Physics

Title of thesis:&nbsp;Don't Judge a Malware by its Binary&nbsp; Similarity-Based Embeddings for Efficient Malware Analysis
Abstract:<span style="font-size: 12.0pt; font-family: 'Aptos',sans-serif; mso-fareast-font-family: Calibri; mso-fareast-theme-font: minor-latin; mso-bidi-font-family: Aptos; mso-ansi-language: SV; mso-fareast-language: SV; mso-bidi-language: AR-SA;">Cyberattacks are projected to cost the global economy more than $10 trillion annually by 2025. Malware classification still leans on manually engineered binary-level features, a process that is expensive and brittle. We investigated whether Windows PE samples could be mapped into a continuous “behavior space” without running each file in a Sandbox, with the aim of enabling finer-grained distinctions than existing malware family labels provide. Static Endgame Malware BEnchmark for Research (EMBER) feature vectors were paired with dynamic behavior reports from Recorded Future’s Triage Sandbox. A deep metric-learning model with triplet loss (FaceNet-style) was trained to project EMBER vectors into clusters defined by behavioral similarity. The model achieved a weighted F1 score of 0.95 for malware-family classification, proving the model could create valuable embedding spaces for classifying malware by family. Text embeddings for the reports were computed with both BM25 combined with cosine similarity and a transformer encoder. When we projected Sandbox reports into text-embedding space, both BM25 combined with cosine similarity and a transformer encoder revealed finer-grained behavioral structure. In contrast, static EMBER feature vectors showed almost no alignment with dynamic behavior, indicating that they carry insufficient behavioral features. Rich behavioral embeddings can be built directly from Sandbox reports using transformer encoders, scaling more efficiently with corpus size than BM25 combined with cosine similarity.
Examiner: Mats Granath Supervisor: Anders Hansson Opponents: Rikard Radovac, Carolina Rönnewall
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#7241f650-27c9-11ef-a21b-017390dd2bcd" target="_blank" title="Chalmers maps" data-anchor="#7241f650-27c9-11ef-a21b-017390dd2bcd">von Bahr, Soliden 3</a><a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers Maps" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2"></a><a rel="noopener" href="https://maps.chalmers.se/#36a610c1-e9b3-48bd-b01c-9501f9af79fe" target="_blank" title="Chalmers map" data-anchor="#36a610c1-e9b3-48bd-b01c-9501f9af79fe"></a>
Online: <a rel="noopener" href="https://chalmers.zoom.us/j/65417104675" target="_blank" title="Zoom">Zoom</a>

Title of thesis:&nbsp;Tendon Cells from Younger and Older Patients Exhibit Similar Mechanical and Biological Properties in a 3D In Vitro Model
Abstract:&nbsp;The Achilles tendon (AT) is the most frequently injured tendon in the human body, with incidence rates on the rise, particularly among older individuals. The healing capacity decline with age based on functional outcomes post an Achilles tendon rupture. However, the cellular mechanisms behind this remains largely unknown as there is a limited number of studies in humans. The aim of this study was therefore to investigate mechanical and biological differences in new tendon tissue formation between young (n=6, mean age=40.5 years, range=21-53 years) and old (n=7, mean age=70, range=57-76 years) patients with chronic Achilles tendon ruptures. To achieve this, primary human cells extracted from tendon tissue were cultured in a 3D fibrin hydrogel model to form tendon constructs. Pull-to-failure tests were performed to determine the mechanical properties, including the maximum force, stress, stiffness, and Young's modulus. The results show that there is no significant difference between the two groups in any of the measured parameters. These findings imply that neither the structural integrity nor the quality of the tendon construct seems to be affected by increasing age. The biological characteristics were assessed through cell health assays, histology, migration assay, and gene expression analysis. By examining metabolic cell activity and cytotoxicity it could be concluded that there was no change in cell health with age. Histology revealed no differences in cell density, circularity, eccentricity, or proteoglycan content nor in collagen type I and type III ratio. The migration assay showed that no difference could be found in cell motility with age. Analysis of the gene expression showed significant difference in two of the studied genes, namely DCN and S100A4, at one time-point of collection. The compiled results from the biological tests indicate that cells extracted from healing tissue does not seem to behave differently due to increasing age of the patients, except for the gene expression of two of the investigated genes. The mechanical and biological test results indicate that the capacity to repair and form new tissue does not decline with age, which is an encouraging outcome from a patient perspective. To the best of our knowledge, this is the first study to investigate age-related differences in healing Achilles tendons using patient derived primary human cells in an in vitro model, providing novel insights into Achilles tendon healing capacity as a function of age.
Examiner: Julie Gold,&nbsp; Supervisors: Pernilla Eliasson and Christer Thomsen Opponent: &nbsp;Lisa Aronsson&nbsp;
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#96de5091-96fe-4e70-8def-1faa831680ae" target="_blank" title="Chalmers maps" data-anchor="#96de5091-96fe-4e70-8def-1faa831680ae">F Room Raven &amp; Fox</a><a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers map" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2"></a>

Title of thesis:&nbsp;Characterization of solid electrolytes and sodium metal anodes
Abstract:&nbsp;Solid-state sodium batteries are possible candidates for next-generation electrochemical energy storage, with improved sustainability and safety over current lithium-ion technologies. In addition, solid electrolytes could potentially enable the use of sodium metal anodes for increased energy density. However, these electrodes currently have short lifetime due to electro-chemo-mechanically coupled degradation mechanisms, which need to be well understood in order to mitigate the problems and extend the lifetime of solid-state sodium batteries. This thesis investigates a sulfide solid electrolyte, with a particular interest in processes at the interface with sodium metal. The study involves electrochemical characterization techniques and imaging with X-ray tomographic microscopy, in order to gain insights into the processes that limit cell lifetime and efficiency. Results show rapid polarization of the cells, likely caused by chemical degradation between sodium metal and the electrolyte. The reaction products seem to form an interphase layer that fails to passivate the electrode, leading to capacity loss due to the continuous consumption of sodium metal. Protective interlayers could be a promising path of future research in order to improve the interface stability and extend the lifetime of sodium metal electrodes.
Examiner: Aleksandar Matic Supervisors: Elin Dufvenius Esping Opponent: Simon Dovrén
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers map" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2">PJ-salen</a>

Title of thesis: <span style="font-size: 12.0pt; font-family: 'Aptos',sans-serif; mso-fareast-font-family: Calibri; mso-fareast-theme-font: minor-latin; mso-bidi-font-family: Aptos; color: black; mso-ansi-language: SV; mso-fareast-language: SV; mso-bidi-language: AR-SA;">Water-induced phase separation in amorphous solid dispersions studied with SAXS
Abstract:&nbsp;Amorphous solid dispersions (ASDs) are pharmaceutical formulation in which a drug is molecularly dispersed within a carrier matrix. ASDs are designed to address the challenges of drugs with low solubility in their crystalline state. The carrier matrix, typically a polymer, helps stabilizing the drug in its amorphous phase, which allows ASDs to enhance solubility and dissolution rates. This thesis examines ASDs composed of the drug indomethacin and the polymer PVPVA as carrier, with the objective to correlate water-induced phase separation to drug release behaviour. Samples of varying drug concentrations, both in their dry state and after exposure to a neutral buffer, were analysed using small-angle X-ray scattering (SAXS), to understand the phase separation processes on the nanoscale. &nbsp;The mobility of the samples was analysed with differential scanning calorimetry (DSC) and the drug release was measured through dissolution tests. Nanoscale phase separation occurred upon exposure to a neutral buffer, for drug loadings &gt;10%, after absorbing &gt;40% buffer. Higher drug loadings and buffer concentration led to a higher degree of phase separation with distinct interphases. This impacted the dissolution properties, as samples with 5% and 10% drug loading had faster drug release, in comparison to the 20% and 30% samples. &nbsp;These results offer insight into the phase behaviour of ASDs upon dissolution, though additional investigation of ASDs in acidic conditions would enhance the understanding of in vivo phase separation processes.
Examiner: Aleksandar Matic Supervisors: Martina Olsson Opponent: Andreas Erlandsson
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers map" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2">PJ-salen</a>

Title of thesis:&nbsp;<span style="font-size: 11.0pt; font-family: 'Aptos',sans-serif; mso-fareast-font-family: Calibri; mso-fareast-theme-font: minor-latin; mso-bidi-font-family: Aptos; color: #212121; mso-ligatures: standardcontextual; mso-ansi-language: EN-US; mso-fareast-language: SV; mso-bidi-language: AR-SA;" lang="EN-US">Extracting Fission observables in solenoidal spectrometers&nbsp;
Abstract:&nbsp;
Modern subatomic physics experiments often require large accelerator facilities in order to produce and study exotic particles and isotopes. Experiments with highly segmented detectors as well as sophisticated analysis software are nowadays state-of-the-art. This project is motivated by an upcoming experiment in the ISOLDE facility at CERN that aims to study the fission barrier of 230Ac through 229Ac(d,pf) reactions. This is a first step to explore exotic neutron rich isotopes, and the role fission has in the astrophysical rapid neutron capture process (r-process).
This master thesis focuses on development and tests of the analysis software to include newly added detectors before the start of the experiment. By converting simulated data to mimic experimental data it is possible to test the analysis software in advance. Comparing the data output from the software with the same data before the conversion, allows to spot bugs in the software. Thus, the software can be prepared and tested before the experiment, as it removes the need to collect measured data. This means that the software can be used as a diagnostic tool already during the setup phase prior to the experiment.
Examiner: Andreas Heinz Supervisor: Andreas Heinz Opponent: <span style="font-size: 11.0pt; font-family: 'Aptos',sans-serif; mso-fareast-font-family: Calibri; mso-fareast-theme-font: minor-latin; mso-bidi-font-family: Aptos; color: #212121; mso-ligatures: standardcontextual; mso-ansi-language: EN-US; mso-fareast-language: SV; mso-bidi-language: AR-SA;" lang="EN-US">Hazel Ndulue
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers map" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2">PJ-salen</a>

Title of thesis: Creation of a digital twin for simulation of melt pool geometry in metal additive manufacturing
Abstract:&nbsp;Additive manufacturing (AM) is a production technique that allows for the creation of complex geometries and components in near net shape by fusing the material layer by layer. It is a versatile but expensive method and due to this cost, simulating the process is of utmost interest. Since it requires complex multi-physics models at a very low scale, producing simulations of desired accuracy takes a lot of time and computational power. Simpler analytical models have shown promise as they run in a fraction of the time, but at the cost of accuracy. The present work explores the possibility to train an analytical model by adjusting material and process parameters to compensate for the lack of physics. The goal is a digital twin of the more complex model, which generates melt pool geometries comparable to the multi-physics model in a fraction of the time. This was accomplished by minimising a cost function based on the width and depth of a number of isotherms in a simulated melt pool in an electron beam melting AM process. The work demonstrates the feasibility of using a low resolution analytical model with modified material and process parameters as a digital twin to a multi-physics based simulation. Thus demonstrating the feasibility of bridging the gap between computational efficiency and model accuracy. It also establishes a foundation for more cost-effective and scalable simulations in additive manufacturing, advancing the potential for innovation in the field.
Examiner: Mattias Thuvander Supervisor: Erik Granhed Opponent: Filip Bergqvist
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#dc1c9950-1f4d-11f0-a594-0dc76ca9af78" target="_blank" title="Chalmers Maps" data-anchor="#dc1c9950-1f4d-11f0-a594-0dc76ca9af78">F-6004</a>
Online: <a rel="noopener" href="https://chalmers.zoom.us/j/2989797445?omn=64612038301" target="_blank" title="Zoom" data-anchor="?omn=64612038301">Zoom</a>

