Sitemap
A list of all the posts and pages found on the site. For you robots out there, there is an XML version available for digesting as well.
Pages
Posts
Future Blog Post
Published:
This post will show up by default. To disable scheduling of future posts, edit config.yml and set future: false.
Blog Post number 4
Published:
This is a sample blog post. Lorem ipsum I can’t remember the rest of lorem ipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.
Blog Post number 3
Published:
This is a sample blog post. Lorem ipsum I can’t remember the rest of lorem ipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.
Blog Post number 2
Published:
This is a sample blog post. Lorem ipsum I can’t remember the rest of lorem ipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.
Blog Post number 1
Published:
This is a sample blog post. Lorem ipsum I can’t remember the rest of lorem ipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.
my-students
Mood Classification in Music Tracks
Defense Date 2018
The thesis is about realization and comparison of multiclass classification methods. Important part of the work was to create an application for gaining and presenting music tracks’ data. The data is used as training and test set for the classifiers. Moreover, it describes classification theory and describes chosen models: naive Bayes classifier, k-nearest neighbors method, decision tree, support vector machine and random forest. The work also presents tools used during the thesis development and guides through data preparation process. It also describes tests that were made and their results.
Use of hash functions in solving machine learning tasks
Defense Date 2018
This thesis presents following subjects: usage of hash functions while preparation of data for machine learning tasks and the tries of optimization of selected methods. At the beginning, hashing functions are introduced. Their definition and classification is given with samples. The functions that were used in this thesis like MurmurHash3 are presented with more details. The principals of machine learning algorithms that are used in the thesis are also explained. In the further part, the experiment on classification accuracy in dependence of the reduced number of dimensions and number of collisions, which happened because of same hash function values for different inputs, is described. The duration of classification and memory consumption was also considered. Then the χ2 test values were verified between all the categorical variables in the task and the observations classes in dependence of possible hash function output values, when the feature values were hashed. It was affirmed that these relations clearly exist. It is proposed to create a method for choosing dimensionality basing on χ2 test without necessity to train the models to do this selection. Two methods that use hash functions for preparing data for machine learning tasks have been optimized.
Researching snake movement strategies created using machine learning techniques
Defense Date 2018
The aim of this thesis is to create a simulation environment that would allow to conduct and visualize experiments on how snakes move. Different gait strategies, automatically developed using artificial intelligence techniques, are examined and compared with the strategies used by these animals in the real world environment. Record ID
Efficient implementation of induction of decisions rules
Defense Date 2019
This thesis titled ‘‘Efficient implementation of induction of decision rules” introduces different approaches for implementing the rule induction algorithm based on IREP++ and RIPPER algorithms. This algorithm is used to solve classification problems. Its main principle is to create rules, based on data, which will allow to assign class to an example clearly. The purpose of this thesis was to implement this algorithm in a way to obtain high performance together with high accuracy and effective model. Two different implementations are shown in this thesis, presenting varied approaches to data storage and the algorithm. The first one is quite simple and uses dictionaries. One dictionary is created for every conditional variable. Each dictionary contains pairs of values: example index and a value of a variable for this example. Afterwards, using all unique values of all variables, new literals are created and then transformed into rules. The second approach assumes creating a map for every unique value of a variable, which contains only indices of those examples, which have this value. Afterwards, maps are joined using conjunctions and alternatives in order to create rules. Those two ways are implemented in the Python language. Both implementations are tested on quality and performance level. Achieved results are presented in this thesis. There is a huge variety of data sets used in experiments in order to rate the performance and effectiveness of a model in a reliable way.
Mobile application for recognition and translation of text
Defense Date 2019
The aim of this thesis is to design and implement a mobile application for text recognition and translation. This thesis provides a description of theoretical basics related to optical characted recognition and machine translation. It also describes artifficial inteligence models that are used in those areas, focusing on artifficial neural networks. This thesis also depicts realization of the task and contains characterization of every component building the application. It also presents results of conducted tests. At the end of this thesis there is a summary that describes possible further enhancements.
