Experiment: Cooperative Decision Making via Reinforcement Learning
Berka, Milan
2018 - English This report inspects cooperative decision making task using reinforcement learning. It serves for comparison with methodology based on fully probabilistic design of decision strategies. Keywords: decision making; reinforcement learning; cooperation Fulltext is available at external website. Experiment: Cooperative Decision Making via Reinforcement Learning

This report inspects cooperative decision making task using reinforcement learning. It serves for comparison with methodology based on fully probabilistic design of decision strategies.

Berka, Milan
Ústav teorie informace a automatizace, 2018

Balancing Exploitation and Exploration via Fully Probabilistic Design of Decision Policies
Kárný, Miroslav; Hůla, František
2018 - English Adaptive decision making learns an environment model serving a design of a decision policy. The policy-generated actions influence both the acquired reward and the future knowledge. The optimal policy properly balances exploitation with exploration. The inherent dimensionality\ncurse of decision making under incomplete knowledge prevents the realisation of the optimal design. Keywords: Exploitation; Exploration; Bayesian estimation; Adaptive systems; Fully probabilistic design; Kullback-Leibler divergence; Decision policy; Markov decision process Fulltext is available at external website. Balancing Exploitation and Exploration via Fully Probabilistic Design of Decision Policies

Adaptive decision making learns an environment model serving a design of a decision policy. The policy-generated actions influence both the acquired reward and the future knowledge. The optimal policy ...

Kárný, Miroslav; Hůla, František
Ústav teorie informace a automatizace, 2018

Appearance Acquisition and Analysis of Effect Coatings
Filip, Jiří; Maile, F. J.
2017 - English Keywords: effect coatings; appearance capturing; polychromatic; particle orientation Fulltext is available at external website. Appearance Acquisition and Analysis of Effect Coatings

Filip, Jiří; Maile, F. J.
Ústav teorie informace a automatizace, 2017

Multi-period Factor Model of a Loan Portfolio
Šmíd, Martin; Dufek, J.
2017 - English We construct a general dynamic model of losses of a large loan portfolio, secured by collaterals. In the model, the wealth of a debtor and the price of the corresponding collateral depend each on two factors: a common one, having a general distribution, and an individual one, following an AR(1) process. The default of a loan happens if the wealth stops to be su cient for repaying the loan. We show that the mapping transforming the common factors into the probability of default (PD) and the loss given default (LGD) is one-to-one twice continuously differentiable. As the transformation is not analytically tractable, we propose a numerical technique for its computation and demonstrate its accuracy by a numerical study.\nWe show that the results given by our multi-period model may differ signi cantly from\nthose resulting from single-period models, and demonstrate that our model naturally replicates\nthe empirically observed decrease of PDs within a portfolio in time. In addition, we give a formula for the overall loss of the portfolio and, as an example of its application, we formulate a simple optimal scoring decision problem and discuss its solution. Keywords: Credit Risk; Structural Factor Models; Loan Portfolio Management Fulltext is available at external website. Multi-period Factor Model of a Loan Portfolio

We construct a general dynamic model of losses of a large loan portfolio, secured by collaterals. In the model, the wealth of a debtor and the price of the corresponding collateral depend each on two ...

Šmíd, Martin; Dufek, J.
Ústav teorie informace a automatizace, 2017

Diffusion MCMC for Mixture Estimation
Reichl, Jan; Dedecius, Kamil
2016 - English Distributed inference of parameters of mixture models by a network of cooperating nodes (sensors) with computational and communication capabilities still represents a challenging task. In the last decade, several methods were proposed to solve this issue, predominantly formulated within the expectation-maximization framework and with the assumption of mixture components normality. The present paper adopts the Bayesian approach to inference of general (non-normal) mixtures via the Markov chain Monte Carlo simulation from the parameter posterior distribution. By collaborative tuning of node chains, the method allows reliable estimation even at nodes with significantly worse observational conditions, where the components may tend to merge due to high variances. The method runs in the diffusion networks, where the nodes communicate only with their adjacent neighbors within 1 hop distance. Keywords: Mixture; mixture estimation; MCMC Fulltext is available at external website. Diffusion MCMC for Mixture Estimation

Distributed inference of parameters of mixture models by a network of cooperating nodes (sensors) with computational and communication capabilities still represents a challenging task. In the last ...