Title of thesis:&nbsp;Fuel cell catalyst coating quality for reliable ORR activity results
Abstract: Proton exchange membrane fuel cells (PEMFCs) are energy conversion devices that could play a key role in the reduction of greenhouse gas emissions. However, one major problem for PEMFCs is the slow oxygen reduction reaction (ORR), which causes large overpotentials and requires a large amount of expensive platinum catalyst to be sped up. For that reason, evaluation of the ORR activity of catalysts is important, and the rotating disk electrode (RDE) method is commonly used for this purpose. However, there are still problems with achieving reliable results of the ORR activity with RDE, specifically due to the activity’s dependence on the quality of the catalyst coating on the electrode. This thesis investigates how different parameters of a catalyst ink influence the catalyst coating and how to achieve a uniform thin coating. For one catalyst, inks with different alcohol contents and platinum loadings were tested, and for a second catalyst, different alcohol contents and pH levels were tested. The alcohol content did not appear to have a significant impact on the ORR activity or the homogeneity of the thin films. A high alcohol content facilitated the spreading out of the ink thanks to a lower surface tension, but it was also harder to keep it from spreading outside the electrode disk. With platinum loading, there seemed to be a lower limit between 10 and 20 μgPt/cm2 for the catalyst to cover the entire disk homogenously. Below this limit, the mean specific ORR activity was about 20 % lower. At 34.8 μgPt/cm2, the results were about the same as for 20 μgPt/cm2, meaning no upper limit appears to have been reached. For the second catalyst, stability issues were observed, which did not significantly change with the alcohol content. Increasing the pH of the ink by adding KOH improved ink stability, but it was still not possible to produce a completely homogeneous coating. Further investigations on how to achieve a homogeneous coating with this catalyst are therefore needed.
Examiner: Björn Wickman Supervisor: Marika Männikkö Opponents: William Colliander, Victor Enevold&nbsp;
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#96de5091-96fe-4e70-8def-1faa831680ae" target="_blank" title="Chalmers Map" data-anchor="#96de5091-96fe-4e70-8def-1faa831680ae">Raven &amp; Fox</a><a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers map" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2"></a>

Title of thesis:&nbsp;Establishing rheological methodology and analyzing data for PEMFC catalyst ink &nbsp;
Abstract:The shift toward cleaner, low-emission energy solutions are pacing up and hydrogen powered Proton Exchange Membrane Fuel Cells (PEMFCs) are emerging as one of the leading technologies driving this transformation. Catalyst ink is a key component of membrane assembly in the PEMFC, which is composed of platinum based catalyst particles supported on carbon and an ionomer. Composition of catalyst inks, preparation, dispersion, application of catalyst layers and storage largely affect the PEMFC performance. In this study, catalyst inks of varying compositions were tested using the rheometer. Different types of rheology tests were conducted on catalyst inks samples to understand the viscous-elastic responses and their possible impact during electrode coating application. &nbsp;Ionomer to Carbon (I/C) ratio of the catalyst powder was noted to be the main factor in bringing stable viscosity response under shear conditions for the catalyst ink. Solid content follows as a second important factor, however combination of higher I/C ratio with 45-60% content of N-Propanol (NPA) as solvent responds well during the oscillatory sweep tests results. Catalyst ink of such composition shows dominant elastic modulus, higher relaxation time and cross over point (Elastic Modulus G’= Loss Modulus G’’) with stable behavior in linear visco-elastic region (LVER) region. Furthermore, the key parameter Deborah (De) was found to be close to 1 for the best performing catalyst inks, which in rheology is used to characterize the fluidity vs. elasticity of a material under deformation or flow. These findings are also validated and supported by similar results obtained from subscale testing of these CIs membrane assemblies. While these tests and analysis helped make headway in initial understanding of catalyst inks used at PowerCell Group, more research work is needed to understand morphology correlation and response of particle interactions at the nano scale using advanced techniques like cryo-TEM, SEM and SAXS.
Examiner: Björn Wickman Supervisor: Marcus Liljenberg Opponents: Lorensu Hewa Welle Kamkanamge Sujith Priyantha
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#96de5091-96fe-4e70-8def-1faa831680ae" target="_blank" title="Chalmers Map" data-anchor="#96de5091-96fe-4e70-8def-1faa831680ae">Raven &amp; Fox</a>

Title of thesis:&nbsp;Uncertainty-Aware Weather-Related Road Surface Condition Classification
Abstract: Autonomous vehicles rely on collaborative systems for decision-making during operation, one of which is machine learning algorithms. Using cameras and sensors mounted on the car, these algorithms can analyse captured data to interpret its surroundings, enabling it to make informed and safe driving decisions. The algorithms are trained on data gathered in the field. In real-time use, samples that have never been seen before, referred to as Out-Of-Distribution (OOD) samples, can occur, which the algorithm then confidently uses to make driving decisions. In a worst-case scenario, this can lead to vehicle failure, resulting in injuries to passengers and/or property. The primary objective of this thesis is to implement and analyse methods for detecting out-of-distribution (OOD) samples in a road surface condition classifier. The machine learning algorithm was a convolutional neural network, trained to detect dry, wet and snowy road conditions. The methods used to detect OOD samples were Maximum Softmax Probability, Energy-Based OOD detection, Outlier Exposure, Virtual Outlier Synthesis, Rectified Activations, Deep Nearest Neighbour, and Virtual Logit Matching. The datasets used for evaluation of the methods were three datasets, cars (car in front of vehicle), slush (slushy road condition), glare (sun causes glare on the road surface), constructed from the same database as the data the model was trained on and are considered as near OOD data along side with two publicly available datasets, CIFAR10 and Textures, are considered far OOD data. The methods that performed the best were Deep Nearest Neighbour and Virtual Logit Matching when evaluated on CIFAR10 and Textures datasets, achieving near-perfect results. When tested on the cars dataset, Virtual Logit Matching exhibited a notable deviation in performance compared to the other methods, although it still performed poorly. In general, except as previously mentioned, the detectors did not outperform the two baseline detectors, Energy-Based OOD detection and Maximum Softmax Probability.
Examiner: Daniel Midtvedt Supervisor:&nbsp; Opponents: William Nilsen, Gustav Wikman
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#eece58ce-9a1e-4ab1-af3d-d4fc2cb1126c" target="_blank" title="Chalmers maps" data-anchor="#eece58ce-9a1e-4ab1-af3d-d4fc2cb1126c">Nexus</a>

Title of thesis:&nbsp;A Method for Learning Automated Driving Systems and Data Consistency Analysis
Abstract:&nbsp;In this report a method for developing a deep learning network to predict signals in automated driving systems is presented. The goal is to incorporate the network in a larger data consistency tool aimed to find faults in vehicle testing data. To this end, a neural network was constructed and trained on simulations of vehicle dynamics with the purpose of predicting the dynamics of a real test vehicle. A parallel goal was to compare the properties of the network to a simulation software. The project was made in collaboration with Zenseact, an AI and software company developing the complete software stack for automated driving and advanced driver-assistance systems. The network did predictions replicating a simulation software's Controllers, Vehicle model and the two combined. To evaluate which type of training data gave the best result, three datasets with different driving schemes were generated by the simulation software. On a test set constructed from 21 log files from real test drives, the network got the best results predicting the Controllers and Vehicle model combined. In this case, for only predicting a vehicle's lateral motion, a training set consisting of constant longitudinal velocity generated the best results. If both lateral and longitudinal are desired, a dataset with sinusoidally changing acceleration performed best overall. The other purpose was to compare the time aspects of the network to the simulation software. This was done by firstly evaluating &nbsp;the time to simulate twelve 65 second scenarios, resulting in 8.5 to 9 minutes over all tests. Secondly, training the network on this data consistently took 61 to 62 seconds in total. Thirdly, the prediction time on the 84495 batches was between 6 to 7 seconds, or an average of 75.9 microseconds per batch. It was concluded that, apart for the intended goal of being a core in a data consistency tool, the main use case for the network could be in applications where fast data generation times are crucial, for example in near real time predictions during test drives.
Examiner: Mohsen Mirkhalaf Supervisor: Mohsen Mirkhalaf Opponents:&nbsp;Ludvig Tvingby, Maria Björkman
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#7241f650-27c9-11ef-a21b-017390dd2bcd" target="_blank" title="Chalmers maps" data-anchor="#7241f650-27c9-11ef-a21b-017390dd2bcd">Eva von Bahr</a>

Title of thesis: Advancing Sub-Visible Particle Characterization in Inhaled Biologics
Subtitle:&nbsp;Micro-Flow Imaging for Particle Analysis and Sub-Population Differentiation
Abstract: Protein aggregation poses a significant immunogenicity risk in inhaled biologics, where aggregated proteins may trigger adverse immune responses. Conventional methods, such as Light Obscuration, are used to quantify sub-visible particles, but often face challenges in detecting small and translucent particles.
This thesis explores how continuous flow microscopy, specifically Micro-Flow Imaging, can be used to differentiate particle sub-populations by analyzing morphological and optical data. Drug products and excipients are examined to identify and quantify these sub-populations, their composition, aiming to enhance the understanding of the particulate environment in the formulations.
The results reveal that different formulations exhibit particles with distinct characteristics. Particle distributions have been established, resulting in unique profiles that could potentially be used to identify the presence of specific formulations in unknown samples. However, a deeper understanding of the interactions and complexities among various substances in the formulations is necessary to enhance the accuracy of the identification test.
Examiner: Jan Swenson Supervisors: <span style="font-size: 12.0pt; font-family: 'Aptos',sans-serif; mso-fareast-font-family: 'Times New Roman'; mso-bidi-font-family: Aptos; mso-ansi-language: SV; mso-fareast-language: SV; mso-bidi-language: AR-SA;">Noemi Gaglianone, Gerard Masdeu Opponent: <span style="font-size: 12.0pt; font-family: 'Aptos',sans-serif; mso-fareast-font-family: 'Times New Roman'; mso-bidi-font-family: Aptos; mso-ansi-language: SV; mso-fareast-language: SV; mso-bidi-language: AR-SA;">Olivia Jonsson
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers map" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2">PJ-salen</a>