The use of machine learning in construction of the Learned Bloom Filter
Defense Date 2020
The aim of this work is to construct and compare the effectiveness of the Learned Bloom Filter to a traditional Bloom filter. For this purpose, the results on three different datasets, divided into sets with categorical and text data are compared. The theoretical basis of the Bloom filter and the Learned Bloom filter designs are presented, along with their properties. The details of the design of the binary classifiers based on artificial neural networks, used in the Learned Bloom Filter, are discussed, with particular emphasis on the choice of architecture of the networks in terms of their size. The experiments were chosen to highlight problems and their solutions encountered during the design of the Learned Bloom Filter. Based on the experiments, the features of the dataset for which the use of a Learned Bloom Filter has the potential to bring benefits are briefly discussed.
Entity labeling in business documents based on contextual information
Defense Date 2020
In this thesis, I raise the issue of automatic entity labeling in business documents. I propose a solution to this task for the document corpora prepared for this purpose, which correspond with their layout and content to typical texts in the domain. The designed and implemented mechanism improve the results of the reference solution, which is one of the information extraction system components. The annotated data is used to train the named entity recognition model. I develop the approach using contextual information about entity mentions. My main aim is to improve the precision. I use natural language processing and machine learning methods, in particular pretrained language models, clustering and outliers detection. I choose the used algorithms and techniques to be the most suitable for the specified datasets. After thorough experiments I compare the final results of the created mechanism with the reference solution. I also propose many ways to develop this project.
Application aiding in testing resistance of classifier models to attacks using adversarial examples
Defense Date 2020
This thesis presents the development of an application designed to assist in testing the robustness of classifier models against adversarial attacks. The work focuses on evaluating the resistance of machine learning models to adversarial examples - specially crafted inputs designed to fool classification algorithms. The application provides tools for generating adversarial examples and systematically testing model vulnerabilities.
Reversible learning of artificial neural networks
Defense Date 2021
Hyperparameter tuning is one of the main challenges that has to be tackled in order to create an efficient machine learning model. The optimal values of the hyperparameters differ depending on the datasets and the tasks that have to be performed on them. This thesis approaches the topic of gradient-based hiperparameter tuning, using the reversibility of the learning process conducted with the Classic Momentum algorithm. This feature enables the recreation of the learning trajectory and memory-efficient hyperparameter gradient computation. Obtaining such gradient facilitates the construction of more powerful models. The implementation of this solution is going to be presented in the Python programming language with the inclusion of the Tensorflow library.
Generating code autocompletions for Integrated Development Environments
Defense Date 2021
Code autocompletion is a feature that offers suggestions for what a software developer may want to write. Correct suggestions improve efficiency of programmers job. There are a lot of forms of code autocompletion, for example using dictionary methods or parsing code online. In this paper, I propose approach for code autocompletion using machine learning. It generates ranked by its probabilities of occurrences list of suggestions, which can be used by software developer at edit time. I describe explored architectures and compare them to each other in order to select the best one. I also discuss challenges regarding this approach. The final model was trained using 9000 Python source files from publicly available repositories from service GitHub and evaluated on 4500 files. The evaluation results in 68% accuracy for top 5 suggestions, and mean time of response resulted in 18 ms. The system is available as a plugin for integrated development environment SublimeText 3.
Neural network learning algorithms with smoothing
Defense Date 2021
The aim of the work is to investigate the impact of the use of weight smoothing in the neural network model during training on the quality of prediction and the time needed to train it. Certain types of calculating moving averages were used for smoothing. The proposed smoothing algorithms were tested on the VGG19, Wide Residual Networks 28x10 and Densenet 100x12 models. The aim of the work is to examine the prediction results of models with smoothed weights in comparison with models trained without the use of weight averaging, also taking into account the time needed to train the model.
