Rozdziały
Stały URI dla kolekcjihttp://hdl.handle.net/11652/4775
Przeglądaj
Pozycja Simulation of the Quantum Heat Engine in the Quantum Register(Wydawnictwo Politechniki Łódzkiej, 2023) Ostrowski, MarcinThis paper investigates whether a quantum computer can efficiently simulate the transfer of excitation between a pair of quantum systems with energy loss caused by photon or phonon emission. The main contribution of our work is an algorithm that enables the simulation of time evolution of such a system, implemented on a standard two-input gates. The paper examines the properties of the proposed algorithm and then compares the obtained results with theoretical predictions.Pozycja Increasing Skin Lesions Classification Rates using Convolutional Neural Networks with Invariant Dataset Augmentation and the Three-Point Checklist of Dermoscopy(Wydawnictwo Politechniki Łódzkiej, 2023) Milczarski, Piotr; Borowski, Norbert; Beczkowski, MichałIn the paper, we show how to tackle the problem of lack of the rotation invariance in CNN networks using the authors’ Invariant Dataset Augmentation (IDA) method. The IDA method allows to increase the classification rates taking into account as an example the classification of the skin lesions using a small image set. In order to solve the problem of the lack of rotation invariance, IDA method was used and the dataset was increased in an eightfold and invariant way. In the research, we applied the IDA methods and compared the results of VGG19, XN and Inception-ResNetv2 CNN networks in three skin lesions features classification defined by wellknown dermoscopic criterions e.g. the Three-Point Checklist of Dermoscopy or the Seven-Point Checklist. Due to Invariant Dataset Augmentation, the classification rate parameters like true positive rate by almost 20%, false positive rate as well as the F1 score and Matthews correlation coefficient have been significantly increased opposite to type II error that has significantly decreased. In the paper, the confusion matrix parameters result in: 98-100% accuracy, 98-100% true positive rate, 0.0-2.3% false positive rate, tests F1=0.95 and MCC=0.95. That general approach can provide higher results while using CNN networks in other applications.Pozycja A Hybrid Fuzzy-Rough Approach to Handling Missing Data in a Fall Detection System(Wydawnictwo Politechniki Łódzkiej, 2023) Mroczek, Teresa; Gil, Dorota; Pękala, BarbaraPozycja Grounded HyperSymbolic Representations Learned through Gradient-Based Optimization(Wydawnictwo Politechniki Łódzkiej, 2023) Łuczak, Piotr; Ślot, Krzysztof; Kucharski, JacekHyperdimensional computing is a novel paradigm, capable of processing complex data structures with simple operations. Its main limitations lie in the conversion of real world data onto hyperdimensional space, which due to lack of a universal translation scheme, oftentimes requires application-specific methods. This work presents a novel method for unsupervised hyperdimensional conversion of arbitrary image data. Additionally, this method is augmented by the ability of creating HyperSymbols, or class prototypes, provided that such class labels are available. The proposed method achieves promising performance on MNIST dataset, both in translating individual samples as well as producing HyperSymbols for downstream classification task.Pozycja Machine Learning for Water Leak Detection and Localization in the WaterPrime Project(Wydawnictwo Politechniki Łódzkiej, 2023) Głomb, Przemysław; Romaszewski, Michał; Cholewa, Michał; Koral, Wojciech; Madej, Andrzej; Skrabski, Maciej; Kołodziej, KatarzynaWe present an integrated approach for water leak detection and localization developed for the WaterPrime project. Proposed method is based on telemetric monitoring of a District Metered Areas (DMA), using first an application of anomaly detection on sensors’ data and then building a ‘digital twin’ of a DMA state using a combination of hydraulic simulator and machine learning algorithms. This approach leads to reduction of time of leak location estimation from the order of weeks/months to days, and provides a significant reduction in quantity of water lost, as was preliminary verified in two waterworks associated with the project.Pozycja A Novel DNN-based Image Watermarking Algorithm(Wydawnictwo Politechniki Łódzkiej, 2023) Kovačević, Slavko; Pavlović, Kosta; Djurović, IgorDNN architecture for image watermarking that balances the tasks of embedding and detecting a watermark is presented. The system consists of two networks: the embedder and the detector. A loss function based on a structural similarity index measure minimizes the difference between the original and watermarked signal. The average SSIM is 0.98 while the accuracy is 99.99%.Pozycja A Reinforcement Learning Framework for Motion Planning of Autonomous Vehicles(Wydawnictwo Politechniki Łódzkiej, 2023) Orłowski, Mateusz; Skruch, PawełThe paper introduces a framework that has been developed for the design and verification of motion planning algorithms for autonomous