Papers by Urszula Markowska-Kaczmar
Pozyskiwanie symbolicznej wiedzy z sieci neuronowych
Artificial intelligence review, Feb 14, 2024
This article proposes a novel concept for a two-step Ki-67/lymphocytes classification cell detect... more This article proposes a novel concept for a two-step Ki-67/lymphocytes classification cell detection pipeline on Ki-67 stained histopathological slides utilizing commonly available and undedicated, in terms of the medical problem considered, deep learning models. Models used vary in implementation, complexity, and applications, allowing for the use of a dedicated architecture depending on the physician's needs. Moreover, generic models' performance was compared with the problem-dedicated one. Experiments highlight that with relatively small training datasets, commonly used architectures for instance segmentation and object detection are competitive with a dedicated model. To ensure generalization and minimize biased sampling, experiments were performed on data derived from two unrelated histopathology laboratories.
Pozyskiwanie symbolicznej wiedzy z sieci neuronowych
Prace Naukowe Akademii Ekonomicznej we Wrocławiu, 2000
Interpretacja symboliczna wiedzy z sieci neuronowej
An Overview of Few-Shot Learning Methods in Analysis of Histopathological Images
Intelligent systems reference library, 2023
Frontiers in Neuroscience, 2019
An author name was incorrectly spelled as "Urszula Markowska-Kacznar." The correct spelling is "U... more An author name was incorrectly spelled as "Urszula Markowska-Kacznar." The correct spelling is "Urszula Markowska-Kaczmar." The authors apologize for this error and state that this does not change the scientific conclusions of the article in any way. The original article has been updated.
American Sign Language Fingerspelling Recognition Using Wide Residual Networks
Artificial Intelligence and Soft Computing, 2018
Despite existing solutions for accurate translation between written and spoken language, sign lan... more Despite existing solutions for accurate translation between written and spoken language, sign language is still not well-studied area. A reliable, robust and working in real-time translator of American Sign Language is a crucial bridge to facilitate communication between deaf and hearing people. In this paper we propose a method of sign language fingerspelling recognition using a modern architecture of convolutional neural network called Wide Residual Network trained with Snapshot Learning procedure. The model was trained on augmented datasets available at Surrey University and Massey University web pages using transfer learning. The final result is a robust classifier of all alphabet letters, which beats current state-of-the-art results. The outcomes encourage further research in this field for creating fully usable sign language translator.
3D robotic navigation using a vision-based deep reinforcement learning model
Applied Soft Computing, 2021
Abstract In this paper, we address a problem of vision-based 3D robotic navigation using deep rei... more Abstract In this paper, we address a problem of vision-based 3D robotic navigation using deep reinforcement learning for an Autonomous Underwater Vehicle (AUV). Our research offers conclusions from the experimental study based on one of the RoboSub 2018 competition tasks. However, it can be generalized to any navigation task consisting of movement from a starting point to the front of the next station. The presented reinforcement learning-based model predicts the robot’s steering settings using the data acquired from the robot’s sensors. Its Vision Module may be based on a built-in convolutional network or a pre-trained TinyYOLO network so that a comparison of various levels of features’ complexity is possible. To enable evaluation of the proposed solution, we prepared a test environment imitating the real conditions. It provides the ability to steer the agent simulating the AUV and calculate values of rewards, used for training the model by evaluating its decisions. We study the solution in terms of the reward function form, the model’s hyperparameters and the exploited camera images processing method, and provide an analysis of the correctness and speed of the model’s functioning. As a result, we obtain a valid model able to steer the robot from the starting point to the destination based on visual cues and inputs from other sensors.
Capsule Network Versus Convolutional Neural Network in Image Classification
Computational Science – ICCS 2021, 2021
2009 International Multiconference on Computer Science and Information Technology, 2009
This paper presents the method of cancer localization in the breast tissue digital images. The me... more This paper presents the method of cancer localization in the breast tissue digital images. The method is implemented and tested in order to be included in the image analysis system which aim is to support a surgeon in interoperative probe of pathological areas in a breast tissue. In future it will be supplemented with information about cancerous areas acquired from dielectric maps. The idea of the whole system is described in the paper. Next, the method of cancer localization in the breast tissue digital images is presented. This method enables a detection of the cancerous changes on the basis of Otsu's binarization method and the saturation information from a breast tissue image. It is worth mentioning that after small changes this method can be used to segmentate images due to more complex criteria.