Implementation of the nDES evolutionary strategy on multiple graphics processing units
Defense Date 2021
Classical gradient-based optimization methods used for training neural networks limit the class of possible to use architectures and are the source of several issues. Those include the exploding or vanishing gradient problem. One of the approaches to overcome them assumes using methods that are not based on gradient. The thesis majorly focuses on the evolutionary strategy nDES, which belongs to the group of such non-gradient-based methods. The original implementation assumes computations with a single graphics processing unit. However, the algorithm is scalable and the thesis introduces implementation parallelized across multiple graphics cards. The thesis also goes through the literature and describes some popular neural network architectures, classical optimization methods as well as those rooting from metaheuristics (including evolution strategies). The nDES algorithm is thoroughly described with the included pseudocode. The thesis presents the details of the parallelized implementation. Moreover, there is a description of conducted experiments showing speed-up over the original method and comparing nDES with the classical gradient-based methods. Finally, prospective ways of further development are discussed.
An Investment Assistant Using Machine Learning Methods
Defense Date 2022
The aim of this work is to prepare an investment assistant to support speculative investing in the cryptocurrency market. The developed application provides information about people’ moods and estimates the prices of three cryptocurrencies: Bitcoin, Dogecoin and Ethereum based on social network traffic analysis. Evaluation of public opinion on selected virtual assets is carried out using sentiment analysis methods of posts published on the social network Reddit. Price prediction is made through a machine learning model based on historical market data and sentiment analysis results. The application presents an interactive report constisting of the actual and forecasted price of each cryptocurrency as well as information on social sentiment determined on the basis of the emotional nature of the analysed statements.
Training sets processing neural models
Defense Date 2022
The primary objective of this thesis is to describe and evaluate the efficiency of different models that use sets as inputs, the challenges that they face in comparison to other models, and the problems that they were designed to solve. The main focus was on the Set Transformer, which utilizes the attention mechanism. The thesis addresses the question of whether adding additional complexity to the attention mechanism would enhance the performance of the model. In the scope of the research, several experiments were carried out. The mentioned model was tested on image classification and generating movie recommendations along with a DeepSets model for comparison purposes. The models have been implemented in Python using TensorFlow and Keras libraries. The experiments were run on Google Colab Platform, which allows writing and executing Python code directly in the browser, as well as providing access to GPU power. There were additional neural networks added to the attention implementation, and the modified mechanism was used either in the input layer of the model or all the layers. The networks varied in size and complexity so that the best setup was found that allowed for optimal results. A change in the input layer of the model did indeed enhance its performance for tabular datasets. When images were used as input, however, the results were not significantly better.
Image inpainting using artificial intelligence
Defense Date 2022
The goal of the image inpainting process is to reconstruct images in such a way that the completed area is undetectable by the observer. Recently, with the development of image processing techniques, image reconstruction methods have become an important and challenging research topic. In this paper, the topic of image inpainting using artificial intelligence is introduced. State-of-the-art image reconstruction methods were analyzed and compared with each other, and then one of them was chosen as the base method for the experiments presented in this paper. In the practical part, the influence of the perceptual loss function on the missing data completion process was investigated. The described methods from the deep learning-based group are Context Encoders and Globally and Locally Consistent Image Completion. Those belonging to the attention-based methods are Shift-Net, Contextual Attention, and Coherent Semantic Attention. Finally, all methods were compared with each other and, as they gave the best results, Coherent Semantic Attention was chosen as the base method for further research. The experiments performed and their qualitative and quantitative results are described. Quality metrics, previously described, were used to evaluate the correctness of each experiment.
Metric learning based music recommendation system
Defense Date 2022
Recommender systems use the field of artificial intelligence extensively. Specifically, music recommender systems have experienced a huge expansion, thanks to the emergence of successful online streaming services, like Spotify, Apple Music, or Pandora. As this area of recommendations grows, so do the expectations of users who want to explore the vast amounts of music available at their fingertips. Music recommender systems are faced with many challenges - aspiring to provide better and more personalized recommendations requires to go beyond simple content-based recommendations and tune into listeners’ needs and preferences. This work describes preparation of an engineering thesis being a music recommendation system which uses the Metric Learning approach. It is an attempt to implement a system which delves more into personalized recommendations for users’ listening sessions.