driving. The framework allows for the use of reinforcement learning for autonomous driving that requires complex and computationally intensive simulations. The key element in the presented approach plays a multi-agent closed-loop simulation of the traffic environment. Using the framework, the training process can be performed in parallel on high-performance computing clusters. Therefore, the framework provides an easy way to explore the potential of reinforcement learning for autonomous driving applications.Pozycja Towards Detection of Unknown Polymorphic Patterns Using Prior Knowledge(Wydawnictwo Politechniki Łódzkiej, 2023) Kucharski, Przemysław; Ślot, KrzysztofThe presented paper proposes a novel approach for detecting unknown polymorphic patterns in sequences composed of random symbols and of known polymorphic patterns. We propose to represent rules that drive pattern generation as regular expressions. To detect unknown patterns, we first incorporate knowledge on known rules into a Convolutional Autoencoder (CAE), then we train the CAE with additional objective to prevent weights from learning the already known patterns. Analysis of training results provides statistically significant information on presence or absence of polymorphic patterns that were not previously known.Pozycja Loss Function Influence on Uncertainty Estimation for White Matter Lesions 3D Segmentation in a Shifted Domain Setting(Wydawnictwo Politechniki Łódzkiej, 2023) Kaczmarska, Marta; Majek, KarolThe aim of this study is to address the problem of distributional shift for white matter Multiple Sclerosis lesion segmentation models. The impact of loss function on models performance and uncertainty estimation is evaluated. The evaluation is performed on two in-domain and one out-ofdomain dataset consisting of 3D FLAIR Magnetic Resonance images. Our experiments show that application of segmentation losses (eg. Dice) translate into reduced models robustness and poorer uncertainty estimation compared with classification losses (eg. CE). The source code is publicly available.Pozycja A New Approach to Learning of 3D Characteristic Points for Vehicle Pose Estimation(Wydawnictwo Politechniki Łódzkiej, 2023) Nowak, Tomasz; Skrzypczyński, PiotrThis article discusses the challenges of estimating the pose of a vehicle from monocular images in an uncontrolled environment. We propose a new neural network architecture that learns 3D characteristic points of vehicles from image crops and coordinates of 2D keypoints on images. To facilitate supervised training of this network, we pre-process the ApolloCar3D dataset to obtain labelled 3D characteristic points of different car models. We evaluate our approach on the ApolloCar3D benchmark and demonstrate results competitive to state-of-the-art methods.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 Customer Churn Analytics Using Monotonic Rules(Wydawnictwo Politechniki Łódzkiej, 2023) Szeląg, Marcin; Słowiński, RomanUsing bank customer churn data, we demonstrate the explanatory and predictive capacity of monotonic decision rules. Since the data are partially ordinal, they are structured by a new version of the Variable Consistency Dominance-based Rough Set Approach before the induction of monotonic decision rules. The induced rules characterize loyal customers and the ones who left the bank. Such an approach is in line with explainable AI, aiming to obtain a transparent and understandable decision model. In the course of a computational experiment, we compare the predictive performance of monotonic rules with several well-known machine learning models.Pozycja Lessons Learned from a Smart City Project with Citizen Engagement(Wydawnictwo Politechniki Łódzkiej, 2023) Ernst, Sebastian; Zaworski, Konrad; Sokołowski, Piotr; Salwa, GrzegorzThe paper discusses the experiences gained in a joint research project by AGH and the commune of Siechnice. Two main areas are discussed: collecting data from heterogenous sensor devices as well as input from citizens, and development of analytic procedures in a way which guarantees integration between day-to-day and research operations. The most prominent outcomes of the project include the development of a living lab as well as automation of multi-aspect inference, which would normally have to be carried out by a team of experts.Pozycja Recognition of Shoplifting Activities in CCTV Footage Using the Combined CNN-RNN Model(Wydawnictwo Politechniki Łódzkiej, 2023) Kirichenko, Lyudmyla; Pichugina, Oksana; Sydorenko, Bohdan; Yakovlev, SergiyThe recognition of human activities through surveillance has numerous applications across various fields. This article presents a proposed approach to identify shoplifting in camera-recorded video data using a neural classifier that combines two neural networks, specifically, convolutional and recurrent networks. The hybrid architecture consists of two parallel streams: initial and processed video fragments (histogram of oriented gradients and optical flow). The convolutional network extracts