A5030600070- Tensor decomposition and applications
- Natural Language Processing Techniques
- Machine Fault Diagnosis Techniques
- Advanced Neural Network Applications
- Algorithms and Data Compression
- Machine Learning and Algorithms
- Computational Physics and Python Applications
- Ultrasonics and Acoustic Wave Propagation
- Speech Recognition and Synthesis
- Genomics and Phylogenetic Studies
- RNA and protein synthesis mechanisms
- Elasticity and Wave Propagation
- Parallel Computing and Optimization Techniques
- Antenna Design and Optimization
- Structural Integrity and Reliability Analysis
- Vehicle License Plate Recognition
- Gear and Bearing Dynamics Analysis
- Model Reduction and Neural Networks
- Machine Learning in Bioinformatics
- Advanced Adaptive Filtering Techniques
- Neural Networks and Applications
- Structural Health Monitoring Techniques
- Digital Filter Design and Implementation
- Mechanical Engineering and Vibrations Research
- Text and Document Classification Technologies
Wrocław University of Science and Technology
2019-2023
AGH University of Krakow
2019-2023
Local damage of bearings can be detected as a weak cyclic and impulsive component in highly noisy measured signal. A key problem is how to extract the signal interest (SOI) from raw signal, i.e., identify design an optimal filter. To tackle this problem, we propose use stochastic sampled orthogonal non-negative matrix factorization for extracting frequency-based features spectrogram The proposed algorithm finds selective filter that tailored frequency band SOI. We show our approach...
The split-based method in a weighted context-free grammar (WCFG) induction was formalised and verified on comprehensive set of languages. WCFG is learned using novel grammatical inference method. proposed learns from both positive negative samples, whereas the weights rules are estimated Inside–Outside Contrastive Estimation algorithm. results showed that our approach outperforms terms F1 scores other state-of-the-art methods.
Convolutional neural networks (CNNs) are among the most widely used machine learning models for computer vision tasks, such as image classification. To improve efficiency of CNNs, many CNNs compressing approaches have been developed. Low-rank methods approximate original convolutional kernel with a sequence smaller kernels, which leads to reduced storage and time complexities. In this study, we propose novel low-rank compression method that is based on direct tensor ring decomposition...
The topic of convolutional neural networks (CNN) compression has attracted increasing attention as new generations become larger and require more computing performance. This computational problem can be solved by representing the weights a network with low-rank factors using matrix/tensor decomposition methods. study presents novel concept for compressing nested In this approach, we alternately perform fine-tuning network. numerical experiments are performed on various CNN architectures,...
Convolutional neural networks (CNN) play a crucial role and achieve top results in computer vision tasks but at the cost of high computational performance storage complexity. One way to solve this problem is approximation convolution kernel using tensor decomposition methods. In way, original replaced with sequence kernels lower-dimensional space. This study proposes novel CNN compression technique based on hierarchical Tucker-2 (HT-2) makes an important contribution field network low-rank...
Time-frequency representation (TFR) is often used for non-stationary signal analysis. The most intuitive and interpretable TFR the spectrogram. Recently, a concept of non-negative matrix factorization (NMF) has been successfully applied to local damage detection in rolling elements bearings via spectrogram factorization. NMF allows one find an informative frequency band, which could be further as filter characteristic. However, obtained characteristics mostly detect also encompasses lot...