Application of graph neural networks in the expansion of graph sets
Defense Date 2023
This study combines two issues, namely the expansion of sets and the graph neural network (GNN), the latter being a relatively new type of artificial neural networks. The sets are expanded by separating from a large data set a relatively small number of elements similar to each other which are jointly referred to as an expanded set. In the residual data set, referred to as a candidate set, the search of further elements of the expanded set, i.e. those similar to the previously separated elements, is continued. With regard to the expansion of sets it is of great importance that the expanded set is much less numerous than the candidate set and that the latter is contaminated with elements of the expanded set. The other of the methods consists in training semi-supervised networks which consists in labeling elements of the expanded set as 1, increasing the candidate set by a similar number of class 1 elements and labeling all the elements of the candidate set as 0. The network was trained with reference to two types of sets: the two-class Mutagenesis set and the six-class Enzymes set. In the case of the Mutagenesis set the expanded set contained chemical compounds of one type only, and the candidate set – chemical compounds of both types. In the case of the Enzymes set the expanded set contained enzymes of one from among six types, and the candidate set – enzymes of all six types. According to the study results the GNN network is able to learn despite the candidate set being contaminated with elements of the expanded set, while at the same time the number of elements of the expanded set is small compared to the candidate set.
Learning from dependent aggregate observations
Defense Date 2023
The aim of the thesis is to extend the range of applications of a method of learning from aggregate observations [1]. A method of learning from dependent aggregates is introduced. The examined case assumes dependence of the aggregates in form of feature values commonality in aggregated elements. Experiments consist of verification of the method of learning from aggregate observations, examination of the influence of observation dependencies on learning capabilities, introduction of the method for solving the problem of aggregate interdependence along with confirmation of its effectiveness. The results indicated that the proposed method allows learning from dependent observations.
Explaining machine learning models predictions in the context of adversarial data
Defense Date 2023
In my thesis, I investigate the impact of robust learning on the quality of explanations of machine learning models. The use of models in many areas of life causes an increase in the demand for robustness and credibility of predictions of the models used. To this end, I propose 3 experimental methods for assessing the quality of explanations and analysis 4 methods for generating explanations. I compare the robust and classic variants of the random forest and XGBoost models using the proposed evaluation methods and compare the explanations obtained on the 3 data sets with the information from the data mining process.
Tuning small semantic segmentation models via knowledge distillation
Defense Date 2023
The demand for accurate and efficient driver assistance systems is increasing with the rapid development of the autonomous driving industry. Semantic segmentation models play a critical role in enabling the safe and reliable operation of self-driving vehicles. However, the computational requirements of semantic segmentation models pose challenges for real-time deployment on resource-constrained devices. Developers are thus compelled to utilize less complex model architectures, generally achieving inferior results. This thesis aims to enhance the accuracy of a lightweight semantic segmentation model designed for autonomous vehicle applications, through the application of knowledge distillation.
Predictive models trained on aggregated data
Defense Date 2023
The aim of this work is to expand the range of methods trained on aggregated data, which are combinations of all possible attribute pairs. The work compares the results obtained by logistic regression models and the naive Bayes classifier for different datasets based on fine-grained data with the results achieved by modified models based on the aforementioned aggregates. Furthermore, this work presents and examines a new method called RanAgg, based on aggregated data. As part of the research described in this work, the properties of the mentioned methods and the impact of their individual parameters on the speed and quality of predictions based on defined measures were investigated. The results obtained indicate that the proposed modified logistic regression models and the naive Bayes classifier, as well as the new RanAgg method, allow for prediction based solely on aggregated data.
