrate the advantages of the proposed method in the problem of reconstructing sea surface temperature, which requires processing of a real-world dataset with 10^6 observations.",
responsibleauthor = "Luttinen, Jaakko",
url = "http://jmlr.csail.mit.edu/proceedings/papers/v22/luttinen12.html",
flags = "public COIN",
il = "no",
year = "2012",
unitcode = "T306-100",
impactfactor = "A4",
pages = "741--750"
}
- Luttinen Jaakko, Ilin Alexander, Karhunen Juha. Bayesian Robust PCA for Incomplete Data. In Proceedings of the 8th International Conference on Independent Component Analysis and Blind Signal Separation, ICA'2009, pages 66-73, 2009. [More info] [Pdf] [BibTeX]
@inproceedings{
Luttinen09rpca,
author = "Luttinen, Jaakko and Ilin, Alexander and Karhunen, Juha",
doi = "10.1007/978-3-642-00599-2_9",
title = "{B}ayesian Robust {PCA} for Incomplete Data",
booktitle = "Proceedings of the 8th International Conference on Independent Component Analysis and Blind Signal Separation, ICA'2009",
address = "Paraty, Brazil",
corerank = "NA",
abstract = "We present a probabilistic model for robust principal component analysis (PCA) in which the observation noise is modelled by Student-$t$ distributions that are independent for different data dimensions. A heavy-tailed noise distribution is used to reduce the negative effect of outliers. Intractability of posterior evaluation is solved using variational Bayesian approximation methods. We show experimentally that the proposed model can be a useful tool for PCA preprocessing for incomplete noisy data. We also demonstrate that the assumed noise model can yield more accurate reconstructions of missing values: Corrupted dimensions of a ``bad'' sample may be reconstructed well from other dimensions of the same data vector. The model was motivated by a real-world weather dataset which was used for comparison of the proposed technique to relevant probabilistic PCA models.",
month = "March",
responsibleauthor = "Luttinen, Jaakko",
flags = "public AIRC",
year = "2009",
keywords = "",
pdf = "luttinen_ica09.pdf",
impactfactor = "D3",
pages = "66--73"
}
- Luttinen Jaakko, Ilin Alexander, Raiko Tapani. Transformations for Variational Factor Analysis to Speed up Learning. In Proceedings of the 14th European Symposium on Artificial Neural Networks, ESANN'2009, pages 77-82, 2009. [More info] [Pdf] [BibTeX]
@inproceedings{
Luttinen09esann,
author = "Luttinen, Jaakko and Ilin, Alexander and Raiko, Tapani",
title = "Transformations for Variational Factor Analysis to Speed up Learning",
url = "http://eprints.pascal-network.org/archive/00006403/",
booktitle = "Proceedings of the 14th {E}uropean Symposium on Artificial Neural Networks, ESANN'2009",
address = "Bruges, Belgium",
corerank = "NA",
abstract = "We propose simple transformation of the hidden states in variational Bayesian (VB) factor analysis models to speed up the learning procedure. The transformation basically performs centering and whitening of the hidden states taking into account the posterior uncertainties. The transformation is given a theoretical justification from optimisation of the VB cost function. We derive the transformation formulae for variational Bayesian principal component analysis and show experimentally that it can significantly improve the rate of convergence. Similar transformations can be applied to other variational Bayesian factor analysis models as well.",
month = "April",
responsibleauthor = "Luttinen, Jaakko",
flags = "public AIRC",
year = "2009",
keywords = "",
pdf = "luttinen_esann09.pdf",
impactfactor = "D3",
pages = "77--82"
}
- Luttinen Jaakko, Ilin Alexander. Variational Gaussian-Process Factor Analysis for Modeling Spatio-Temporal Data. In Advances in Neural Information Processing Systems 22, NIPS'2009, pages 1177-1185, 2009. MIT Press [More info] [Pdf] [BibTeX]
@inproceedings{
Luttinen09nips,
author = "Luttinen, Jaakko and Ilin, Alexander",
publisher = "MIT Press",
title = "Variational {Gaussian}-Process Factor Analysis for Modeling Spatio-Temporal Data",
url = "http://papers.nips.cc/paper/3805-variational-gaussian-process-factor-analysis-for-modeling-spatio-temporal-data",
booktitle = "Advances in Neural Information Processing Systems 22, NIPS'2009",
address = "Cambridge, MA, USA",
corerank = "A+",
abstract = "We present a probabilistic factor analysis model which can be used for studying spatio-temporal datasets. The spatial and temporal structure is modeled by using Gaussian process priors both for the loading matrix and the factors. The posterior distributions are approximated using the variational Bayesian framework. High computational cost of Gaussian process modeling is reduced by using sparse approximations. The model is used to compute the reconstructions of the global sea surface temperatures from a historical dataset. The results suggest that the proposed model can outperform the state-of-the-art reconstruction systems.",
responsibleauthor = "Luttinen, Jaakko",
flags = "public AIRC",
year = "2009",
keywords = "",
pdf = "luttinen_nips2009.pdf",
impactfactor = "D3",
pages = "1177--1185"
}
- Luttinen Jaakko. Gaussian-process factor analysis for modeling spatio-temporal data. Master's thesis, 2009. Helsinki University of Technology. [Pdf] [BibTeX]
@mastersthesis{
Luttinen:2009:MSc,
author = "Luttinen, Jaakko",
school = "Helsinki University of Technology",
title = "Gaussian-process factor analysis for modeling spatio-temporal data",
abstract = "The main theme of this thesis is analyzing and modeling large spatio-temporal datasets, such as global temperature measurements. The task is typically to extract relevant structure and features for predicting or studying the system. This can be a challenging problem because simple models are often not able to capture the complex structure suffiently well, and more sophisticated models can be computationally too expensive in practice. \par This thesis presents a novel spatio-temporal model which extends factor analysis by setting Gaussian process priors over the spatial and temporal components. In contrast to factor analysis, the presented model is capable of modeling complex spatial and temporal structure. Compared to standard Gaussian process regression over the spatio-temporal domain, the presented model gains substantial computational savings by operating only in the spatial or temporal domain at a time. Thus, it is feasible to model larger spatio-temporal datasets than with standard Gaussian process regression. \par The new model combines the modeling assumptions of several traditional techniques used for analyzing spatially and temporally distributed data: kriging is used for modeling spatial dependencies; empirical orthogonal functions reduce the dimensionality of the problem; and temporal smoothing finds relevant features from time series. \par The model is applied to reconstruct missing values in a historical sea surface temperature dataset. The results are promising and suggest that the proposed model may outperform the state-of-the-art reconstruction systems.",
corerank = "NA",
month = "December",
responsibleauthor = "Luttinen, Jaakko",
flags = "public",
year = "2009",
pdf = "luttinen_msc.pdf",
impactfactor = "E"
}
