Search

In this paper, we propose an evolutionary learning approach for flexible identification of custom composite models for classification problems. To solve this problem in an efficient way, the problem-specific evolutionary operators are proposed and the effectiveness of different modifications of the common genetic programming algorithm is investigated. Also, several implementations of caching for the fitted models were compared from the performance point of view. To verify the proposed algorithm, both synthetic and real-world classification cases are examined. The implemented solution can identify the structure of the composite models from scratch, as well as be used as a part of automated machine learning solutions. more...

The effectiveness of the machine learning methods for real-world tasks depends on the proper structure of the modeling pipeline. The proposed approach is aimed to automate the design of composite machine learning pipelines, which is equivalent to computation workflows that consist of models and data operations. The approach combines key ideas of both automated machine learning and workflow management systems. It designs the pipelines with a customizable graph-based structure, analyzes the obtained results, and reproduces them. The evolutionary approach is used for the flexible identification of pipeline structure. The additional algorithms for sensitivity analysis, atomization, and hyperparameter tuning are implemented to improve the effectiveness of the approach. Also, the software implementation on this approach is presented as an open-source framework. The set of experiments is conducted for the different datasets and tasks (classification, regression, time series forecasting). The obtained results confirm the correctness and effectiveness of the proposed approach in the comparison with the state-of-the-art competitors and baseline solutions. more...

In this paper, we describe the concept of generative design approach applied to the automated evolutionary learning of mathematical models in a computationally efficient way. To formalize the problems of models&rsquo, design and co-design, the generalized formulation of the modeling workflow is proposed. A parallelized evolutionary learning approach for the identification of model structure is described for the equation-based model and composite machine learning models. Moreover, the involvement of the performance models in the design process is analyzed. A set of experiments with various models and computational resources is conducted to verify different aspects of the proposed approach. more...

The problem of the weather forecasting still exists in urban agglomerations, in part because of local factors on a city scale. It is important to take into acount urban processes, to make a high-resolution weather forecast. In this study, we suggest an approach to use crowdsourced meteorological data from citizen weather stations (CWS) to improve weather forecasting. The Weather Research and Forecasting model was used to simulate forecasting for two cities Saint Petersburg and Moscow. Using processed CWS data we propose a data-driven model to correct numeric modeling for short-time forecasting. We propose a procedure for statistical validation of citizen stations using observations from professional meteostations. Experiments have shown that in quality they can be compared with the government ones, the numerical model WRF need to be corrected for representing winter heat island. We considered the possibility of using text data from social media posts to predict weather conditions in the case of air temperature. more...

Metocean modeling with short output fields timestep, for example, hourly average fields output, generates a large number of pictures and requires extended computational power. Often, during the simulation different types of artifacts can appear due to the inappropriate model tuning or errors in boundary and initial data and, therefore, expert’s supervision and validation are required. When the number of images is increasing it becomes difficult or even impossible to check all output images manually. Therefore, it is required to use machine learning algorithms to reduce a time for expert’s validation. Thereby, it would be useful to develop a system that allows detecting anomalies in generated data automatically during the experiment. In the paper, we provide a method of anomalies detection for the geospatial data. Data in climatographic archives is available in restricted amount and therefore, full Arctic images are divided into sub-zones, which allows one to increase training set. Moreover, this division can be used to account for the spatial dependency, which is required for ice images. An advantage of the approach is the ability to detect anomalies completely in automatic mode without involving a domain expert and manual labeling. more...

Exploring customer interests from open source information has become a significant issue. On the one hand, consumers deepen their engagement with the brands which values matter to them. On the other hand, annoying marketing calls and polls do not reflect real customers’ needs and wants. This article considers topic modeling in application to social media analysis. We have received interpretable topics related to users preferences. Crawled posts texts and texts obtaining from images by an optical character recognition were used as datasets. Focusing on two approaches: probabilistic (LDA, ARTM) and neural network based (doc2vec, word2vec), we suggest the combined model deARTM. Hierarchical ARTM model allows us to obtain relations between texts in several abstraction levels which we used as vector representation. To avoid misspelling sensitivity, our model includes document embedding. In the experimental part, we show that our model can improve results of topic modeling on social media datasets. more...

Summary In this paper, the authors investigated the approach of Bayesian networks to the analysis of parameters of oil and gas fields. The study of existing approaches and methods of probabilistic modeling in relation to the problems of analyzing the parameters of oil and gas fields based on production data showed that the best approach should have a number of important properties: the interpretability of the model, the ability to work with various types of data, and the ability to process distributions of sufficient dimension in a reasonable time. Bayesian networks were chosen as the main tool of work, since they allow to develop models that are understandable to the specialist and allow to do this entirely on data with minimal involvement of expert knowledge. The experiments have shown that Bayesian networks are able to simulate multidimensional distributions of field parameters; the developed model can also be used to reconstruct data gaps and assess the significance of variables. A comparison was made of different architectures of Bayesian networks with different score functions. It also shows an example of assessing the information significance of the parameters of deposits. more...