Improving transfer learning under distribution shift
Defense Date 2025
Transfer learning is a popular tool used to create machine learning models in fields such as natural language processing or computer vision. A simple example is the use of a model trained on a certain, typically large, dataset as a feature extractor for another model on a target dataset. A model built this way can achieve better results than a model trained only on the target dataset. A common issue with this approach is the distribution shift occurring between the two datasets, which negatively impacts modeling results. The aim of this work is to introduce the reader to the topic of transfer learning and to compare various knowledge transfer strategies in terms of their robustness to issues arising from distribution shift. The concept of distribution shift is discussed in both mathematical and practical contexts. Three experiments were conducted to compare the effectiveness of three model fine-tuning methods. In two of these experiments, distribution shift was modeled, while in the third, three datasets with varying degrees of shift were used. The results of the experiment suggest that the studied issue paves the way for more effective and efficient training of machine learning models.
Machine learning in chess
Defense Date 2025
The thesis focuses on the application of algorithms and machine learning methods in game simulation and position analysis. Specifically, it describes the process of data collection, designing models capable of generating chess moves, and analyzing given positions. The potential for combining search algorithms with machine learning algorithms was also explored. The initial part of the thesis presents the theoretical foundations of applying machine learning in strategic games, particularly in chess. Key approaches, such as reinforcement learning and supervised learning of deep neural networks, which form the basis of modern chess systems, are described. The practical part of the thesis involves designing and training machine learning algorithms capable of generating legal chess moves based on position or evaluation functions. Open data sources were utilized, including grandmaster-level games and engine-evaluated positions. The models were trained using the PyTorch framework, and the quality of their generated moves was assessed through simulated games against the current top chess engine. The outcome of the thesis is a set of algorithms capable of playing chess games. The summary lists potential future development steps, advantages, and disadvantages of the algorithms.
Synthetic Data Generation for Machine Learning Algorithms
Defense Date 2025
Machine learning has the potential to improve decisions and outcomes in different scientific and real world areas, yet there are many application areas that lack sufficient data for analyses, simulations, and the development of analytical approaches. One way to overcome the issue of data availability is to use synthetic data as an alternative to real data. Synthetic data are simulated from real data by using the underlying statistical properties of the real data to produce synthetic datasets that exhibit these same statistical properties. In this research, we will implement a framework that will generate synthetic data for music-streaming service using the Monte Carlo Simulation, which allows us to pre define the statistical properties of the generated datasets, and then test it on both predictor and recommender models to evaluate the level of control we have over the data and it’s impact on the result.
A study on the usefulness of the SHAP method in the process of hyperparameter tuning for machine learning methods
Defense Date 2025
This master’s thesis focuses on the analysis and assessment of the potential of the SHAP method (SHapley Additive exPlanations) in the context of tuning hyperparameters for machine learning methods. Originating from the field of XAI (eXplainable Artificial Intelligence), the SHAP method allows for a better interpretation of the results generated by machine learning models, highlighting the contribution of individual features to the decision-making process of the model. This work introduces an innovative approach to using the SHAP method in hyperparameter optimization. Instead of traditional methods like grid search or random search, a method leveraging SHAP was proposed, wherein the hyperparameter tuning process is based on information obtained from SHAP. This helps identify which hyperparameters have the most significant impact on the model’s performance in the context of a specific problem.
Analysis of listed companies’ reports using language models
Defense Date 2025
The aim of this engineering thesis is to develop a tool that supports stock market investors in analyzing corporate financial statements by leveraging advanced natural language processing and data extraction technologies. The presented solution enables automatic processing of financial documents using Azure Document Intelligence services and the ChatGPT model, allowing for rapid acquisition and presentation of key information contained in financial reports in a user-friendly manner. The application is designed for investors employing fundamental analysis, enabling them to efficiently process extensive and complex financial documents. The user interface was developed using the Streamlit framework, ensuring ease of use and interactivity. The main system components-namely user interaction, text extraction, and data processing-are described in detail along with the rationale for the selection of the applied tools and technologies. The implementation of these solutions significantly improves the efficiency of financial data analysis, resulting in time savings and enhanced accuracy in investment decision-making.