Title of thesis:&nbsp;Microscopy for pH mapping in nanofluidic devices
Abstract: Nanotechnology, along with related fields such as nanofluidics, is an area of science with much potential. Research into material properties at the nano-scale is driving innovation across various fields, such as the development of new drugs in medicine, and improvements to solar cells in the development of sustainable energy.&nbsp; This thesis examines the possibility of creating and measuring pH gradients in nanofluidic channels, as a novel method to study nanoparticles. Such a method could potentially be used to study nanoparticles subjected to different pH environments. The presented process relies on darkfield microscopy and a pair of double Amici prisms. When observing fluorescent particles or dyes in the nanochannels, the fluorescence is angularly dispersed by the prisms. This enables measurements that capture spatial distribution of particles along the nanochannels, and corresponding spectral information dispersed perpendicularly.&nbsp; The experiments confirm that it is possible both to generate a pH gradient and to measure local pH values within the nanochannels. The presented results are largely based on the use of the pH-sensitive fluorescent dye SNARF-1, and a foundation for future study with FRET nanoparticle probes is also provided. 
Examiner: Christoph Langhammer Supervisor: Bohdan Yeroshenko&nbsp; Opponents: Jacob Billvén and Nikhil Saraswat
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#96de5091-96fe-4e70-8def-1faa831680ae" target="_blank" title="Chalmers Maps" data-anchor="#96de5091-96fe-4e70-8def-1faa831680ae">Raven &amp; Fox</a><a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers Maps" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2"></a><a rel="noopener" href="https://maps.chalmers.se/#36a610c1-e9b3-48bd-b01c-9501f9af79fe" target="_blank" title="Chalmers map" data-anchor="#36a610c1-e9b3-48bd-b01c-9501f9af79fe"></a>
Online: <a rel="noopener" href="https://chalmers.zoom.us/j/8250031978" target="_blank" title="Zoom">Zoom</a>

Title of thesis:&nbsp;Sequential Graph-Based Decoding of the Surface Code using a Hybrid Graph and Recurrent Neural Network Model
Abstract:In order to achieve reliable quantum computation with noisy qubits, quantum error correction (QEC) is necessary. Quantum error-correcting codes mitigate the inherent noise in quantum systems by distributing the logical state over several qubits, thereby introducing redundancy. One such promising code is the surface code. It encodes the logical qubit using a two-dimensional lattice of physical data qubits and ancilla qubits. By taking and decoding measurements on the ancilla qubits of the surface code, one can deduce whether a logical bit- or phase-flip has occurred. However, this is a complex and potentially time-consuming task. Multiple decoding algorithms exist, such as the classical minimum weight perfect matching (MWPM) decoder. In recent years, data-driven algorithms have been shown to decode the surface code with a high degree of accuracy. In this thesis, we present a machine learning approach to decoding the surface code using a combination of graph neural networks (GNN) and recurrent neural networks (RNN). Specifically, graph representations are constructed over a short, sliding time window of syndrome measurement data. Each representation is processed by a GNN and its output is used as a learned high-dimensional embedding for an RNN. This enables continuous decoding of measurement patterns over longer time series. While the decoder is trained on relatively short syndromes, it is able to generalize for new data and longer syndromes, outperforming the classical MWPM algorithm across both short and long time series. This work opens up a new approach to reliable and potentially fast decoding of QEC codes.
Examiner: Mats Granath Supervisor: Mats Granath Opponent: Tobias Wallström&nbsp;
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#37d8de3e-1806-4420-92e7-b0b773f4fc88" target="_blank" title="Chalmers Maps" data-anchor="#37d8de3e-1806-4420-92e7-b0b773f4fc88">FL41</a><a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers Maps" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2"></a><a rel="noopener" href="https://maps.chalmers.se/#36a610c1-e9b3-48bd-b01c-9501f9af79fe" target="_blank" title="Chalmers map" data-anchor="#36a610c1-e9b3-48bd-b01c-9501f9af79fe"></a>
Online: <a rel="noopener" href="https://chalmers.zoom.us/j/62345043051" target="_blank" title="Zoom">Zoom</a>

Title of thesis: Pd-Pt nanoparticles for plasmonic sensing of hydrogen in humid environments
Abstract: Using hydrogen as energy storage and as fuel in fuel cells are attractive ideas that have been investigated for decades. One of the big problems with using hydrogen for this purpose is that hydrogen, due to its small size, can easily escape, and only low concentrations are required for it to be explosive. To help detect leaks and prevent accidents, a fast and reliable method for the detection of hydrogen gas is necessary. One of the most commonly used materials for this purpose is Pd. The problem with Pd is that it is vulnerable to deactivation from several different compounds, one of which being water. This thesis aimed to investigate a solution to this by incorporating Pt into the system. Several different compositions were fabricated (Pd70Pt30, Pd85Pt15, Pd90Pt10 and Pd95Pt5), and tested both as a bi-layered&nbsp; structure and alloyed together, in a humidity reactor. All systems show potential as hydrogen sensors in humid environments. However, how to clearly translate the complex signal output into readings of H2 concentration remains unclear.&nbsp;
 Examiner: Christoph Langhammer Supervisor: Carl Andersson, Athanasios Theodoridis, Sara Nilsson&nbsp;&nbsp; Opponent: Hugo Mårtensson&nbsp;
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#96de5091-96fe-4e70-8def-1faa831680ae" target="_blank" title="Chalmers Maps" data-anchor="#96de5091-96fe-4e70-8def-1faa831680ae">Raven &amp; Fox</a><a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers Maps" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2"></a><a rel="noopener" href="https://maps.chalmers.se/#36a610c1-e9b3-48bd-b01c-9501f9af79fe" target="_blank" title="Chalmers map" data-anchor="#36a610c1-e9b3-48bd-b01c-9501f9af79fe"></a>

Title of thesis:&nbsp;Optimal Control of a SNAIL-resonator
Abstract: In recent experiments done at Chalmers, universal control of a bosonic mode was demonstrated for a superconducting cavity terminated by a Superconducting Nonlinear Asymmetric Inductive eLement (SNAIL), while simultaneously cancelling Kerr effects by applying an external magnetic flux to the SNAIL. The universal control was demonstrated by generating a cubic phase state; however, generating other interesting non-Gaussian states remains a non-trivial issue. The aim of this work is to perform optimal control on the SNAIL-resonator. Due to the large Hilbert space of the system and its fast time dependence, an effective Hamiltonian is derived from the previously known full Hamiltonian in the regime of resonant driving, which significantly speeds up the optimization process. The drives of the system are then optimized to reach a target cubic phase state, both by simply tuning the amplitudes and phases of the pulses used in the Chalmers experiment and by pulse shaping optimization. Then, optimal control is used to generate two-photon cat states. The results are analyzed in terms of the fidelity to target states and its susceptibility to circuit parameter deviations. Finally, the potential for using optimal control to find experimentally viable methods of generating non-Gaussian states is discussed.
 Examiner: Göran Johansson Supervisor: Tahereh Abad &nbsp; Opponent: Joel Sandås
Location of presentation:<a rel="noopener" href="https://maps.chalmers.se/#167d1b5f-8ec8-4e86-b511-0b3c50e3c06a" target="_blank" title="Chalmers Maps" data-anchor="#167d1b5f-8ec8-4e86-b511-0b3c50e3c06a">Luftbryggan</a>
Online: <a rel="noopener" href="https://chalmers.zoom.us/j/5525350795" target="_blank" title="Zoom">Zoom</a>

Title of thesis:&nbsp;Towards Event Sequence Foundation Models - Leveraging Temporal Point Process Transformers for Power Grid Fault Prediction
Abstract:&nbsp;Interruptions in electric power systems cause significant disruptions and economic losses. The growing availability of fault data from electric power grids, combined with advances in machine learning, particularly deep learning, is opening up new opportunities for fault prediction. Transformer architectures are especially well-suited to model sequential data, like the temporal sequence of historical disturbances in the power grid. This thesis proposes a foundation model approach to fault prediction, in which a transformer-based model is pretrained on large-scale event sequence datasets from diverse domains. We then fine-tune the model on downstream fault prediction tasks in a continual learning manner, using historical fault data segmented into 3-day sequences. While an equivalent specialized fault model outperforms this smaller foundation model by achieving 21.8% higher average precision, fine-tuning on multiple proxy tasks - common in foundation model training - significantly improved fault prediction performance of both models. These results suggest that while foundation models for event sequences are still emerging, the idea of proxy task fine-tuning is already beneficial to existing models. Notably, the foundation model exhibited less forgetting and higher forward transfer than the specialized model, indicating superior stability and plasticity. Scaling up foundation models remains a promising path towards reliable and scalable fault prediction systems for the power grid.
Examiner: Mohsen Mirkhalaf Supervisors:&nbsp;Viktor Olsson, Jakob Lindqvist&nbsp; Opponents: Anton Hedén and Anthon Odengard
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#7241f650-27c9-11ef-a21b-017390dd2bcd" target="_blank" title="Chalmers maps" data-anchor="#7241f650-27c9-11ef-a21b-017390dd2bcd">Eva von Bahr</a>

Title of thesis:&nbsp;Optimizing Power Cable Routing using AI&nbsp;
Abstract: Planning underground transmission cable routes is traditionally a labor-intensive and expert-driven task involving multiple, often conflicting, objectives. To meet increasing energy demands, scalable and intelligent methods are needed to enhance efficiency while accounting for diverse spatial constraints. This study investigates how AI-driven optimization algorithms can improve underground cable routing by integrating them with Geographic Information Systems (GIS) data. Key geospatial constraints, such as soil type, land use, and vegetation, are identified and systematically encoded into a unified cost surface using the Analytic Hierarchy Process (AHP). Three algorithmic approaches are implemented and evaluated: Dijkstra’s algorithm (including single- and bi-objective variants), the A* algorithm with weighted heuristics, and a bi-objective Ant Colony Optimization (ACO) algorithm. These are assessed across synthetic and real-world environments, including the planned Gotland link in Sweden. While each model exhibits distinct strengths and limitations, all produce competitive and adaptable routes compared to manual planning, effectively balancing traversal cost and distance, and identifying Pareto-optimal solutions that highlight strategically important areas. . This suggests that the integration of AI and GIS has strong potential to automate and improve cable routing processes. At the same time, the study underscores the importance of tuning both model architecture and environmental representation to maximize real-world applicability, revealing that model performance is not only algorithm-dependent but also highly sensitive to the spatial structure and scaling of the input data.
Examiner: Mats Granath Supervisor: Josef Abriren&nbsp; Opponent: Pontus Kjellström (MPCAS), Niklas Lundblad (MPCAS)&nbsp;
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#37d8de3e-1806-4420-92e7-b0b773f4fc88" target="_blank" title="Chalmers Maps" data-anchor="#37d8de3e-1806-4420-92e7-b0b773f4fc88">FL41</a><a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers Maps" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2"></a><a rel="noopener" href="https://maps.chalmers.se/#36a610c1-e9b3-48bd-b01c-9501f9af79fe" target="_blank" title="Chalmers map" data-anchor="#36a610c1-e9b3-48bd-b01c-9501f9af79fe"></a>
Online: <a rel="noopener" href="https://chalmers.zoom.us/j/63341168344" target="_blank" title="Zoom">Zoom</a>

