Rozdziały
Stały URI dla kolekcjihttp://hdl.handle.net/11652/4775
Przeglądaj
Pozycja 3D Reconstruction of Non-Visible Surfaces of Objects from a Single Depth View – Comparative Study(Wydawnictwo Politechniki Łódzkiej, 2023) Staszak, Rafał; Michałek, Piotr; Chudziński, Jakub; Kopicki, Marek; Belter, DominikScene and object reconstruction is an important problem in robotics, in particular in planning collision-free trajectories or in object manipulation. This paper compares two strategies for the reconstruction of nonvisible parts of the object surface from a single RGB-D camera view. The first method, named DeepSDF predicts the Signed Distance Transform to the object surface for a given point in 3D space. The second method, named MirrorNet reconstructs the occluded objects’ parts by generating images from the other side of the observed object. Experiments performed with objects from the ShapeNet dataset, show that the view-dependent MirrorNet is faster and has smaller reconstruction errors in most categories.Pozycja Aaron Earned an Iron Urn: Speech-to-IPA Models Improve Diagnostic of Pronunciation(Wydawnictwo Politechniki Łódzkiej, 2023) Olejnik, Franciszek; Stachowiak, Rafał; Krysińska, Izabela; Morzy, MikołajLearning the proper pronunciation is one of the key aspects of foreign language acquisition. Assessment of the correctness of pronunciation requires the involvement of expert phoneticians and linguists, severely limiting the scalability of learning solutions. However, the recent adaptation of the Transformer architecture to the audio domain opens the way for automatic model-based assessment of pronunciation. In this paper, we present the pronunciation diagnostic tool developed at PUT and we experimentally evaluate the correlation between expert human assessment and automatic model assessment. By combining the Wav2Vec model and the IPA representation, we prove that pronunciation assessment can be performed automatically with high precision.Pozycja AI-driven Ecodriving and ETA Solutions for Truck Transport(Wydawnictwo Politechniki Łódzkiej, 2023) Lipiński, Piotr; Morawski, Michał; Napieralski, Piotr; Nowok, Paweł; Zawiślak, Bartosz; Hojdys, Leszek; Lazar, Marcin; Lazarek, Przemysław; Zając, Norbert; Pizoń, Sylwester; Jakubiec, Rafał; Sienkiewicz, Jacek; Gołąbek, Sebastian; Kabocik, Mateusz; Fedrizzi, Szymon; Kuliga, Michał; Frączkiewicz, Mateusz; Malarz, Mirosław; Puchalski, Jarosław; Danysz, Ewa; Grajcarek, MaciejPozycja AloneKnight – Enabling Affective Interaction within Mobile Video Games(Wydawnictwo Politechniki Łódzkiej, 2023) Jemioło, Paweł; Świder, Krzysztof; Storman, Dawid; Adrian, Weronika T.Artificial intelligence is used in various contexts, including video games, where it can enhance the game design and adapt content to players’ emotional states through affective computing. In this paper, we present an example of an affective mobile game and compare participants’ opinions after playing two versions of the game, with and without an affective loop. The game was developed using Unity. In the affective version, physiological data is recorded and analysed to detect emotions based on facial expressions and electrodermal activity, which then affects the game. The study with 11 participants showed positive feedback for the game with affective loop.Pozycja AMUseBot: Towards Making the Most out of a Task-oriented Dialogue System(Wydawnictwo Politechniki Łódzkiej, 2023) Christop, Iwona; Dudzic, Kacper; Krzymiński, MikołajThis paper presents AMUseBot, a task-oriented dialogue system designed to assist the user in completing multi-step tasks. Taking into consideration that the fundamental issues with such systems are poor user ratings and high rates of uncompleted tasks, the main goal of the project is to keep the user focused and provide engaging conversations. We approach these problems by the introduction of dynamic multimodal communication and graph-based task management.Pozycja Analysis of Surface EMG Signals to Control of a Bionic Hand Prototype(Wydawnictwo Politechniki Łódzkiej, 2023) Pieprzycki, Adam; Król, Daniel; Wawryka, Piotr; Łachut, Katarzyna; Hamera, Mateusz; Srebro, BartoszThe aim of the presented project is to develop a comprehensive system for acquiring surface EMG data and carry out time-frequency analysis to determine useful parameters for subsequent gesture classification for a simple bionic hand prosthesis. This system is expected to assist in controlling both the prosthetic hand and the robotic hand in making precise gestures with the fingers on the hand. The article presents the methods for acquiring and processing multi-channel EMG signals and feature extraction for gesture recognition by an artificial neural network (ANN).Pozycja Anonymizer for Polish Language(Wydawnictwo Politechniki Łódzkiej, 2023) Walkowiak, Tomasz; Gniewkowski, Mateusz; Pogoda, Michał; Ropiak, NorbertResearchers and enterprises require anonymization of unstructured text. This is not only due to the GDPR regulation, but also due to the increasing use of large language models (LLMs) such as GPT-3, where there is growing concern about the privacy and security risks associated with these models. The texts to be processed by such models need to be anonymized beforehand, and very often they need to be anonymized at the data providers’ premises rather than at the machine learning teams. In this paper, we present an effective anonymization pipeline for Polish. It provides a modular and configurable solution that employs different modes, including the challenging pseudo-anonymization mode in languages with complex inflectional systems. The system