Evaluation of Few-Shot Methods from the Perspective of Class Scaling
Defense Date 2025
Modern artificial intelligence models in computer vision applications can consist of billions of parameters, and their training may take several days. This entails the need for vast amounts of training data and significant energy consumption during training process. In the context of real-world business problems, there is a growing demand for models that are more flexible, faster, and have lower hardware requirements. In response to these needs, recent research has focused on developing universal methods that do not require costly retraining. Few-shot learning models have gained considerable interest due to their efficiency. Most of them emphasize the use of high-quality image feature extractors. However, an important yet often overlooked aspect of these methods is their scalability with respect to the increasing number of classes in classification tasks. Existing research rarely addresses this issue, instead focusing on achieving top results on the most popular benchmarks. This thesis aims to analyze and compare selected state-of-the-art few-shot learning methods in terms of their ability to scale with the number of classes in image classification problems. It contrasts large models with approaches that use feature extractors and classic clustering algorithms, highlighting the significant impact of the chosen extractor on final performance.
The use of large language models in active learning methods
Defense Date 2025
In this thesis, I address the issue of using large language models (LLMs) as oracles in the active learning process, where an LLM labels samples selected by specific strategies. To this end, I focus on currently used LLMs and appropriate “Prompt Engineering” techniques to obtain the best possible labels. While researching the topic, I analyzed the available scientific literature, searching for similar solutions. I observed that current research may suffer from the problem of data leakage. To mitigate this issue, I prepared my own completely new dataset to minimize the chance of this phenomenon occurring. I confirmed that for this new data, the use of an LLM in the active learning process yields good results. Analyzing the subject of active learning, I selected the best and most diverse strategies to compare how they perform in the presence of noise introduced by the LLM. The experiments show that for the selected datasets, an uncertainty-based strategy performs best; therefore, the implementation complexity and computational overhead do not justify the introduction of the other strategies. Additionally, this work consolidates the scattered and rapidly changing knowledge in this field, while simultaneously conducting uniform experiments at a single point in time.
portfolio
Portfolio item number 1
Short description of portfolio item number 1
Portfolio item number 2
Short description of portfolio item number 2
publications
Adversarial examples: A survey
Published in 2018 Baltic URSI Symposium (URSI), 2018
Conference paper providing a comprehensive survey of adversarial examples in machine learning.
Recommended citation: Zawistowski, P. (2018). "Adversarial examples: A survey." 2018 Baltic URSI Symposium (URSI), CFP18N89-ART, 295–298.
Download Paper
Paraphrase generation and evaluation: a view from the trenches
Published in Proceedings of SPIE: Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments 2019, 2019
Conference paper presenting practical insights into paraphrase generation and evaluation methods.
Recommended citation: Franus, W., Twardowski, B., Zawistowski, P., & Nowak, R. M. (2019). "Paraphrase generation and evaluation: a view from the trenches." W R. Romaniuk & M. G. Linczuk (Redaktorzy), Proceedings of SPIE: Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments 2019.
Download Paper
Machine learning models for predicting customer decision in motor claims settlements
Published in Proceedings of SPIE: Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments 2019, 2019
Conference paper on applying machine learning models to predict customer decisions in motor insurance claims settlements.
Recommended citation: Nowak, R. M., Neumann, Ł., Franus, W., Dąmbski, M., Smółkowski, A., & Zawistowski, P. (2019). "Machine learning models for predicting customer decision in motor claims settlements." W R. Romaniuk & M. G. Linczuk (Redaktorzy), Proceedings of SPIE: Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments 2019.
Download Paper
Deep Learning Optimization Tasks and Metaheuristic Methods
Published in Fundamenta Informaticae, 2019
Journal article on applying metaheuristic methods to deep learning optimization tasks.
Recommended citation: Biedrzycki, R., Zawistowski, P., & Twardowski, B. (2019). "Deep Learning Optimization Tasks and Metaheuristic Methods." Fundamenta Informaticae, 168, Article 2–4.
Download Paper
Automatic hyperparameter tuning in on-line learning: Classic Momentum and ADAM
Published in 2020 International Joint Conference on Neural Networks (IJCNN), 2020
Conference paper on automatic hyperparameter tuning methods for online learning algorithms, focusing on momentum and ADAM optimizers.