Reichl, Jan; Dedecius, Kamil
Ústav teorie informace a automatizace, 2016

Adaptive Blind Separation of Instantaneous Linear Mixtures of Independent Sources
Šembera, Ondřej; Tichavský, Petr; Koldovský, Zbyněk
2016 - English In many applications, there is a need to blindly separate independent sources from their linear instantaneous mixtures while the mixing matrix or source properties are slowly or abruptly changing in time. The easiest way to separate the data is to consider off-line estimation of the model parameters repeatedly in time shifting window. Another popular method is the stochastic natural gradient algorithm, which relies on non-Gaussianity of the separated signals and is adaptive by its nature. In this paper, we propose an adaptive version of two blind source separation algorithms which exploit non-stationarity of the original signals. The results indicate that the proposed algorithms slightly outperform the natural gradient in the trade-off between the algorithm’s ability to quickly adapt to changes in the mixing matrix and the variance of the estimate when the mixing is stationary. Keywords: blind separation; algorithms; block gaussian separation Fulltext is available at external website. Adaptive Blind Separation of Instantaneous Linear Mixtures of Independent Sources

In many applications, there is a need to blindly separate independent sources from their linear instantaneous mixtures while the mixing matrix or source properties are slowly or abruptly changing in ...

Šembera, Ondřej; Tichavský, Petr; Koldovský, Zbyněk
Ústav teorie informace a automatizace, 2016

Basic facts concerning extreme supermodular functions
Studený, Milan
2016 - English Elementary facts and observations on the cone of supermodular set functions are recalled. The manuscript deals with such operations with set functions which preserve supermodularity\nand the emphasis is put on those such operations which even preserve extremality (of a supermodular function). These involve a few self-transformations of the cone of supermodular set functions. Moreover, projections to the (less-dimensional) linear space of set functions for a subset of the variable set are discussed. Finally, several extensions to the (more-dimensional) linear space of set functions for a superset of the variable set are shown to be both preserving supermodularity and extremality. Keywords: supermodular function; standardizations; extreme supermodular function Fulltext is available at external website. Basic facts concerning extreme supermodular functions

Elementary facts and observations on the cone of supermodular set functions are recalled. The manuscript deals with such operations with set functions which preserve supermodularity\nand the emphasis ...

Studený, Milan
Ústav teorie informace a automatizace, 2016

Comparison of mixture-based classification with the data-dependent pointer model for various types of components
Likhonina, Raissa; Suzdaleva, Evgenia; Nagy, Ivan
2016 - English The presented report is devoted to the analysis of a data-dependent pointer model, whether it brings some advantages in comparison with a data-independent pointer model at simulation and estimation of components referring to different types of distribution, including categorical, uniform, exponential and state-space components for a dynamic data-dependent model, and normal components for a static data-dependent pointer model. Keywords: mixture-based classification; data-dependent pointer; recurisive mixture estimation Fulltext is available at external website. Comparison of mixture-based classification with the data-dependent pointer model for various types of components

The presented report is devoted to the analysis of a data-dependent pointer model, whether it brings some advantages in comparison with a data-independent pointer model at simulation and estimation of ...

Likhonina, Raissa; Suzdaleva, Evgenia; Nagy, Ivan
Ústav teorie informace a automatizace, 2016

Linear ARX and state-space model with uniform noise: computation of first and second moments
Jirsa, Ladislav
2016 - English This report collects technical procedures used for computations of various estimates and keeps them in one place for internal purposes. The context concerns application of estimation of unknown parameters and states of linear model with uniformly distributed noise. Keywords: uncertainty; bounded variable; uniform noise; linear model; model identification; state estimation Fulltext is available at external website. Linear ARX and state-space model with uniform noise: computation of first and second moments

This report collects technical procedures used for computations of various estimates and keeps them in one place for internal purposes. The context concerns application of estimation of unknown ...

Jirsa, Ladislav
Ústav teorie informace a automatizace, 2016

Sparse robust portfolio optimization via NLP regularizations
Branda, Martin; Červinka, Michal; Schwartz, A.
2016 - English We deal with investment problems where we minimize a risk measure under a condition on the sparsity of the portfolio. Various risk measures are considered including Value-at-Risk and Conditional Value-at-Risk under normal distribution of returns and their robust counterparts are derived under moment conditions, all leading to nonconvex objective functions. We propose four solution approaches: a mixed-integer formulation, a relaxation of an alternative mixed-integer reformulation and two NLP regularizations. In a numerical study, we compare their computational performance on a large number of simulated instances taken from the literature. We deal with investment problems where we minimize a risk measure\nunder a condition on the sparsity of the portfolio. Various risk measures\nare considered including Value-at-Risk and Conditional Value-at-Risk\nunder normal distribution of returns and their robust counterparts are\nderived under moment conditions, all leading to nonconvex objective\nfunctions. We propose four solution approaches: a mixed-integer formulation,\na relaxation of an alternative mixed-integer reformulation and\ntwo NLP regularizations. In a numerical study, we compare their computational\nperformance on a large number of simulated instances taken\nfrom the literature. Keywords: Conditional Value-at-Risk; Value-at-Risk; risk measure Fulltext is available at external website. Sparse robust portfolio optimization via NLP regularizations

We deal with investment problems where we minimize a risk measure under a condition on the sparsity of the portfolio. Various risk measures are considered including Value-at-Risk and Conditional ...

Branda, Martin; Červinka, Michal; Schwartz, A.
Ústav teorie informace a automatizace, 2016