Title of thesis:&nbsp;Advancing emergency operations using sensor fusion with AI features&nbsp;
Abstract: In recent years, society has experienced rapid technological advancements across all fields, pushing the boundaries of what was once thought possible. Emergency services, such as fire departments, have largely been left behind. Operating in hazardous, smoke-filled environments places firefighters at extreme risk, where no compromises can be made on the tools and technologies that support them. This project addresses that gap by exploring the development of a Multi-Environment Camera (MEV-Cam) system designed to aid firefighters during rescue operations in visually impaired environments. The study focuses on the fusion of LiDAR and thermal imaging, investigating how the sensors can complement each other weaknesses to provide more reliable scene understanding.&nbsp; To address these challenges, four tailored datasets were created to simulate the intended application scenario, each containing synchronized data from the sensors. Fusion based models were developed to enable the sensors to complement each other’s weaknesses. For scene understanding, deep learning models were investigated to both LiDAR and thermal data, using architectures based on YOLO and PointNet. Additionally, pose estimation was explored using monocular visual odometry. A major focus of the project was on enhancing the raw collected data to ensure the sensors could be effectively used in this system. In parallel, a hardware prototype was developed to enable efficient data collection in real world environments. Results show that airborne particles significantly degrade LiDAR performance, while thermal sensors remain relatively unaffected. However, sensor fusion can compensate for these limitations. Deep learning models demonstrated the ability to accurately interpret scene structure under degraded conditions. After effective data enhancement, the MEV-Cam system showed improved performance across its various modules. While these results highlight the promise of the technology, they also reveal current limitations and suggest several innovative directions for future work. This project marks the beginning of a new project to support life saving operations, by utilizing the latest sensor technology and AI.
Examiner: Magnus Karlsteen Supervisors: Magnus Karlsteen Opponents: Simon Nygren, Gustav Mosséen&nbsp;
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers map" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2">PJ-salen</a>

Title of thesis:&nbsp;Deep Leakage in Federated Learning: Understanding Privacy Vulnerabilities
Abstract:&nbsp;
Deep Leakage attacks in Federated Learning have traditionally relied on FedSGD, yet real-world deployments commonly adopt FedAVG due to its reduced communication overhead. This study investigates the feasibility and limitations of executing DL attacks within a FedAVG setting. A custom FL framework was developed to support FedAVG and state-of-the-art DL techniques to operate on shared model weights instead of gradients. Experiments conducted using the CIFAR-10 dataset revealed that while DL attacks are possible under FedAVG, their success diminishes as local training (batch size and epochs) increases, due to degraded gradient approximations. Additionally, model initialization strategies, dataset size, and image resolution significantly impact reconstruction quality. These findings highlight critical trade-offs between privacy and performance in FL systems, emphasizing the need for cautious design choices in real-world applications.
Examiner: Kristian Gustafsson Supervisor:&nbsp;Håkan Warston Opponent: Albin Steen 
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#7241f650-27c9-11ef-a21b-017390dd2bcd" target="_blank" title="Chalmers maps" data-anchor="#7241f650-27c9-11ef-a21b-017390dd2bcd">Eva von Bahr</a>

Title of thesis:&nbsp;Universal interface growth in substrate etching processes
Abstract: Etching is a process that has been used both for art and in industry for hundreds of years. Although well understood on larger spatial scales, when acting on a small-scale two-dimensional lattice other effects seem to take place. This project has aimed at using a Monte Carlo simulation to study the effects on a honeycomb lattice and study how the system evolves over time depending on the underlying crystal structure and different relative decay rates for solitary atoms.
The implementation shows that during short times the crystal structure very much plays a role in the shapes that develop creating shapes likening a hexagon with slightly rounded corners. As time passes, the holes created become more circular and by increasing reactivity for the atoms that are more solitary, the shape becomes nearly completely circular at large times. This is supported by several metrics such as area error measurement as well as radial error measurements. Moreover, the radius grows linearly with time, indicating that as more atoms become exposed, the more reactions occur, but as the interface grows larger the more atoms need to be annihilated to increase the radius. Longer time frames need to be studied further for more reactive conditions, to see where the hexagonal error will plateau.
Examiner: Johannes Hofmann Supervisors: Johannes Hofmann Opponents: Ole Fjeldså and Gustaf Jonasson Johansson
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers map" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2">PJ-salen</a>

Title of thesis:&nbsp;Pushing the Boundary: Simulating Interface Growth in Bacterial Colonies
Abstract: Bacteria are life forms that exist all over Earth, thriving in both the human body and in nature. Creating models to simulate the complex behaviors that may arise in bacterial colonies is a valuable asset in microbiology and biomedicine, for example these models can be used to study antibiotic resistance. This thesis aims to develop a computational model to simulate bacterial colony growth. The model is based on a lattice growth model and is grown with Eden dynamics. In addition there is a nutrient field representing nutrient concentrations. This nutrient field adds a more realistic complement to the Eden model and can model environmental constraints that impact how and where the colony can grow. The colony grows as bacterial cells consume nutrients and divide, pushing the colony's boundary outward. The colony is more likely to expand into areas with higher nutrient availability. The parameter determining the growth rate and colony longevity is the nutrient consumption coefficient, which controls at what rate nutrients are consumed during growth. The results show how this coefficient affects the colony's size, radial expansion, and roughness of the boundary. Additionally, a stability analysis is conducted to determine the threshold value of the nutrient consumption coefficient required for the colony to sustain itself.
Examiner: Johannes Hofmann Supervisor: Johannes Hofmann Opponent: Rakel Hellberg
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#7241f650-27c9-11ef-a21b-017390dd2bcd" target="_blank" title="Chalmers maps" data-anchor="#7241f650-27c9-11ef-a21b-017390dd2bcd">Eva von Bahr</a>

Title of thesis:&nbsp;Tracking meta-dynamic recrystallization in a Ni-based superalloy using time-resolved synchrotron X-ray diffraction
Abstract:&nbsp;
The drive towards more sustainable aviation has gained traction as the global aviation industry expands alongside increasing demands to reduce carbon emissions. This drive has pushed research in materials science aimed at developing alloys capable of withstanding high operating temperatures while maintaining metallurgical stability. The nickel-based superalloy Haynes 282 has emerged as a promising material for use in aero-engine applications. Recrystallization plays a critical role in determining the final microstructure and, consequently, the mechanical properties of metallic materials. This project aims to develop computational methods for studying the dynamics of meta-dynamic recrystallization in Haynes 282, to gain a deeper understanding of how the recrystallization process evolves. The superalloy samples were first deformed at elevated temperatures. Right after the deformation stopped, in-situ synchrotron X-ray diffraction (XRD) measurements were performed while maintaining the constant deformation temperature. Diffraction patterns were recorded as a function of time. A method was developed to separate the signal into contributions from the deformed and recrystallized grains. This enabled the calculation of the recrystallized fraction, and the tracking of recrystallized grain evolution over time. These analyses were applied to samples deformed at different temperatures. For the sample deformed at 1100°C , the recrystallized fraction increased rapidly, from an initial average of approximately 5% to 90% within 120 seconds. Higher deformation temperatures resulted in faster recrystallization kinetics and higher final fractions. The progression of new grains and their size distribution was also studied, showing an increase in both the number of grains and their average size over time.
The results were compared to previous experimental measurements of Haynes 282, where electron backscatter diffraction was used to quantify the recrystallization and grain growth. The findings of this project align with conclusions previously drawn. The methodology developed in this work demonstrates that time-resolved synchrotron XRD data can be used to effectively study recrystallization, enabling high temporal resolution and potentially greater sensitivity to small grains.
Examiner: Magnus Hörnqvist Colliander Supervisors: Magnus Hörnqvist Colliander Opponent: Jonathan Borsander&nbsp;
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers map" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2">PJ-salen</a>

Title of thesis:&nbsp;Recurrent Neural Networks for Lagrangian Tracking of Microorganisms
Abstract: <span style="font-size: 12.0pt; font-family: 'Aptos',sans-serif; mso-fareast-font-family: 'Times New Roman'; mso-bidi-font-family: Aptos; color: black; mso-ansi-language: SV; mso-fareast-language: SV; mso-bidi-language: AR-SA;">Research in microbiology is crucial for development of antibiotics, vaccines and other medicines that cure diseases and prevent spread of viral infections. One method of studying microorganisms is Lagrangian tracking, where the movement of single microorganisms is tracked over long periods of time, which is important when studying for example chemotaxis. Lagrangian tracking has previously been implemented using deep learning, showing promising results. However, the model, a convolutional neural network (CNN), struggled with handling overlapping microorganisms, which resulted in failure of whole experiments when the model switched which microorganism was currently being tracked. This thesis aimed to create a model for Lagrangian tracking of microorganisms that could accurately track over long periods of time as well as handle overlapping microorganisms. The method included simulation of fluorescence microscopic data as well as design, training and evaluation of a recurrent neural network (RNN) using the simulated data. The results showed that the RNN gave lower distribution of errors and was able to handle overlapping microorganisms better compared to the CNN implemented for benchmarking. An analysis of the influence of features of the microorganisms indicated that the tracking was harder when surrounding microorganisms were close to the focal plane or had higher intensity compared to the microorganism that was currently tracked. Although testing the RNN in an experimental setup remains, the results suggest that replacing a CNN with an RNN can improve the accuracy of the Lagrangian tracking and to greater extent avoid losing the microorganism during an overlap. In turn, improving the accuracy of the Lagrangian tracking contributes to the possibility of tracking single microorganisms over long periods of time and gain more knowledge about for example chemotaxis.
Examiner: Giovanni Volpe Supervisors: Daniel Midtvedt Opponent: Erik Broback&nbsp;
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#37d8de3e-1806-4420-92e7-b0b773f4fc88" target="_blank" title="Chalmers maps" data-anchor="#37d8de3e-1806-4420-92e7-b0b773f4fc88">FL-41</a>