can be easily integrated with existing systems and deployed in different environments using a microservices architecture solution with a REST interface.Pozycja Application of Pawlak’s Conflict Model to Generate Coalitions of Local Tables with Similar Values on Conditional Attributes(Wydawnictwo Politechniki Łódzkiej, 2023) Przybyła-Kasperek, Małgorzata; Kusztal, KatarzynaGenerating a model or pattern based on dispersed data available in many different tables is difficult because there can be numerous inconsistencies in the data. One way to deal with such a problem is to analyze conflicts and generate coalitions of consistent local tables. This paper proposes a model in which coalitions of tables with consistent data are created using Pawlak’s conflict analysis approach. A model, decision tree, is created based on the aggregated data within the coalition. This way, we get rules that better describe the concepts found in consistent local tables.Pozycja Are Quantified Boolean Formulas Hard for Reason-Able Embeddings?(Wydawnictwo Politechniki Łódzkiej, 2023) Potoniec, JędrzejWe aim to establish theoretical boundaries for the applicability of reason-able embeddings, a recently proposed method employing a transferable neural reasoner to shape a latent space of knowledge graph embeddings. Since reason-able embeddings rely on the ALC description logic, we construct a dataset of the hardest concepts in ALC by translating quantified boolean formulas (QBF) from QBFLIB, a benchmark for QBF solvers. We experimentally show the dataset is hard for a symbolic reasoner FaCT++, and analyze the results of reasoning with reason-able embeddings, concluding that the dataset is too hard for them.Pozycja Autoregressive Label-Conditioned Autoencoder for Controllable Image-To-Video Generation(Wydawnictwo Politechniki Łódzkiej, 2023) Kubicki, Kacper; Ślot, KrzysztofGenerating videos from a single image with user-controlled attributes is a complex challenge in the field of computer vision, despite the significant advancements recently made in the field. This paper presents a novel approach to tackle this issue, leveraging a convolutional autoencoder with supervised principal component analysis and autoregressive inference step. The efficacy of the proposed method is evaluated on two datasets – MNIST handwritten-digits and time-lapse photos of the sky. Results from both quantitative and qualitative analyses show that the proposed approach produces high-quality videos of variable duration with user-defined attributes, while preserving the integrity of original image contents.Pozycja BDOT10k-seg: A Dataset for Semantic Segmentation(Wydawnictwo Politechniki Łódzkiej, 2023) Kos, Aleksandra; Majek, KarolIn this work, we describe BDOT10k-seg, a novel aerial dataset for semantic and instance segmentation. Our data covers almost the entire territory of Poland (314,000 km2) and provides precise pixel-level annotations for 286 classes of topographical objects, including buildings, roads, rivers, lakes, airports, agricultural areas, and forests. BDOT10-seg consists of 60,718 images with a resolution of 3 to 75 centimeters per pixel, and more than 40 million object instances. The average image size is 12,367 px because, unlike other publicly available datasets, we do not modify the source orthoimages. The code for generating the BDOT10k-seg dataset is publicly available.Pozycja Beacon-based Swarm Search and Rescue(Wydawnictwo Politechniki Łódzkiej, 2023) Ratnayake, Sunil; Figat, MaksymPozycja Brief Overview of Selected Research Directions and Applications of Process Mining in KRaKEn Research Group(Wydawnictwo Politechniki Łódzkiej, 2023) Kluza, Krzysztof; Zaremba, Mateusz; Sepioło, Dominik; Wiśniewski, Piotr; Adrian, Weronika T.; Gaudio, Maria Teresa; Jemioło, Paweł; Adrian, Marek; Jobczyk, Krystian; Ślażyński, Mateusz; Stachuta-Terlecka, Bernadetta; Ligęza, AntoniProcess mining allows for exploring processes using data from event logs. By providing insights into how processes are actually executed, rather than how they are supposed to be executed, process mining can be used for optimizing business processes and improving organizational efficiency. In this exploratory paper, we report on selected research threads related to process mining carried out within KRaKEn Research Group at AGH University of Science and Technology. We introduce a collection of initial ideas that require further exploration. Our research threads are concerned with the use of process mining techniques 1) for discovering processes from unstructured data, specifically text from e-mails, 2) for explaining black-box machine learning models, using process models as a global explanation, and 3) for analyzing data from different food industry systems to identify inefficiencies and provide recommendations for improvement.Pozycja Building Energy Use Intensity Prediction with Artificial Neural Networks(Wydawnictwo Politechniki Łódzkiej, 2023) Stokfiszewski, Kamil; Sztoch, Przemysław; Sztoch, Ryszard; Wosiak, AgnieszkaIn this paper the authors propose the construction and examine the performance of the artificial neural network for energy use intensity prediction for residential buildings. The network’s type is the standard multilayer perceptron and its training dataset contains the data of 768 residential buildings where the training pattern for an individual building consists of 8 parameters describing the building’s geometry along with its lighting and glazing conditions while the only output value is the building’s actual energy use intensity characteristics. Experimental study shows that the mean absolute