Lasse Holmström
University of Oulu, Department of Mathematical Sciences, Faculty Member
Research Interests:
Pattern classification using neural networks and statistical methods is discussed. We first give a tutorial overview that groups popular classifiers according to their underlying mathematical principles into several distinct categories.... more
Pattern classification using neural networks and statistical methods is discussed. We first give a tutorial overview that groups popular classifiers according to their underlying mathematical principles into several distinct categories. Starting from the Bayes classifier, one division is whether the classifier is explicitly estimating class conditional densities, or directly estimating the posterior probabilities by regression. Another criterion is the flexibility of the architecture in the sense of how rich the discriminant function family is. Still one dimension is neural vs. nonneural learning: neural learning is characterized by simple local computations in a number of real or virtual processing elements. Based on these comparisons, a number of classification methods were selected for a case study that uses handwritten digit data. An effort was made to get fair estimates of their true classification performance, thus training set cross-validation was extensively used to design t...
We report a method for reconstructing temperatures from proxy data based on Bayesian multi-scale smoothing. The main features of the method is that it makes possible a simultaneous utilization of multiple proxies and sites to achieve a... more
We report a method for reconstructing temperatures from proxy data based on Bayesian multi-scale smoothing. The main features of the method is that it makes possible a simultaneous utilization of multiple proxies and sites to achieve a regional `consensus' reconstruction of temperature, it enables various types of uncertainties (e.g., chronology) to be taken into account, and that it also allows significant features to be detected from the emerging time series. Applying the method to multiple proxy data derived from lakes in Finnish Lapland suggests that, contrary to the conventional view, Holocene climate cooling in the North Atlantic region started already c. 8.5 kyr, and that climate since then was punctuated by several quasi-cyclical climate events, the forcing mechanisms of which are not yet fully understood. Our study suggests that inconsistencies in climate reconstructions between different proxy sources may be at least partly spurious; there is probably no single narrati...
Research Interests:
We give an overview of pattern recognition, concentrating on the problem of pattern classification. Several popular discrimination methods are reviewed using decision theory as a unifying framework. Copyright © 2010 John Wiley & Sons,... more
We give an overview of pattern recognition, concentrating on the problem of pattern classification. Several popular discrimination methods are reviewed using decision theory as a unifying framework. Copyright © 2010 John Wiley & Sons, Inc.For further resources related to this article, please visit the WIREs website.
This report describes an electromagnetic scattering model for the computation of remote sensing quantities from a forested region. Several scattering models have been developed in the past with di#ering degrees of complexity. A large... more
This report describes an electromagnetic scattering model for the computation of remote sensing quantities from a forested region. Several scattering models have been developed in the past with di#ering degrees of complexity. A large class of models utilizes a radiative transfer formulation to compute scattering cross sections (see [12] for a general overview). An example is the model used by Pullianen et al. [9, 10] for comparison with microwave measurements of boreal forests. This model can be extended to handle quite complicated layer structures, as with the Michigan Microwave Canopy Scattering Model (MIMICS) applied in [7] and [14]. While the radiative transfer approach is valid and can yield good comparisons with data, it ignores the coherent scattering that may occur within the tree structure. A discrete scatterer model has been used in [1] to model the volume scattering and a branching model was employed in [16] to account for structural e#ects. The random nature of the trees...
LASSO is a penalized regression method that facilitates model fitting in situations where there are as many, or even more explanatory variables than observations, and only a few variables are relevant in explaining the data. We focus on... more
LASSO is a penalized regression method that facilitates model fitting in situations where there are as many, or even more explanatory variables than observations, and only a few variables are relevant in explaining the data. We focus on the Bayesian version of LASSO and consider four problems that need special attention: (i) controlling false positives, (ii) multiple comparisons, (iii) collinearity among explanatory variables, and (iv) the choice of the tuning parameter that controls the amount of shrinkage and the sparsity of the estimates. The particular application considered is association genetics, where LASSO regression can be used to find links between chromosome locations and phenotypic traits in a biological organism. However, the proposed techniques are relevant also in other contexts where LASSO is used for variable selection. We separate the true associations from false positives using the posterior distribution of the effects (regression coefficients) provided by Bayesi...
Research Interests:
Probability density estimation using the probabilistic neural network or the kernel method is considered. In its basic form this method can be computationally prohibitive, as all training data need to be stored and each individual... more
Probability density estimation using the probabilistic neural network or the kernel method is considered. In its basic form this method can be computationally prohibitive, as all training data need to be stored and each individual training vector gives rise to a new term of the estimate. Given an original training sample of size N in a d-dimensional space, a simple
Research Interests:
Research Interests:
This paper is concerned with detecting image features that appear in different scales or resolutions. A new ap- proach is proposed that uses Bayesian statistical modeling and simulation based inference. The method can be viewed as a... more
This paper is concerned with detecting image features that appear in different scales or resolutions. A new ap- proach is proposed that uses Bayesian statistical modeling and simulation based inference. The method can be viewed as a further development of SiZer technology, originally de- signed for nonparametric curve fitting. A strength of the Bayesian approach is its flexibility that facilitates
Research Interests:
Research Interests:
Research Interests:
A density estimation approach to statistical pattern recognition is discussed. The pattern vector is split into two parts factoring a high dimensional class density function into a product of two lower dimensional density functions. The... more
A density estimation approach to statistical pattern recognition is discussed. The pattern vector is split into two parts factoring a high dimensional class density function into a product of two lower dimensional density functions. The first factor, corresponding to the non-Gaussian structure in the data, is modeled nonparametrically. The second factor is modeled as a multivariate Gaussian conditionally on the
Research Interests:
Research Interests:
We report a method for reconstructing temperatures from proxy data based on Bayesian multi-scale smoothing. The main features of the method is that it makes possible a simultaneous utilization of multiple proxies and sites to achieve a... more
We report a method for reconstructing temperatures from proxy data based on Bayesian multi-scale smoothing. The main features of the method is that it makes possible a simultaneous utilization of multiple proxies and sites to achieve a regional `consensus' reconstruction of temperature, it enables various types of uncertainties (e.g., chronology) to be taken into account, and that it also allows significant features to be detected from the emerging time series. Applying the method to multiple proxy data derived from lakes in Finnish Lapland suggests that, contrary to the conventional view, Holocene climate cooling in the North Atlantic region started already c. 8.5 kyr, and that climate since then was punctuated by several quasi-cyclical climate events, the forcing mechanisms of which are not yet fully understood. Our study suggests that inconsistencies in climate reconstructions between different proxy sources may be at least partly spurious; there is probably no single narrat...
Research Interests:
The computational cost of multivariate kernel density estimation can be reduced by prebinning the data. The data are discretized to a grid and a weighted kernel estimator is computed. We report results on the accuracy of such a binned... more
The computational cost of multivariate kernel density estimation can be reduced by prebinning the data. The data are discretized to a grid and a weighted kernel estimator is computed. We report results on the accuracy of such a binned kernel estimator and discuss the computational complexity of the estimator as measured by its average number of nonzero terms.