Title of thesis: An Investigation of using Simulated Data for Machine Learning
Abstract: 
Monitoring meiofauna is an effective way of assessing the effects of pollution in an environment, but the current method of manually extracting data from samples is time consuming. If one instead uses a machine learning model for image recognition and segmentation a lot of this manual work can be automated. Furthermore, if one can use simulated data for training the model then new models can be created more easily. This work therefore investigates the possibility of simulating two groups of meiofauna, Nematodes and Nodosarias, and finds promising results and methodology. The findings also highlight the fact that simulations are not necessarily easily created and require new effort for every new group that should be simulated.
Examiner: Giovanni Volpe Supervisor: Daniela Pérez Guerrero Opponent: Mathilda Gustafsson
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#37d8de3e-1806-4420-92e7-b0b773f4fc88" target="_blank" title="Chalmers maps" data-anchor="#37d8de3e-1806-4420-92e7-b0b773f4fc88">FL-41</a>

Title of thesis: Design and Evaluation of a RAG Chatbot in an Industrial Setting
Abstract: The rapid advancements in large language models (LLM) have sparked industrial interest in leveraging generative AI for information access and decision support. However, the limitations of standalone LLMs, such as hallucinations and a lack of domain-specific or up-to-date knowledge, pose significant barriers in enterprise settings. Retrieval-Augmented Generation (RAG) architectures offer a promising solution by integrating external document retrieval into the generative process. This thesis presents the design, implementation, and evaluation of a prototype RAG chatbot developed in collaboration with Volvo Group Trucks Technology. The chatbot enables users to interact conversationally with internal data comprising multiple modalities, including text and images. The system architecture was built from the ground up, focusing on document parsing, vector-based retrieval using FAISS, and multimodal integration. Evaluation covered functionality, generation quality, and usability, with positive feedback from end users supporting the system’s practical viability. This work contributes to applied research on RAG systems in industrial environments, highlighting challenges related to multimodal data handling, data security, and ethical considerations such as GDPR compliance. It serves both as a proof of concept for internal use and as a blueprint for broader AI adoption within the organization and similar industrial contexts.&nbsp;
Examiner: Giovanni Volpe Supervisor: Martin Claesson Opponent: Viktor Örnbratt
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#37d8de3e-1806-4420-92e7-b0b773f4fc88" target="_blank" title="Chalmers maps" data-anchor="#37d8de3e-1806-4420-92e7-b0b773f4fc88"></a><a rel="noopener" href="https://maps.chalmers.se/#eece58ce-9a1e-4ab1-af3d-d4fc2cb1126c" target="_blank" title="Chalmers maps" data-anchor="#eece58ce-9a1e-4ab1-af3d-d4fc2cb1126c">Nexus</a>

Title of thesis: Applying Conformal Prediction for LLM Multi-Label Text Classification
Abstract: Large language models are increasingly deployed in real-world natural language processing systems to perform multi-label classification tasks, such as identifying multiple forms of toxicity in online content. However, most models output raw probabilities without an exact way to quantify uncertainty, increasing the risk of over-prediction in high-stakes applications. In this thesis, inductive conformal prediction is integrated with the binary relevance approach to produce statistically valid prediction sets, with label-wise guarantees. Using a modified Wikipedia toxic comments dataset, this framework is evaluated across varying significance levels ($\epsilon$), incorporating calibration-set-aware thresholds to address label imbalances. The results showcase that a binary relevance-based conformal prediction maintains valid marginal coverage while enabling flexible control over prediction set size (efficiency). Even in the presence of rare labels, the framework provides practical uncertainty estimates and where the prediction can be abstained in uncertain cases via empty sets. These findings support the feasibility of binary relevance inductive conformal prediction-based uncertainty calibration for robust, scalable and interpretable automation in multi-label natural language processing systems.&nbsp;
Examiner: Giovanni Volpe Supervisor: Johan Hallberg Szabadváry Opponent: Rasmus Nilsson, Oscar Rosman
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#37d8de3e-1806-4420-92e7-b0b773f4fc88" target="_blank" title="Chalmers maps" data-anchor="#37d8de3e-1806-4420-92e7-b0b773f4fc88"></a><a rel="noopener" href="https://maps.chalmers.se/#eece58ce-9a1e-4ab1-af3d-d4fc2cb1126c" target="_blank" title="Chalmers maps" data-anchor="#eece58ce-9a1e-4ab1-af3d-d4fc2cb1126c">Nexus</a>
Online: <a rel="noopener" href="https://chalmers.zoom.us/j/64082323944" target="_blank" title="Zoom">Zoom</a>

Title of thesis: Inverse Design of FSS Using Variational Autoencoder
Abstract:&nbsp;Frequency selective surfaces (FSS) are critical for modern electromagnetic filtering, but their design traditionally relies on costly trial-and-error simulations. We present a generative machine learning framework for the inverse design of FSS unit cell patterns. By utilizing a conditional variational autoencoder (cVAE), the proposed method directly maps desired electromagnetic scattering parameters (S-parameters) to FSS patterns, thereby circumventing the traditional energy- and time-expensive trial-and-error approach inherent in FSS design. A dataset of 10,000 simulated (pattern, S-parameter) samples was generated using Ansys HFSS over the 2 to 8 GHz frequency range to train both a surrogate neural network, which accurately predicts the S-parameters from a given pattern, and a cVAE-based generator that synthesizes novel pattern designs conditioned on target frequency responses.
The integrated framework employs a gradient-based optimization strategy in the latent space to minimize the deviation between the predicted and desired S-parameter responses, with particular emphasis on preserving the frequency valley. Benchmarking on the test dataset demonstrates that the surrogate model achieves low mean absolute errors from 0.5 dB at 2 GHz to 1.9 dB at 8 GHz, while the optimization loop refines designs to yield deviations as low as 0.0-0.2 GHz at the frequency valley for half of the samples. These results underscore the promising potential of generative machine learning for rapid FSS inverse design.
Examiner: Giovanni Volpe Supervisors: Giovanni Volpe, Ahmed Gouda Opponent: Abubakar Abukar
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#37d8de3e-1806-4420-92e7-b0b773f4fc88" target="_blank" title="Chalmers maps" data-anchor="#37d8de3e-1806-4420-92e7-b0b773f4fc88">FL-41</a>

Title of thesis:&nbsp;Quantification of C-Reactive Protein Using a Miniaturized SPR Platform
Abstract: C-reactive protein (CRP) is a key biomarker for inflammation and infection and is therefore widely used in clinical diagnostics. The growing number of patients being treated at home has increased the demand for portable, low-cost diagnostic tools that support remote health monitoring. However, current CRP testing methods often rely on centralized laboratory infrastructure or bulky and expensive point-of-care (PoC) systems, limiting its applicability in decentralized settings. This thesis explores the adaptation of a miniaturized surface plasmon resonance (SPR) platform for CRP quantification, with the goal of enabling accessible diagnostics in these types of applications. The platform employs a compact optical setup using metagrating-integrated vertical-cavity surface-emitting lasers (VCSELs) to achieve angle-resolved SPR sensing without bulky optics or mechanical components. The system was integrated with a microfluidic system to deliver analyte solutions across the sensing surface in a controlled manner. Its bulk performance was characterized by using glycerol standards, demonstrating a sensitivity of 97.72°/RIU and a resolution of 5.11 × 10-6 RIU, which indicates a performance sufficient for many biosensing applications. A surface functionalization protocol for anti-CRP antibody immobilization was designed and implemented using self-assembled monolayers and EDC/NHS chemistry. However, its performance could not be evaluated due to a hardware malfunction in the laser chip that prevented subsequent functionalization and CRP binding experiments. Despite this challenge, the work provides a foundation for future development and proposes strategies for optimizing and validating the assay in future work, with the long-term goal of enabling CRP detection in remote or home-based settings using compact, low-cost instrumentation.
Examiner: Andreas Dahlin Supervisors: Hana Jungová Opponent: Maria Virginia Ramos Marca&nbsp;
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers map" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2">PJ-salen</a>

Title of thesis:&nbsp;Forecasting Production Plan on Limited Dataset
Abstract:&nbsp;Forecasting production plans with limited data is a significant challenge, especially for smaller firms or new products with short production histories. This thesis aims to predict actual production outcomes based on historical data and to understand the relationships between and connect planned and executed production volumes. By exploring various forecasting approaches, including ARIMA and Long Short-Term Memory (LSTM) networks, the study focuses on methods designed to perform well with smaller datasets. The research employs a hierarchical model architecture that decomposes the forecasting task into three components: production forecasting, quarterly mapping, and unfolding to a monthly plan estimate. The models are evaluated against baseline strategies, including naive predictions and basic LSTM and ARIMA models. Results show that the proposed hierarchical models outperform baseline models, capturing the general behavior of the data more effectively. The deep learning-based model excels at capturing extremes in the time series, while the regression-based model provides stable and accurate forecasts. However, the models struggle with highly erratic production plan patterns, indicating the need for further refinement. This thesis contributes to more robust and scalable production planning solutions for data-constrained environments, offering valuable insights for both academic research and practical applications.
Examiner: Giovanni Volpe Supervisors: Di Xue Opponent: Oliver De Rosa &amp; Johan Larsson
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#37d8de3e-1806-4420-92e7-b0b773f4fc88" target="_blank" title="Chalmers maps" data-anchor="#37d8de3e-1806-4420-92e7-b0b773f4fc88">FL41</a>

Title of thesis:&nbsp;From magnetic field seed generation to dynamo in collisionless plasmas
Abstract: The origin of the cosmic magnetization present throughout the universe, greatly impacting the development of a diverse set of astrophysical systems, has long been a subject of research. Observations of cosmic magnetic fields, for example in the weakly collisional intracluster medium, show field magnitudes consistent with having been produced through the dynamo process, which causes small seed magnetic fields in plasmas to grow exponentially through the conversion to magnetic energy from kinetic energy in turbulent flows. Magnetohydrodynamics, the standard method of modeling dynamo, is not valid for weakly collisional systems since it assumes that a high collisionality dominates over competing small-scale processes, such as magnetic field seed-generating plasma instabilities. However, a fully kinetic treatment of the problem in electron-proton plasmas is prohibitively expensive numerically. &nbsp;
In this thesis an advanced collisionless fluid plasma model, with a complexity between that of magnetohydrodynamics and a kinetic model, is used to study magnetic field growth in a plasma in the presence of driven turbulent flows. A significant decoupling of the ion and electron dynamics is observed for the employed mass-ratio, allowing access to previously unexplored physics. The simulations demonstrate both the dynamo process and magnetic field seed generation through the electron Weibel instability. The fluid and magnetic Reynolds numbers, key quantities in classical dynamo theory, are generalized for a collisionless system. This is done by studying the dependence of flow and magnetic energy damping rates on the strength of the pressure isotropization encoded in the ion and electron heat flux closures of the fluid model. A scan over the effective magnetic Reynolds number is performed for the dynamo simulations by adjusting the strength of the electron closure. Strong pressure isotropization from the closure leads to behavior of the magnetic field growth expected from classical dynamo theory, whereas a small isotropization results in seed generation from pressure anisotropy-driven instabilities dominating, as previously observed in kinetic simulations of electron-positron plasmas.&nbsp;
Examiner: Istvan Pusztai Supervisor: Istvan Pusztai Opponent: Liam Antonsson 
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#7241f650-27c9-11ef-a21b-017390dd2bcd" target="_blank" title="Chalmers maps" data-anchor="#7241f650-27c9-11ef-a21b-017390dd2bcd">Eva von Bahr</a>
Online: <a rel="noopener" href="https://chalmers.zoom.us/j/62452572114?pwd=daFN9pigWaUHkDRRGGsavI9lqzEoPN.1&amp;from=addon" target="_blank" title="Zoom" data-anchor="?pwd=daFN9pigWaUHkDRRGGsavI9lqzEoPN.1&amp;from=addon">Zoom</a>