Granular Representation of Temporal Signals Using Differential Quadratures
Lecture Notes in Computer Science, 2011
This article presents the general idea of granular representation of temporal data, particularly ... more This article presents the general idea of granular representation of temporal data, particularly signal sampled with constant frequency. The core of presented method is based on using fuzzy numbers as information granules. Three types of fuzzy numbers are considered, as interval numbers, triangular numbers and Gaussian numbers. The input space contains values of first few derivatives of underlying signal, which are computed using certain numerical differentiation algorithms, including polynomial interpolation as well as polynomial approximation. Data granules are constructed using the optimization method according to objective function based on two criteria: high description ability and compactness of fuzzy numbers. The data granules are subject to the clustering process, namely fuzzy c-means. The centroids of created clusters form a granular vocabulary. Quality of description is quantitatively assessed by reconstruction criterion. Results of numerical experiments are presented, which incorporate exemplary biomedical signal, namely electrocardiographic signal.
Fast learning method of interval type-2 fuzzy neural networks
2014 14th UK Workshop on Computational Intelligence (UKCI), 2014
2009 International Multiconference on Computer Science and Information Technology, 2009
In the paper the method of creating investment strategies for a profitable trading system is desc... more In the paper the method of creating investment strategies for a profitable trading system is described. This method is based on artificial intelligence techniques and technical analysis tools. Created strategies describe the investment signal and amount of cash or stocks, which should be used at a given moment. The carried out experiments allow to find values of parameters for generating investment strategies and to define an influence of provisions. An important element of this work is to check the possibility of investment strategies generalization. The possibility of generalization determines a new direction of the research on the model. On this basis, in future the stock which is currently worth to invest or this one which is better to sell could be identified.
Support vector machines in handwritten digits classification
5th International Conference on Intelligent Systems Design and Applications (ISDA'05), 2005
Abstract In the paper our approach to classify handwritten digits by using support vector machine... more Abstract In the paper our approach to classify handwritten digits by using support vector machines is described. Because of the unsatisfying, long time of training of SVM we propose to apply k-nearest neighbours algorithm with Manhattan distance to obtain reduced size of ...
The influence of parameters in evolutionary based rule extraction method from neural network
5th International Conference on Intelligent Systems Design and Applications (ISDA'05), 2005
Abstract In the paper the experimental study of the influence of parameters on the final results ... more Abstract In the paper the experimental study of the influence of parameters on the final results of the rule extraction method from neural network for classification problem is described. The method is based on evolutionary approach, where for each class evolves ...
IISA 2014, The 5th International Conference on Information, Intelligence, Systems and Applications, 2014
Ahstract-The automatic method of information extraction from heatmaps based on OCR, image process... more Ahstract-The automatic method of information extraction from heatmaps based on OCR, image processing and image recognition techniques is proposed. It is composed of the sequence of steps. First, the heatmap area is separated from other elements of the heatmap image. Next, the key and axis are recognized. To produce quick answers for a user query, the heatmap is stored in the form of a tree. The method was tested on the basis of several diverse heatmaps. The results are promising.
Creativity of Neural Networks
Lecture Notes in Computer Science, 2006
In the paper the ability of neural networks in creativity is tested. The creation of new words wa... more In the paper the ability of neural networks in creativity is tested. The creation of new words was chosen as an example task of creativity. Three different approaches based on the neural networks were designed and implemented to perform experiments. From all ...
GA-Based Pareto Optimization for Rule Extraction from Neural Networks
Studies in Computational Intelligence, 2006
The chapter presents a new method of rule extraction from trained neural networks, based on a hie... more The chapter presents a new method of rule extraction from trained neural networks, based on a hierarchical multiobjective genetic algorithm. The problems associated with rule extraction, especially its multiobjective nature, are described in detail, and techniques ...
Extraction of Emotional Content from Music Data
2008 7th Computer Information Systems and Industrial Management Applications, 2008
Abstract This paper presents the system for automatic emotion detection from music data stored in... more Abstract This paper presents the system for automatic emotion detection from music data stored in MIDI format files. First, the piece of music is divided into independent segments that potentially represent different emotional states. For this task the method of segmentation is used. The most important part is a features extraction from the music data. On this basis similar emotional parts are grouped by clustering algorithm. Music domain knowledge is used to extract features which are then grouped hierarchically by agglomerative clustering ...
Data Mining Techniques in e-Learning CelGrid System
2008 7th Computer Information Systems and Industrial Management Applications, 2008
Abstract The paper presents e-learning an system as a source of large datasets that can be analyz... more Abstract The paper presents e-learning an system as a source of large datasets that can be analyzed by data mining techniques. Proposed data mining techniques can be used as a didactic content recommendation system, feedback tool, intrusion detection tools etc. All techniques are applied to make learning process more effective (taking into account time consuming aspects and resource usage). The paper describes data mining tasks and techniques that can be applied to CelGrid system. A particular attention is given to the ...
Uploads
Papers by Urszula Markowska-Kaczmar