Examensarbeten för masterexamen
Länka till denna samling:
Browse
Senast publicerade
- PostEvaluation of Neural-Network and Large-Language Model Approaches for Generating Instructions for Animations(2025) BARLETTARO, ELISABETTA; ERIKSSON, EMMA; Chalmers tekniska högskola / Institutionen för fysik; Chalmers University of Technology / Department of Physics; Mehlig, Bernhard; Fröjd, MartinConversational agents are used more and more in customer service, health care, for educational purposes. The fundamental problems of conversational agents are many, including limitations in interpretation of complex queries and lack of emotional intelligence. Despite this, there are distinct advantages of conversational agents, such as efficient data analysis, reduction of operational costs and aid in interactive learning for personalized teaching. The most significant challenge this project aims to undertake is to generate realistic and complex animations in the context of interactive learning with a real-time constraint. The investigation includes how to select machine learning tools and models to aid in the advancement of animation generation, by using both Large-Language Models and purposely constructed Neural Networks. While Large-Language Models are convenient when used in straightforward conditions, Neural Networks are more dependable in an operative application thanks to their consistent format, adaptability and specifically developed purpose.
- PostGenerative Models for Context Dependent Urban Planning(2025) Napieralski, Wojciech; Chalmers tekniska högskola / Institutionen för fysik; Chalmers University of Technology / Department of Physics; Granath, Mats; Radne, AlexanderBuilding footprints represent the total area of coverage of a physical building. Footprints can be used to give an overview of a planned developed area in the early stages of urban planning. This thesis investigates the possibility of training a generative AI model to generate building footprints in a designated area based on surrounding, already existing building footprints. Such a generative model would be a useful tool for architects in the early stages of urban planning, as it would allow the rapid generation of footprint suggestions in an area designated for development. Two image inpainting networks trained for general image reconstruction were fine-tuned using a dataset of building footprints to improve on the task of footprint generation. One of the networks was also modified and trained to be able to accept a desired density of footprints in the generated area as an additional input. Two different masking algorithms were used during training and evaluation: A simple square approach and a more sophisticated algorithm that finds and masks city blocks. FID and LPIPS were used to evaluate and compare the trained models. It was shown that image inpainting networks form a good basis for context dependent footprint generation and that fine-tuning improves performance on this task. Furthermore, it was demonstrated that an image inpainting network can be modified to accept and adhere to density requirements providing a proof-of-concept for other types of user guidance.
- PostMeasurement of Complex Permittivity and Permeability Through a Cavity- Perturbation Method(2015) Rydholm, Tomas; Chalmers tekniska högskola / Institutionen för fysik; Chalmers University of Technology / Department of Physics; Tassin, Philippe; Johansson, JoakimFabrication of antenna and microwave devices demands good knowledge of certain material parameters such as permittivity and permeability, including their corresponding losses. The project presented in this master’s degree thesis aims to develop an experimental setup to measure these properties at microwave frequencies. A cylindrical cavity resonator was designed for a cavity perturbation method. In this method, it is studied how the resonance frequencies shift as a sample is inserted into the cavity resonator. The permittivity and permeability can be determined from this frequency shift and the loss parameters can be obtained from the broadening of the resonance peaks in the spectrum. An alternative method is developed, in which the measurements are compared to data obtained from simulations. A set of curves or contour lines, describing the resonance frequency of a specified mode, is plotted as function of the permittivity and permeability. By performing two measurements where the sample is placed at two different positions in the cavity, the material properties can be determined from the point where the two curves cross each other. We refer to this new method as the “curve-set method”. Simulations indicate that the cavity-perturbation method can be used together with the designed cavity resonator to measure the permittivity and dielectric loss tangent accurately for nonmagnetic materials. However, it seems difficult to measure the permeability and magnetic loss tangent. On the other hand, the curve-set method appears to be a possible way to determine both the permittivity and the permeability, given that the simulations represent experiments accurately.
- PostPrioritizing Structures for Neuroevolution Potentials Improving training data selection for machine-learned interatomic potentials(2024) Strandby, Carl; Chalmers tekniska högskola / Institutionen för fysik; Chalmers University of Technology / Department of Physics; Erhart; Erhart
- PostNeural Network-based study on background for the Dark Leptonic Scalar model at NA64(2024) Zaya, Emil; Chalmers tekniska högskola / Institutionen för fysik; Chalmers University of Technology / Department of Physics; Cederwall, Martin; Crivelli, PaoloThe search for a particle candidate that could explain the origin of dark matter is a central goal in modern astro-particle physics. Numerous experiments employing various measurement strategies are being developed to try and understand this elusive phenomenon. The NA64 experiment situated at the north area of CERN, utilizing the CERN Super Proton Synchrotron (SPS), is an active target experiment aiming to look for signatures like missing energies with hopes of finding signals that correspond to Dark Matter (DM) particles. These dark particles are modelled to explain the physical process of kinetic mixing between the Standard Model (SM) and the hypothesised corresponding Dark Sector (DS). The main purpose of this project is to study the background for a Dark Leptonic Scalar model (DLS) using a highly accurate Monte Carlo simulation for the NA64 experiment. More precisely, the GEANT4 particle simulator was used for the NA64 experiment to simulate the results of the experimental setup used in 2023. The results of this was compared with real data taken in 2023, and a first step was benchmarking the simulation which was done by using dimuon (μμ) events. Furthermore, the simulation results were used as a means of perfecting the methods of event selection. The main source of background for DLS particle φ are μμ production, kaon κ and pion π decay. The main purpose of this thesis is to produce a trained Neural Network (NN) model that can be used for optimizing the selection of events. The background for the DLS φ was simulated and trained on a NN for selecting μμ events as a means of benchmarking the method. The selection of μμ using a trained NN is compared to traditional methods of selection, where an increase of 36 % of the final state events is seen with the NN selected data. A future study could be to simulate the DLS φ particles and train them on a NN to use for event selection. The hopes are to gain a higher signal-to-background ratio and a larger amount of data for the DLS model.