Title of thesis:&nbsp;Modeling of Electron-Phonon Coupling in Perovskites
Abstract:&nbsp;Electron-phonon coupling plays a central role in determining key material properties, such as carrier mobility and optical response. However, accurately modeling these interactions remains computationally demanding and methodologically complex. In this work, different modeling approaches to electron-phonon coupling are explored and applied to two prototypical perovskite systems: the halide perovskite cesium lead bromide and the oxide perovskite calcium titanate. For the oxide, the study also includes Ruddlesden–Popper phases. The objective is to understand temperature-dependent electronic properties and to develop models for band edge fluctuations and polarization behavior. For cesium lead bromide, the study focuses on the cubic phase. During molecular dynamics simulations, band gap and valence band maximum data were obtained from density functional theory calculations, and corresponding phonon mode coordinates were extracted. A feature-based model was developed to describe band edge fluctuations as a function of squared mode coordinates. Several feature selection methods were tested, and the model was found to adequately capture fluctuations in the valence band maximum using a subset of the total phonon modes. For calcium titanate, tensorial neuroevolution potentials were trained to predict Born effective charges using density functional theory data from the perovskite structure and two Ruddlesden-Popper phases. Subsequently, predictions were made for Ruddlesden-popper phases outside of the training set. The predicted Born effective charges were used to calculate spontaneous polarization, and the model successfully reproduced the polarization drop across the ferroelectric–paraelectric phase transition.
Examiner: Paul Erhart Supervisors: Erik Fransson Opponent: Felix Uddén
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers map" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2">PJ-salen</a>

Title of thesis: AI-driven single-particle tracking for cancer cell characterization&nbsp; Subtitle: &nbsp;Tracking vesicle-encapsulated nanoparticles as dynamic biomarkers 
Abstract:&nbsp;Prostate cancer exhibits altered intracellular phenotypes that require novel biomarker approaches for accurate diagnosis and treatment stratification. This thesis presents an integrated computational framework to reconstruct and analyze the intracellular dynamics of vesicle-encapsulated gold nanoparticles (AuNPs) from large-scale scattering microscopy videos, with the goal of identifying dynamic indices that can serve as potential biomarkers for distinguishing between prostate cancer cell lines. &nbsp; The proposed framework combines two major components: trajectory reconstruction using deep learning models, specifically the MAGIK graph neural network, and inference of diffusion properties from the reconstructed trajectories. To address the challenges caused by the large scale of the experimental videos, the study introduces a segmentation-based pipeline that processes smaller video sequences, integrates predicted graphs, and builds accurate nanoparticle trajectories. Importantly, the MAGIK model is trained on simulated trajectories that simulate biologically relevant motion types, reducing the need for labor-intensive manual annotations. &nbsp; Subsequently, the study applies a modified version of the MAGIK model to predict point-wise diffusion exponent values for each trajectory, allowing classification of distinct motion types such as directed motion and subdiffusive motion. Among the extracted dynamic indices, nanoparticle velocities during directed motion emerge as promising biomarkers, showing different distributions between the LNCaP and PC3 prostate cancer cell lines. &nbsp; Overall, this work demonstrates the potential of applying deep learning methods to uncover novel dynamic biomarkers in cancer research. Future directions include extending the analysis to additional diffusion properties and expanding the repertoire of dynamic indices with biological significance to further enhance biomarker discovery and improve our understanding of intracellular transport in cancer cells.
Examiner: Giovanni Volpe Supervisors: Mirja Granfors, Jesús Pineda Opponents: Yixuan Xu, Brian Senabulya
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#37d8de3e-1806-4420-92e7-b0b773f4fc88" target="_blank" title="Chalmers maps" data-anchor="#37d8de3e-1806-4420-92e7-b0b773f4fc88">FL-41</a>

Title of thesis:&nbsp;Self-trapped excitons in the surface of BiVO4
Abstract: In recent years, BiVO4 has emerged as one of the most promising photocatalysts for usage in solar-driven water-splitting applications. Its favourable placement of band edges close to the redox potentials of water, together with the physical stability of the material, makes it stand out from many of its competitors. However, previous studies have demonstrated that the bulk structure of BiVO4 suffers from considerable charge localisation, with reduced charge mobility and worsened photocatalytic properties as a result. This study aims to continue these investigations by analysing the formation of localised states at the material-vacuum interface. Using hybrid density functional theory, simulations are performed on two different surface structures in order to identify the most commonly appearing charge localisations, as well as compute their respective formation energies. For both surfaces, several different localised states are found, with self-trapped excitons appearing both in the direct vacuum interface and in the layer immediately below. By providing energy levels deep within the band gap of BiVO4, these localisations are shown to potentially hamper the overall capability of the material to drive the redox reaction
Examiner: Julia Wiktor Supervisors: Tobias Möslinger Opponent: Leo Ögren Westin and Nikhil Saraswat
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers map" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2">PJ-salen</a>

Title of thesis:&nbsp;Electromagnetic modeling of the effect of plasma sheath on radar signatures of hypersonic targets
Abstract: Hypersonic missiles have been suggested to have significantly different radar signatures compared to similar slower vehicles. This has been attributed to the plasma sheath that forms when high-temperature air in the hypersonic shock front partially ionizes air molecules. Under certain conditions, the free electrons in this plasma can absorb electromagnetic radiation. It is unclear whether this effect could dampen the scattered radiation from targeted hypersonic missiles, effectively making them stealthy to radar. However, the physics of plasma-induced stealth of hypersonic vehicles is quite complex, involving the spatial distributions of plasma and electron–neutral collision frequencies as well as the microwave frequency. For this reason, broad claims about the stealth of hypersonic vehicles must be supported by accurate numerical simulations. In this thesis, the effects of a thin plasma sheath have been studied in various settings and geometries with the aid of electromagnetic as well as hydrodynamic simulations.
The ability of a plasma to shield rough surfaces, which otherwise have a large radar signature, was investigated by simulating the electromagnetic wave scattering of a plasma-covered metal slab with surface roughness in a 2D geometry. Somewhat unexpectedly, the plasma did not behave like a mirror for low frequencies and when the radiation was expected to be attenuated it was instead amplified. Additionally, the ability of the plasma sheath to distort scattered electromagnetic waves was studied analytically. The distortion was exaggerated in this simple model compared to a real-world scenario, but still it was negligible for almost every frequency. Therefore the distortion can be neglected. Finally, to study the effect of a plasma sheath with a realistic plasma density, the plasma frequency and collision frequency distributions were computed for a metal sphere flying at speeds of 3-5 km s−1 in an ambient pressure representative of a 40 km altitude. While the calculated radar signature has a limited accuracy below 1 GHz, we can conclude that in that frequency range the presence of the plasma sheath can either increase or reduce the radar signal, depending on the frequency. At some frequencies the decrease was as large as 10 dB. Our results also indicate that the plasma sheath only has minor effects on the radar signatures of hypersonic vehicles traveling below 3 km s−1, for the microwave bands above 1 GHz.
Examiner: Istvan Pusztai Supervisor: Andréas Sundström&nbsp; Opponent: Nils Müller
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#7241f650-27c9-11ef-a21b-017390dd2bcd" target="_blank" title="Chalmers maps" data-anchor="#7241f650-27c9-11ef-a21b-017390dd2bcd">Eva von Bahr</a>
Online: <a rel="noopener" href="https://chalmers.zoom.us/j/68504068099?pwd=ajUWKmwdOePpW61PhdKhSUzA2gBnxq.1&amp;from=addon" target="_blank" title="Zoom" data-anchor="?pwd=ajUWKmwdOePpW61PhdKhSUzA2gBnxq.1&amp;from=addon">Zoom</a>

Title of thesis:&nbsp;Hydrogen in Ferrite Systems with Carbide Precipitates
Abstract:&nbsp;This work investigates the behavior of hydrogen in ferritic steel, two distinct complex metal carbides, and hybrid ferrite–carbide systems using computational methods. A neural network potential was trained on density functional theory data and used to perform molecular dynamics simulations on large-scale systems over nanosecond timescales. Hydrogen diffusivity was evaluated across ferrite and carbide systems of varying compositions. The two carbides were found to exhibit distinct diffusion pathways and hydrogen trapping sites. In the combined ferrite–carbide systems, hydrogen trapping occurred both at the interface and within the carbide bulk, with some differences between the two carbide types.
Examiner: Paul Erhart Supervisors: Erik Fransson Opponent: Olle Lyth Andersson
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers map" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2">PJ-salen</a>

Title of thesis:&nbsp;Kinetic simulations of the proton-alpha instability in collisionless shocks
Abstract: The solar wind is a dynamic stream of collisionless plasma originating from the Sun. When it reaches the Earth's magnetosphere, it abruptly slows down from super-Alfvénic to sub-Alfvénic speeds, creating a bow shock. This compresses and heats the inflowing plasma. The solar wind consists primarily of electrons, protons and alpha-particles, which interact differently with the shock. A relative drift between protons and alpha particles develops, which can trigger a streaming instability, driving waves. Such waves have previously been identified in spacecraft observations. The dependence of the instability on the particle species temperature, density, and flow velocity has previously been investigated using the dispersion relation for an unmagnetized plasma.
In this work, we investigate this proton-alpha streaming instability using two-dimensional fully kinetic Particle-In-Cell simulations. This approach offers the flexibility to freely choose the initial conditions while simulating the full non-linear dynamics. We study the parametric dependence of the instability on the flow velocity of the alpha particles, the density ratio of the ions as well as the influence of a background magnetic field. In particular, we want to quantify the heating of different particle species due to the instability. Additionally, we aim to test the applicability of the simplified linear theory.
Our simulations show that the instability mediates a strong energy exchange from the flow speed of the alphas into plasma heating of the ions. We attribute this to a Landau resonance between the proton and the alpha distribution, which produces large plateaus in velocity space. The electron dynamics do not change significantly. The wave properties are consistent with the linear theory when the external magnetic field is small. Generally, the angle between the wave and the drift direction increases with the flow speed. The fraction of energy converted to plasma heating reaches a maximum at a wave angle of 45 deg to the drift direction. Overall, the existence of a background magnetic field slightly suppresses the instability, due to the magnetization of electrons. We conclude that the proton-alpha streaming instability can play an important role in heating proton and alpha particles across collisionless shocks.
Examiner: Istvan Pusztai Supervisor: Konrad Steinvall Opponent: Lise Hanebring
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#7241f650-27c9-11ef-a21b-017390dd2bcd" target="_blank" title="Chalmers maps" data-anchor="#7241f650-27c9-11ef-a21b-017390dd2bcd">Eva von Bahr</a>
Online: <a rel="noopener" href="https://chalmers.zoom.us/j/66511118346?pwd=AyVGa5AHrnmCp7uK4e1VmtJE55NhQv.1&amp;from=addon" target="_blank" title="Zoom" data-anchor="?pwd=AyVGa5AHrnmCp7uK4e1VmtJE55NhQv.1&amp;from=addon">Zoom</a>