features from each frame of the video fragment, while the recurrent network processes the temporal information from sequences of frames as features to classify the activity.Pozycja Towards Ontology-Driven Verification of Car Claims Settlement(Wydawnictwo Politechniki Łódzkiej, 2023) Pancerz, Krzysztof; Wolski, JacekIn the paper, we outline an intelligent tool enabling the users to automatize the process of verification of the car claims settlement. Two data sources power the tool. The first one is the source of car images in which damaged elements are recognized. The second one is the source of PDF files in which cost estimates are extracted. The designed ontology of car repair, described in the paper, is used both in the pre-processing step and in recognition of a typical situations.Pozycja On Parameters of Migration in PEA Computing(Wydawnictwo Politechniki Łódzkiej, 2023) Biełaszek, Sylwia; Byrski, AleksanderMetaheuristics, such as evolutionary algorithms have been proven to be (also theoretically, see works of Vose [1]) universal optimization methods. Skolicki and DeJong [2] researched impact of migration intervals on island models. In this article, we explore different migration intervals and amounts of migrating indyviduals, complementing Skolicki and DeJong’s research. In our experiments we use different ways of selecting migrants and pave the way for further research, e.g. involving different topologies and neighborhoods. Besides sketching out the background and presenting the idea of the algorithm we show the experimental results and discuss them in detail.Pozycja Supporting Surgical Training with the Help of Computer Vision and Machine Learning Methods(Wydawnictwo Politechniki Łódzkiej, 2023) Forczmański, Paweł; Ryder, Yoonhee C; Mott, Nicole M; Gross, Christopher L.; Yu, Joon B.; Rooney, Deborah M.; Jeffcoach, David R.; Bidwell, Serena; Anidi, Chioma; Rosenthal, Lindsay; Kim, Grace J.The paper presents a novel concept of laparoscopic skills evaluation based on the automated analysis of videos recorded during simulationbased training exercises via an artificial intelligence algorithm. It has been tested on data collected during the training of actual surgeons. Its performance is promising, providing an opportunity to build an automatic system used mainly in developing countries.Pozycja Statistical Method for Photovoltaic Power Forecasting Basing on Signal Components Decomposition(Wydawnictwo Politechniki Łódzkiej, 2023) Parczyk, Paweł; Burduk, RobertSince climate and environmental protection have become an important point for society, the industry and business have focused on increasing the share of renewable energy sources in the energy mix. This brought us new challenges. In this paper, we propose a method for photovoltaic power production forecasting. We compared our model with a state-of-the-art Auto Regressive model. We used Mean Absolute Error and Mean Absolute Percentage Error as metrics. Finally, our model turned out to be statistically better than reference model in generating one-hour and two-and-a-half-hour forecasts.Pozycja NeRF-based RGB-D Images Generation in Robotics – Experimental Study(Wydawnictwo Politechniki Łódzkiej, 2023) Kulecki, Bartłomiej; Belter, DominikMultiple learning-based algorithms in robotics require collecting RGB-D images of the scene from various viewpoints. These procedures are time-consuming, so many methods are trained using synthetic images. Recently, a Neural Radiance Fields (NeRF) model of the scene was proposed. Moreover, recent methods show that this model can be trained in minutes. This opens the possible applications in robotics for training the systems to reconstruct scenes, grasp objects or estimate their 3D poses using RGB-D images generated from a small number of input images. In this paper, we verify the quality of RGB-D images generated by the Instant Neural Graphics Primitives implementation of NeRF. We compare the obtained results from the Instant NeRF with the ground-truth RGB-D images obtained from the Kinect Azure and images generated from the point cloud model of the scene. The results show that the difference between generated RGB-D images and ground truth images is small, especially near the object.Pozycja Optimized Mutation Operator in Evolutionary Approach to Stackelberg Security Games(Wydawnictwo Politechniki Łódzkiej, 2023) Żychowski, Adam; Mańdziuk, JacekIn this paper, we introduce several mutation modifications in Evolutionary Algorithm for finding Strong Stackelberg Equilibrium in sequential Security Games. The mutation operator used in the state-of-the-art evolutionary method is extended with several greedy optimization techniques. Proposed mutation operators are comprehensively tested on three types of games with different characteristics (in total over 300 test games). The experimental results show that application of some of the proposed mutations yields Defender’s strategies with higher payoffs. A trade-off between the results quality and the computation time is also discussed.