Identifying a proper model structure, using methods that address both structural and parameter uncertainty, is a crucial problem within the systems approach to biology. And yet, it has a marginal presence in the recent literature. While many existing approaches integrate methods for simulation and parameter estimation of a single model to address parameter uncertainty, only few of them address structural uncertainty at the same time. The methods for handling structure uncertainty often oversimplify the problem by allowing the human modeler to explicitly enumerate a relatively small number of alternative model structures. On the other hand, process-based modeling methods provide flexible modular formalisms for specifying large classes of plausible model structures, but their scope is limited to deterministic models. Within the project, we have extended the scope of process-based modeling methods to inductively learn stochastic models from knowledge and data. In the epidemiology domain, the method successfully reconstructs previously established models of epidemic outbreaks from real, sparse and noisy measurement data.
F.02 Acquisition of new scientific knowledge
COBISS.SI-ID: 29374503In a medical context, we have introduced a novel methodology for noninvasive assessment of structure and composition of human skin in vivo. The approach combines pulsed photothermal radiometry (PPTR), involving time-resolved measurements of mid- infrared emission after irradiation with a millisecond light pulse, and diffuse reflectance spectroscopy (DRS) in visible part of the spectrum (400–600 nm). The experimental data are fitted simultaneously with respective predictions from a four-layer Monte Carlo (MC) model of light transport in human skin. The described approach allows assessment of the contents of specific chromophores (melanin, oxy-, and deoxy- hemoglobin), as well as scattering properties and thicknesses of the epidermis and dermis. However, the involved multi- dimensional optimization with a numerical forward model (i.e., inverse MC, IMC) is computationally very expensive. In addition, each optimization task is repeated several times to control the inevitable numerical noise and facilitate escape from local minima. Thus, assessment of 14 free parameters from each radiometric transient and DRS spectrum takes several hours despite massive parallelization using CUDA technology and a high-performance graphics card. To alleviate this limitation, we have developed a computationally very efficient predictive model (PM) based on machine learning technology. The PM is an ensemble of decision trees (random forest), trained using ~10,000 "pairs" of various skin parameter combinations and the corresponding PPTR signals and DRS spectra, computed using our forward MC model. While the parameter values predicted by the PM are very similar to the IMC results there are some concerns regarding their accuracy. Therefore, we present here a hybrid model, which combines the described PM and IMC approaches.
F.21 Development of new health/diagnostic methods/procedures
COBISS.SI-ID: 32537639For modeling the population dynamics of aquatic ecosystems, we have used the approach of process-based modeling. We have built both individual process-based models and ensembles thereof [COBISS-ID 29437223]. We modeled the growth of algae in Lake Bled (Slovenia) and Lake Kasumigaura (Japan). We have also used evolutionary algorithms and regression trees for inferential modelling of complex ecological data. Evolutionary algorithms prove to be superior tools for developing short-term forecasting models, revealing ecological thresholds and supporting quantitative meta-analyses as demonstrated exemplarily by means of the hybrid evolutionary algorithm (HEA). A case study of Lake Müggelsee (Germany) illustrates that models developed by HEA enable one to identify ecological thresholds and driving forces that perform short-term forecasting of population growth. The meta-analysis of Lakes Wivenhoe (Australia) and Lake Paranoa (Brazil) exemplifies the capability of models developed by HEA to test hypotheses on forcing functions of population growth across different environmental and climate conditions. Regression trees display fully transparent correlations between habitat properties and ecological entities. The tree induction process does not require prior assumptions, is fast and is not influenced by redundant variables and noise. A case study for Lake Prespa (Macedonia) illustrates the capacity of regression trees to unravel complex ecological relationships.
F.27 Contribution to preserving/protecting natural and cultural heritage
COBISS.SI-ID: 30863911In 2017, we organized an international conference on machine learning ECML PKDD 2017. The conference was held from September 18-22, 2017 in Skopje, Macedonia. This is a prestigious annual event, attended by 600 participants from around the world. There were 6 invited talks and around 100 paper and poster presentations. In addition, there were a number of accompanying events such as workshops, tutorials, an EU Projects forum and a PhD forum. More information on the conference can be found at the website http://ecmlpkdd2017.ijs.si/
B.01 Organiser of a scientific meeting
COBISS.SI-ID: 31153447ProBMoT is a software tool implementating the process-based modeling approach to modeling dynamical systems. At the core of the approach is a formalism for representing models of dynamical systems as well as knowledge for modeling dynamical systems in a particular domain of interest. The models and knowledge are formalized at two abstraction levels. At the higher level, the representation is based on processes that govern the dynamics and entities involved in the processes and allows for reasoning about the causal and structural relationships in the observed system. At the lower level, processes include specific equations and entities include system variables, allowing for simulation of the observed system dynamics. ProBMoT can automatically transform model and knowledge representations between the two levels and provide computational support for the tasks of simulating process-based models, estimating parameters of process models from data, and learning both the structure and parameters of the models from data and knowledge. More information on ProBMoT can be found at the website http://probmot.ijs.si/ ProBMoT has been used to learn epidemiological models of the spread of COVID-19 in Slovenia (http://kt.ijs.si/~ljupco/covid-19-sir/report.nb.html).
F.15 Development of a new information system/databases
COBISS.SI-ID: 31157543