Title of thesis:Interacting particle systems for constrained optimization
Examiner: Axel Ringh Supervisor: Akash Sharma Opponent:&nbsp;Lovisa Åkesson
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#05137ad7-4d34-45e2-9d14-7f970517e2b6" target="_blank" title="Chalmers maps" data-anchor="#05137ad7-4d34-45e2-9d14-7f970517e2b6">MV: L14</a>

Title of thesis:Detection of Scalar Wave Dark Matter with a Levitated Superconductor
Abstract:&nbsp;: The nature of dark matter is one of the most elusive problems in modern physics. With varying candidates and a vast range of possible particle masses, there is still no direct evidence of dark matter through an experimental observation. In this thesis, the possibility of detecting dark matter in a laboratory setup is theoretically investigated. The candidate in question will be the scalar case of the wave-like dark matter, a particle dark matter that belongs in the ultralight mass spectrum. The proposal is that this type of dark matter can be described by a classical wave instead of a quantum field and will interact with neutrons in matter through a simple, linear, non-derivative term in the Lagrangian. This interaction would result in an sinusoidal force that can be detected in a proposed magnetically levitated superconductor experiment. The experimental setup is parametrized as a coupled harmonic oscillator system, where a vibrational shield is installed to minimize vibrational noise. Through stochastic modeling of vibrational noise and numerical methods such as Monte Carlo simulations, the motion of the target superconductor is studied. The end result of this analysis was answering the question of how sensitive such a detection proposal could be, based on the boundaries of the interaction coupling per frequency of the dark matter signal, which in the case of wave dark matter corresponds to a specific mass. We conclude that in a two-stage model, with experimental parameters corresponding to a possible levitation experiment in Chalmers, the vibrational noise dominates over thermal and measurement noise, bringing down the sensitivity. As a comparison we plot the minimum coupling per mass for an ideal case where vibrational noise is scaled down.
Examiner: Riccardo Catena Supervisor: Riccardo Catena Opponent:&nbsp;Konstantinos Topalakis
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#b46f0611-66a4-4fd8-b340-f760175d61f4" target="_blank" title="Chalmers maps" data-anchor="#b46f0611-66a4-4fd8-b340-f760175d61f4">F-N6115</a>

Title of thesis: Chiral Effective Theory of Direct Detection Spin-1 Dark Matter
Abstract: This thesis aims to investigate the elusive dark matter via formulation of its interaction with normal matter. This approach is commonly known as direct detection which is an active field of dark matter research, both theoretically and experimentally. We treat the dark matter as a heavy complex vector field, corresponding to a massive spin-1 particle, which weakly interacts with a nucleus.   The paper utilizes effective field theory and chiral perturbation theory to create a general interaction Lagrangian. Starting in a relativistic approach, quantum mechanical interaction operators are found which in turn can be used to calculate scattering amplitudes. The thesis assumes non-relativistic and heavy dark matter to perform a non-relativistic reduction which aims to investigate what constraints and bounds created in the relativistic regime that lives on in the non-relativistic limit. The results are compared to previous work within the field which leads to a comprehensive analysis of the situation. 
Examiner: Riccardo Catena Supervisor: Riccardo Catena Opponent: Alma Cavallin
Location of presentation: room <a rel="noopener" href="https://maps.chalmers.se/#b46f0611-66a4-4fd8-b340-f760175d61f4" target="_blank" title="Chalmers Maps" data-anchor="#b46f0611-66a4-4fd8-b340-f760175d61f4">F-N6115</a>

Title of thesis:&nbsp;Simulating Light Dark Matter Signals at SHiP
Abstract: Dark matter remains one of the most compelling mysteries in modern physics, with a wide range of astrophysical and cosmological observations providing strong evidence for its existence. Among the many proposed candidates, light dark matter, with masses below the GeV scale, has gained increasing interest, partly due to its accessibility in current and next-generation accelerator-based experiments. One interesting framework for light dark matter involves a massive dark photon mediator A′, which kinetically mixes with the Standard Model photon, enabling interactions between the visible and dark sectors. This thesis focuses on the sensitivity of the SHiP (Search for Hidden Particles) experiment – a planned proton beam-dump experiment at the CERN SPS – to light dark matter scenarios. This experiment is designed to probe feebly interacting particles and is expected to significantly extend the parameter space coverage for light dark sector particles. The analysis adopts a benchmark model characterized by a minimal dark sector with a complex scalar dark matter candidate χ, a dark coupling αD = 0.1, and a mediator mass set to mA′ = 3mχ. Simulations are performed using two independent tools: BDNMC, a standalone dark matter Monte Carlo generator, and MadDump, a plugin for the MadGraph5_aMC@NLO framework tailored for beam dump and fixed-target experiments. A comparative analysis of these tools is conducted in the context of the complex scalar model, highlighting methodological differences and their impact on predicted signal rates. Furthermore, the sensitivity studies for the SHiP experiment is extended to fermionic dark matter scenarios, including both Dirac and Majorana candidates, resulting in updated exclusion contours and emphasizing the experiment’s discovery potential across these different dark matter hypotheses. 
Examiner: Riccardo Catena Supervisors: Riccardo Catena &amp; Taylor Gray Opponent:&nbsp;Lucas Abrahamsson
Location of presentation:&nbsp;room <a rel="noopener" href="https://maps.chalmers.se/#b46f0611-66a4-4fd8-b340-f760175d61f4" target="_blank" title="Chalmers Maps" data-anchor="#b46f0611-66a4-4fd8-b340-f760175d61f4">F-N6115</a>

Title of thesis:&nbsp;Entropy Production of a Harmonically Driven Order Parameter Field Subtitle: A Field Theoretic Approach to Ultrafast Thermodynamics
Abstract: 
Ultrafast thermodynamics is an emerging field of research that aims to model the non-equilibrium dynamics of laser-excited materials. A central quantity of interest is the entropy production, which quantifies how far a system is from equilibrium. While previous theoretical works have modeled the entropy production in terms of lattice vibrations and magnetization dynamics, this work models the entropy production of a scalar order parameter field coupled to a time-dependent harmonic driving field. In doing so, this thesis takes the first steps in bridging the gap between ultrafast thermodynamics and statistical field theory.
The evolution of the field is modeled by the time-dependent Ginzburg-Landau equation, which is solved numerically. In the absence of a driving field, the validity of this method is confirmed through a finite-size scaling analysis in which equilibrium critical exponents are extracted and shown to agree with known values of the Ising/ϕ^4 universality class.
An analytical expression for the entropy production is also proposed. This expression is compared with numerical simulations and, for a selected range of driving frequencies, exhibits clear quantitative agreement in the high-temperature phase. A finite-size scaling analysis of the entropy production further suggests that, for the driving protocol in question, the entropy production converges to a smooth function around the critical temperature, and as such does not exhibit any critical scaling.
Examiner: Matthias Geilhufe Supervisors: Matthias Geilhufe Opponent: Jacob Ljungberg
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers map" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2">PJ-salen</a>

Title of thesis:&nbsp;Method Development of a Multi-Layered Optical Skin Phantom
Abstract:&nbsp;
The increasing research interest in and use of optical medical technologies has led to a growing need for synthetic testing materials with tissue-mimicking optical properties, known as optical phantoms. Many different materials and methods have been used to create such phantoms but the field currently lacks standardization and repeated results. Optical phantoms made for short wavelengths of light are also missing from research.
This project tests and builds on previously established methods for creating optical phantoms made out of PDMS silicone. A method for creating a multi-layered skin phantom in which each layer mimics the thickness and absorption of one of the main layers of human skin is developed and presented. Four versions of this phantom are made: two of them mimic the absorption of fair skin and two of them mimic the absorption of dark skin. For each skin colour, one of the phantoms mimic the absorption at a wavelength of 405 nm and the other at 630 nm.
A correlation between the concentration of the absorbing agent (nigrosin) and the absorption coefficient of the resulting phantoms is determined using spectrometry. This correlation can work as a guide for how to reach a desired absorption coefficient.
Examiner: Julie Gold Supervisors: Marcus Andersson and Sofia Dall'Orso&nbsp; Opponent: Hugo Söderberg
Location of presentation:&nbsp;<a rel="noopener" href="https://maps.chalmers.se/#32e47c30-31d5-11e8-8b55-73c20346dfc4" target="_blank" title="Chalmers maps" data-anchor="#32e47c30-31d5-11e8-8b55-73c20346dfc4">Mildred Dresselhaus</a>
Online:&nbsp;<a rel="noopener" href="https://chalmers.zoom.us/j/62208303717" target="_blank" title="Zoom">Zoom</a>

Title of thesis:&nbsp;Toward an automatic two-qubit-gate calibration
Abstract: Calibrating a large quantum processor is a challenging task as it requires optimizing many parameters for each gate. As the number of gates increases, the amount of human intervention must be reduced to a minimum and an automatic calibration must be developed to carry out all these complex calibrations. The quantum processor developed by the computing group at Chalmers is already complex enough to require such automation, as it contains 25 qubits and 40 couplers. The group has developed an application called Tergite automatic calibration that can perform the single-qubit calibration but is not yet capable of calibrating two-qubit gates. This Master's thesis focuses on this crucial missing functionality; in this work, a fully automatic calibration procedure is presented that can calibrate two-qubit gates for the 25-qubit chip. Two key aspects have been considered in this work; the possibility of reliably finding flux points that can be used for the calibrations for the CZ-gate and the possibility do this for multiple couplers in parallel. Robustness measurements are presented to show that the autocalibration procedure can reliably find the calibration parameters in different conditions. Crucially, the algorithm developed is not only capable of calibrating multiple CZ gates, but it can do in a relatively fast, enough to allow reliable operations of the quantum processor.
Examiner:&nbsp;Jonas Bylander Supervisors: Michele Faucci Giannelli Opponent: Nils Nerpin
Location of presentation:&nbsp;<a rel="noopener" href="https://maps.chalmers.se/#96de5091-96fe-4e70-8def-1faa831680ae" target="_blank" title="Chalmers maps" data-anchor="#96de5091-96fe-4e70-8def-1faa831680ae">Raven &amp; Fox</a>
Online: <a rel="noopener" href="https://chalmers.zoom.us/j/67950794075" target="_blank" title="Zoom">Zoom</a>