Recommended citation: Wawrzyński, P., Zawistowski, P., & Lepak, Ł. (2020). "Automatic hyperparameter tuning in on-line learning: Classic Momentum and ADAM." 2020 International Joint Conference on Neural Networks (IJCNN), 1–8.
Download Paper
Deep Neuroevolution: Training Neural Networks Using a Matrix-Free Evolution Strategy
Published in Neural Information Processing, Part I, 2021
Conference paper on deep neuroevolution methods for training neural networks using matrix-free evolution strategies.
Recommended citation: Jagodziński, D., Neumann, Ł., & Zawistowski, P. (2021). "Deep Neuroevolution: Training Neural Networks Using a Matrix-Free Evolution Strategy." W T. Mantoro, M. Lee, M. A. Ayu, K. W. Wong, & A. N. Hidayanto (Redaktorzy), Neural Information Processing, Part I.
Download Paper
Metric Learning for Session-Based Recommendations
Published in Advances in Information Retrieval, 43rd European Conference on IR Research, ECIR 2021, 2021
Conference paper on applying metric learning techniques to session-based recommendation systems.
Recommended citation: Twardowski, B., Zawistowski, P., & Zaborowski, S. (2021). "Metric Learning for Session-Based Recommendations." W D. Hiemstra, M.-F. Moens, J. Mothe, R. Perego, M. Potthast, & F. Sebastiani (Redaktorzy), Advances in Information Retrieval, 43rd European Conference on IR Research, ECIR 2021, Proceedings.
Download Paper
Testing and Verification of the Deep Neural Networks Against Sparse Pixel Defects
Published in SAFECOMP 2022 Workshops - Lecture Notes in Computer Science, 2022
Conference paper on testing deep neural networks against sparse pixel defects published in SAFECOMP 2022 Workshops proceedings.
Recommended citation: Szczepankiewicz, M., Radlak, K., Szczepankiewicz, K., Popowicz, A., & Zawistowski, P. (2022). "Testing and Verification of the Deep Neural Networks Against Sparse Pixel Defects." W M. Trapp, E. Schoitsch, J. Guiochet, & F. Bitsch (Redaktorzy), Proceedings of the Computer Safety, Reliability, and Security. SAFECOMP 2022 Workshops.
Download Paper
Ground truth based comparison of saliency maps algorithms
Published in Scientific Reports, 2023
Research article published in Scientific Reports comparing saliency map algorithms using ground truth based methodology.
Recommended citation: Szczepankiewicz, K., Popowicz, A., Charkiewicz, K., Nałęcz-Charkiewicz, K., Szczepankiewicz, M., Lasota, S., Zawistowski, P., & Radlak, K. (2023). "Ground truth based comparison of saliency maps algorithms." Scientific Reports, Article 13.
Download Paper
Exploring the microstructure–property relationship in polymer foams using advanced statistical methods, machine learning and deep learning: A review
Published in Computational Materials Science, 2025
This review explores the application of advanced statistical methods, machine learning, and deep learning techniques for understanding microstructure-property relationships in polymer foams.
Recommended citation: Walicki, D., Zawistowski, P., & Ryszkowska, J. (2025). "Exploring the microstructure–property relationship in polymer foams using advanced statistical methods, machine learning and deep learning: A review." Computational Materials Science. 256, 1-14.
Download Paper
talks
Talk 1 on Relevant Topic in Your Field
Published:
This is a description of your talk, which is a markdown file that can be all markdown-ified like any other post. Yay markdown!
Conference Proceeding talk 3 on Relevant Topic in Your Field
Published:
This is a description of your conference proceedings talk, note the different field in type. You can put anything in this field.
teaching
Teaching experience 2
Workshop, University 1, Department, 2015
This is a description of a teaching experience. You can use markdown like any other post.
IUM
Lecture, EiTI, PW, 2025
IUM (Inżynieria Uczenia Maszynowego)