percentage error of prediction of the energy use intensity evaluated for buildings data present in the network’s test set does not exceed 1.8%, what might be considered a highly satisfactory result.Pozycja Carbon Footprint Reduction of a Petrochemical Process Supported by ML and Digital Twin modelling(Wydawnictwo Politechniki Łódzkiej, 2023) Kulikowski, Sławomir; Romanowski, Andrzej; Sierszeń, ArturThis article aims to present a concept of an Artificial Intelligence application in the form of pre-trained Machine Learning modules to reduce the carbon footprint of a chemical recycling process. Chemical recycling of plastic is energy-consuming as it requires relatively high temperatures and calibration cycles based on a constantly changing structure of raw materials. Due to that fact, complex process parameters must be tuned to allow the production of the required fraction of gasoline. In general, the designed IoT system enables a massive collection of technology and environmental data and the processing of parameters to feed the Digital Twin of a petrochemical plant. The scientific part of the project consists of Digital Twin modelling, experiments, simulations, and training of machine learning modules to predict the optimal set of production line parameters based on the specific structure of raw materials to reduce the number of calibrations and lower energy consumption indirectly which will lead to carbon footprint reduction. There is an here is an estimate that that deployed solution will allow reduction of energy consumption on a monthly level of 10-15% which could generate direct savings on a cost of energy but also savings in a field of carbon emission and related credits. The project also includes the evaluation of predictions supported by machine learning modules, training techniques and comparison to expert settings to assess the quality of the application.Pozycja Challenges of Crop Classification from Satellite Imagery with Eurocrops Dataset(Wydawnictwo Politechniki Łódzkiej, 2023) Aszkowski, Przemysław; Kraft, MarekCrops monitoring and classification on a nationwide level provide important information for sustainable agricultural management, food security, and policy-making. Recent technological advancements, followed by Earth observation programmes like Copernicus, have provided plenty of publicly available multispectral data. Combining these data with field annotations allows for continuous crop monitoring from publicly available data. In this paper, we present a solution for crop classification to determine crop type from Sentinel-2 multispectral data, utilizing machine learning techniques. Apart from presenting initial results, we discuss the challenges of crop classification on a Eurocrops dataset and further research directions.Pozycja Clustering Dilemmas – A Study of the Request of Homogenicity within Clusters Versus Diversity Between Clusters(Wydawnictwo Politechniki Łódzkiej, 2023) Kłopotek, Mieczysław AlojzyAn interplay between the requirements of within-cluster-homogenicity and between-clusters-diversity is investigated. It is shown that taking the requirements of homogenicity and diversity makes the clustering an easy task, but these requirements are rarely matched in the practise.Pozycja A Comparison of Shallow Explainable Artificial Intelligence Methods against Grammatical Evolution Approach(Wydawnictwo Politechniki Łódzkiej, 2023) Sepioło, Dominik; Ligęza, AntoniThis paper reports on an ongoing, innovative research in the area of eXplainable Artificial Intelligence (XAI). An XAI task is considered as finding an explanation of the model generated via Machine Learning by identifying the most influential variables for local decision-making. The proposed approach moves the explanatory process to a new, deeper-level dimension. It is oriented towards Model Discovery, i.e. the internal structure and functions of the components. An experiment on Function Discovery via Grammatical Evolution is reported in brief.Pozycja Contextual ES-adRNN with Attention Mechanisms for Forecasting(Wydawnictwo Politechniki Łódzkiej, 2023) Smyl, Sławek; Dudek, Grzegorz; Pełka, PawełIn this study, we propose a hybrid contextual forecasting model with attention mechanisms for generating context information. The model combines exponential smoothing and recurrent neural network to extract and synthesize information at both the individual series and collective dataset levels. The model is composed of two simultaneously trained tracks: context track and main track. The main track generates forecasts and predictive intervals, while the context track generates additional inputs for the main track based on representative time series. Attention mechanisms are integrated into the model in six different variations to adjust the context information to the forecasted series and so increase the predictive power of the model.Pozycja A Convolutional and Recurrent Neural Network-based Approach for Speech Emotion Recognition(Wydawnictwo Politechniki Łódzkiej, 2023) Duch, Piotr; Wiatrowska, Izabela; Kapusta, PawełSpeech emotion recognition (SER) is a crucial aspect of humancomputer interaction. In this article, we propose a deep learning approach, using CNN and RNN architectures, for SER using both convolutional and recurrent neural networks. We evaluated the approach on four audio datasets, including CREMA-D, RAVDESS, TESS, and EMOVO. Our experiments tested various feature sets and extraction settings to determine optimal features for SER. Our results demonstrate that the proposed approach achieves high accuracy rates and outperforms state-of-the-art algorithms.