Title of thesis:&nbsp;AI-Driven Real Estate Valuation in Non-Metropolitan Sweden: A Hybrid Modeling Approach
Abstract: This thesis investigates the potential of machine learning to enhance real estate valuation accuracy in non-metropolitan regions of Sweden, where traditional appraisal methods often falter due to sparse transaction data and diverse property characteristics. Using a dataset of approximately 90,000 residential properties across Östergötland, Småland, and Blekinge, the study evaluates a hybrid modeling approach combining artificial neural networks and gradient boosting. The ANN leverages the entity embeddings for complex categorical variables. Additionally, feature importance and interpretability techniques, such as SHAP values, are employed to uncover valuation drivers and enhance model transparency. Results indicate that ML-based models can significantly outperform traditional hedonic pricing models in both accuracy and robustness, particularly in data-sparse settings. The findings support the use of hybrid AI techniques as viable complements to conventional valuation tools, offering improved scalability, consistency, and insight into property value determinants for stakeholders such as lenders, appraisers, and public agencies.&nbsp;
 Examiner: Mats Granath Supervisor: Magnus Persson&nbsp; Opponent: Robin Böhme, Lucas Kallenberg
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#37d8de3e-1806-4420-92e7-b0b773f4fc88" target="_blank" title="Chalmers Maps" data-anchor="#37d8de3e-1806-4420-92e7-b0b773f4fc88">FL41</a><a rel="noopener" href="https://maps.chalmers.se/#24d8b7be-1f59-4388-be28-a07861b429b2" target="_blank" title="Chalmers Maps" data-anchor="#24d8b7be-1f59-4388-be28-a07861b429b2"></a><a rel="noopener" href="https://maps.chalmers.se/#36a610c1-e9b3-48bd-b01c-9501f9af79fe" target="_blank" title="Chalmers map" data-anchor="#36a610c1-e9b3-48bd-b01c-9501f9af79fe"></a>
Online: <a rel="noopener" href="https://chalmers.zoom.us/j/65533019596" target="_blank" title="Zoom">Zoom</a>

Title of thesis:&nbsp;Correlated Effective Field Theory Truncation Errors: From Neutron-Proton Scattering Amplitudes to Observables
Abstract:&nbsp;
The strong force governing Nucleon-Nucleon (NN) scattering can be modelled with an &nbsp;Effective Field Theory (EFT) of low-energy quantum chromodynamics. The Feynman diagrams contributing to this force are considered order-by-order in an expansion with decreasing importance. NN scattering observables can then be predicted by considering all contributions up to some finite order, while neglected orders result in predictable truncation errors. Contrary to previous studies, the EFT truncation errors of different NN observable types should not be considered independent since all of them depend on only five complex so-called scattering amplitudes. Instead, truncation errors must be correlated not only across energies and scattering angles but also across different observable types. In this thesis, we modelled and investigated such correlations between the uncertainties of different neutron-proton scattering observables resulting from a truncation of the chiral EFT guided by Weinberg power counting. &nbsp;
We applied a Bayesian uncertainty-quantification model, consisting of a fixed-kernel Gaussian process, on the leading order chiral EFT truncation errors of the neutron-proton scattering amplitudes. The resulting errors were then propagated to errors of various observables. We considered both Saclay and helicity conventions for the NN scattering amplitudes with their specific angular boundary constraints. Firstly, we found that the symmetry constraints of amplitudes propagate to observables, resulting in angle- and energy-dependent uncertainties. Secondly, we observed strong angle-dependent correlations and rigid inequalities between different observable types, cutting the joint posteriors into characteristic shapes traced back to their amplitude dependencies. Finally, we examined the extent to which our error model violates the unitarity constraint of the scattering operator by comparing the optical-theorem and integrated total cross section. There was little to no correlation between the uncertainties of these theoretically identical expressions. While we successfully incorporated unitarity on average, future studies are necessary to explicitly implement unitarity constraints into the error model.
This work is an important step in a larger effort to understand correlated EFT truncation errors, considering rigorously the correlation between different NN scattering observable types. The resulting error model could enhance the physical robustness of EFT parameter inference and the reliability of theoretical predictions for nuclear observables.
Examiner: Christian Forssén Supervisors: Christian Forssén and Oliver Thim Opponent: Elias Svensson
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#7241f650-27c9-11ef-a21b-017390dd2bcd" target="_blank" title="Chalmers maps" data-anchor="#7241f650-27c9-11ef-a21b-017390dd2bcd">Eva von Bahr</a>

Title of thesis:&nbsp;Entanglement and Accidental Symmetries in Nuclear Effective Field Theory
Abstract:&nbsp;I
Effective field theory descriptions of nuclear forces automatically account for symmetries that are manifest in the underlying theory of quantum chromodynamics. In contrast, the consequences of accidental nuclear symmetries, such as Wigner SU(4) and Serber symmetries, often remain hidden. Wigner SU(4) symmetry dictates spin- and isospin-independent interactions while the orbital-parity Serber symmetry requires that the interaction is zero in partial waves with odd parity. Furthermore, recent work has highlighted an interesting connection between spin-entanglement suppression and Wigner SU(4) symmetry.
In this thesis, we incorporate these accidental symmetries into chiral effective field theory (ChEFT) and perform a quantitative study of their effects from an entanglement perspective. We construct five modified versions of the leading order Weinberg power counting (WPC) potential and study pure spin entanglement for the neutron-proton system.&nbsp; Consequently, this study serves as a first step in the investigation of a possible entanglement-guided power counting in ChEFT.&nbsp;
We show that a Wigner SU(4) and Serber symmetric interaction results in a scattering-amplitude matrix that is proportional to the zero-entangling identity operator for all scattering angles and for lab energies less than 300 MeV. Meanwhile, a short-distance breaking of Wigner SU(4) symmetry results in entanglement solely at low energies which, in the very-low-energy limit, depends sensitively on the specific values of the S-wave low-energy constants. At intermediate energies, the Wigner-breaking tensor interaction in the leading order WPC potential is responsible for a large contribution to entanglement while Serber breaking in odd-parity spin-singlet waves results in entanglement in forward and backward scattering directions. The order-by-order entanglement pattern for WPC up to next-to-leading order is also presented and compared with the results obtained with the phenomenological NijmegenI potential. We conclude that entanglement provides a new perspective on the relative importance of different terms in the nuclear force. This work therefore advocates for future studies on entanglement-guided power-counting schemes in nuclear effective field theory.
Examiner: Christian Forssén Supervisors: Oliver Thim and Christian Forssén Opponent: Arvid Ryberg
Location of presentation: <a rel="noopener" href="https://maps.chalmers.se/#7241f650-27c9-11ef-a21b-017390dd2bcd" target="_blank" title="Chalmers maps" data-anchor="#7241f650-27c9-11ef-a21b-017390dd2bcd">Eva von Bahr</a>

Architecture and Civil Engineering

Bachelor’s thesis Chemistry and Chemical Engineering

Chemistry and Chemical Engineering

Communication and Learning in Science

Computer Science and Engineering

Electrical Engineering

Industrial and Materials Science

Life Sciences

Mathematical Sciences

Mechanics and Maritime Sciences

Microtechnology and Nanoscience

Space, Earth and Environment

Technology Management and Economics

Presentation of master thesis

Choose subject

Course syllabus for

Subject

Each academic year contains different courses

See programme page at universityadmissions.se

Educational area

Permissible substitutions of compulsory courses

To see your programme syllabus – select the academic year in which you started the programme.

[en: page.studyprogrammepage.syllabusesview.coursesdescription]

Master thesis presentations at Physics

June 2, 2025

9:00- 10:00 August Klynne & Malte Åqvist, MPCAS

10:00- 11:00 Ebba Danielsson & Josefin Ewerman, MPBIO

13:00- 14:00 Andreas Erlandsson, MPPHS

14:00- 15:00 Simon Dovrén, MPPHS

June 3, 2025

9:00- 10:00 Björn Johansson, MPPHS

10:00- 11:00 Oscar Karlsson, MPCAS

10:00- 11:00 Hugo Mårtensson, MPPHS

11:00- 12:00 Ahsin Hafeez, MPSES

11:00- 12:00 Erman Kulaglic & Jakob Bergman, MPCAS

11:00- 12:00 Erik Setterstål, MPCAS

13:00- 14:00 Sophie You, MPMED

13:00- 13:45 Victor Enevold, MPPHS

13:15- 14:15 Ole Fjeldså & Gustaf Jonasson Johansson, MPCAS

13:45- 14:30 William Colliander, MPPHS

15:00- 16:00 Nils Nerpin, MPPHS

15:00- 16:00 Marcus Andersson & Filip Olsson, MPCAS

June 4, 2025

10:00- 11:00 Johan Larsson & Oliver De Rosa, MPCAS

10:30- 11:30 Joel Carlsson, MPCAS & Denis Grahovic, MPSYS

June 5, 2025

8:00- 9:00 Carl Kronqvist & Malte Olsson, MPCAS

14:00- 15:00 Tobias Wallström, MPCAS

15:00- 16:00 Markus Utterström, MPCAS

15:30- 16:30 Lovisa Åkesson, MPPHS

June 9, 2025

14:00- 15:00 Mathilda Gustafsson, MPCAS

15:00- 16:00 Erik Broback, MPCAS

16:00- 17:00 Rasmus Nilsson & Oscar Rosman, MPCAS

17:00- 18:00 Viktor Örnbratt, MPCAS

18:00- 19:00 Francisco Boudagh, MPCAS

June 10, 2025

10:00- 11:00 Cassandra Hellgren, MPMED

13:00- 14:00 Pontus Kjellström & Niklas Lundblad, MPCAS

13:00- 14:00 Lise Hanebring, MPPHS

13:00- 14:00 Leo Ögren Westin, MPPHS

14:00- 15:00 Yuchao He, MPCAS

14:00- 15:00 Olle Lyth Andersson, MPPHS

14:10- 15:10 Liam Antonsson, MPPHS

15:00- 16:00 Felix Uddén, MPPHS

15:20- 16:20 Nils Müller, MPCAS

June 11, 2025

14:15- 15:15 Jonathan Borsander, MPPHS

June 12, 2025

9:00- 10:00 Ioannis Moutsis, MPPHS

10:00- 11:00 Arvid Ryberg, MPPHS

11:00- 12:00 Elias Svensson, MPPHS

14:00- 15:00 Erik Karlsson Öhman, MPPHS

June 13, 2025

9:00- 10:00 Louise Eckerström, MPBIO

11:00- 12:00 Joel Sandås, MPPHS

13:15- 14:15 Leonard Smedenman & Teddy Sallén, MPCAS

13:15- 14:15 Lucas Abrahamsson, MPPHS

14:30- 15:30 Alma Cavallin, MPPHS