20210328 00:00 
DEnFi: Deep Ensemble Filter for Active Learning
Ulrych, Lukáš ; Šmídl, Václav
Deep Ensembles proved to be a one of the most accurate representation of uncertainty for deep neural networks. Their accuracy is beneficial in the task of active learning where unknown samples are selected for labeling based on the uncertainty of their prediction. Underestimation of the predictive uncertainty leads to poor exploration of the method. The main issue of deep ensembles is their computational cost since multiple complex networks have to be computed in parallel. In this paper, we propose to address this issue by taking advantage of the recursive nature of active learning. Specifically, we propose several methods how to generate initial values of an ensemble based of the previous ensemble. We provide comparison of the proposed strategies with existing methods on benchmark problems from Bayesian optimization and active classification. Practical benefits of the approach is demonstrated on example of learning ID of an IoT device from structured data using deepset based networks.
Úplný záznam

20210328 00:00 
ECB monetary policy and commodity prices
Aliyev, S. ; Kočenda, Evžen
We analyze the impact of the ECB monetary policies on global aggregate and sectoral commodity prices using monthly data from January 2001 till August 2019. We employ a SVAR model and assess separately period of conventional monetary policy before global financial crisis (GFC) and unconventional monetary policy during postcrisis period. Our key results indicate that contractionary monetary policy shocks have positive effects on the aggregate and sectoral commodity prices during both conventional and unconvetional monetary policy periods. The effect is statistically significant for aggregate commodity prices during postcrisis period. In terms of sectoral impact, the effect is statistically significant for food prices in both periods and for fuel prices during postcrisis period; other commodities display positive but statistically insignificant responses. Further, we demonstrate that the impact of the ECB monetary policy on commodity prices increased remarkably after the GFC. Our results also suggest that the effect of the ECB monetary policy on commodity prices does not transmit directly through market demand and supply expectations channel, but rather through the exchange rate channel that influences the European market demand directly.
Úplný záznam

20210328 00:00 
Potential Radioactive Hot Spots Induced by Radiation Accident Being Underway of Atypical Low Wind Meteorological Episodes
Pecha, Petr ; Tichý, Ondřej ; Pechová, E.
Hypothetical radioactivity release with potentially high variability of the source strength is examined. The interactions of the radioactive cloud with surface and atmospheric precipitation are studied and possible adverse consequences on the environment are estimated. The worstcase scenario is devised in two stages starting with a calm meteorological situation succeeded by wind. At the first stage, the discharges of radionuclides into the motionless ambient atmosphere are assumed. During several hours of this calm meteorological situation, a relatively significant level of radioactivity can be accumulated around the source. At the second stage, the calm is assumed to terminate and convective movement of the air immediately starts. The pack of accumulated radioactivity in the form of multiple Gaussian puffs is drifted by wind and pollution is disseminated over the terrain. The results demonstrate the significant transport of radioactivity even behind the protective zone of a nuclear facility (up to between 15 and 20 km). In the case of rain, the aerosols are heavily washed out and dangerous hot spots of the deposited radioactivity can surprisingly emerge even far from the original source of the pollution.
Úplný záznam

20210224 00:44 
Úplný záznam

20210224 00:44 
Úplný záznam

20210224 00:43 
Úplný záznam

20210224 00:43 
Bayesian Selective Transfer Learning for PatientSpecific Inference in Thyroid Radiotherapy
Murray, Sean Ernest ; Quinn, Anthony
This research report outlines a selective transfer approach for Bayesian estimation of patientspecific levels of radioiodine activity in the thyroid during the treatment of differentiated thyroid carcinoma. The work seeks to address some limitations of previous approaches [4] which involve generic, nonselective transfer of archival data. It is proposed that improvements in patientspecific inferences may be achieved via transferring external population knowledge selectively. This involves matching the patient to a similar subpopulation based on available metadata, generating a Gaussian Mixture Model within the partitioned data, and optimally transferring a data predictive distribution from the subpopulation to the specific patient. Additionally, a performance evaluation method is proposed and earlystage results presented.
Úplný záznam

20210224 00:43 
RiskSensitivity and Average Optimality in Markov and SemiMarkov Reward Processes
Sladký, Karel
This contribution is devoted to risksensitivity in longrun average optimality of Markov and semiMarkov reward processes. Since the traditional average optimality criteria cannot reflect the variabilityrisk features of the problem, we are interested in more sophisticated approaches where the stream of rewards generated by the Markov chain that is evaluated by an exponential utility function with a given risk sensitivity coefficient. Recall that for the risk sensitivity coefficient equal to zero (i.e. the so called riskneutral case) we arrive at traditional optimality criteria, if the risk sensitivity coefficient is close to zero the Taylor expansion enables to evaluate variability of the generated total reward. Observe that the first moment of the total reward corresponds to expectation of total reward and the second central moment to the reward variance. In this note we present necessary and sufficient risksensitivity and riskneutral optimality conditions for long run risksensitive average optimality criterion of unichain Markov and semiMarkov reward processes.
Úplný záznam

20210224 00:43 
A Note on Stochastic Optimization Problems with Nonlinear Dependence on a Probability Measure
Kaňková, Vlasta
Nonlinear dependence on a probability measure begins to appear (last time) in a stochastic optimization rather often. Namely, the corresponding type of problems corresponds to many situations in applications. The nonlinear dependence can appear as in the objective functions so in a constraints set. We plan to consider the case of static (oneobjective) problems in which nonlinear dependence appears in the objective function with a few types of constraints sets. In details we consider constraints sets “deterministic”, depending nonlinearly on the probability measure, constraints set determined by second order stochastic dominance and the sets given by meanrisk problems. The last case means that the constraints set corresponds to solutions those guarantee an acceptable value in both criteria. To introduce corresponding assertions we employ the stability results based on the Wasserstein metric and L1 norm. Moreover, we try to deal also with the case when all results have to be obtained (estimated) on the data base.
Úplný záznam

20210224 00:43 
Use of the BCC and Range Directional DEA Models within an Efficiency Evaluation
Houda, Michal
The contribution deals with two data envelopment analysis (DEA) models, in particular the BCC model (radial DEA model with variable returns to scale), and the range directional model. The mathematical description of the models are provided and several properties reported. A numerical comparison of the two models on real industrial data is provided with discussion about possible drawbacks of simplifying modeling procedures.
Úplný záznam



