Doctoral Schools WUT

Search Engine for Promoters and Research Areas

Wykaz obszarów badawczych związanych z tagiem Sieci-neuronowe:

# Obszar badawczy Dziedzina naukowa
1 Application of artificial intelligence methods in technical diagnostics, including the issue of selection of diagnostic information and diagnostically oriented signal processing and analysis. Modeling of electric energy storages for electric and hybrid vehicles with the use of computational intelligence methods. Autonomization of vehicles and their functioning in intelligent transport systems.

The growing importance of projects in modern enterprises is accompanied by an increasing need for methods and tools supporting decisions. Computational technologies related to artificial intelligence methods, such as neural networks, fuzzy logic, and hybrid systems, are noteworthy.

3 Time series classification is a popular machine learning task with many practical applications. To date, there have been many papers devoted to the classification of binary and multiclass time series of one and many variables. Thanks to the resources collected on the UEA/UCR Time Series Classification Repository website (, it is possible to get a preliminary overview of the diversity of methods in this area and of the datasets considered as the so-called benchmark sets of the field. The proposed research area concerns the development of new approaches to time series classification, in particular methods based on neural networks. A related topic is the analysis of time series representation schemes and feature extraction techniques that will be the base for building a classifier. A second related topic is early classification.
4 New video compression methods based on neural networks: The research problem is devoted to the development of new architectures of convolutional neural networks and other video signal processing components for effective compression of video sequences. These other elements include the use of orthogonal transformations of the DCT and DWT type, descriptors allowing for the formation of the context. Basically, the network architectures for compression are based on the autoencoder structure. However, the relationship between the layers and the structure of the layers itself depends on many hyper parameters. The goal is to find a solution that will give the best possible compression efficiency at a limited computational cost. It is advisable to use learning in the GAN configuration to adjust the compression to the human vision system.
5 Theory and algorithms of intelligent hypercomplex signal processing: The research is concentrated on the theory and search of practical applications of theoretical results in multidimensional signal processing. Nowadays, the methods of intelligent signal processing are extensively developping as the alternative of conventional methods. Such an approach facilitates elaboration of effective algorithms of multidimensional signal processing using deep learning methods and neural networks. In the first stage of the research, the study of advanced conventional methods of multidimensional signal processing with emphasis on hypercomplex algebras of quaternions and octonions is required. Then, the deep study of the state-of-the art in the domain of intelligent methods of multidimensional signal processing is needed. During the research, new methods will be alaboarted and tested on real 3-D signals.