Chalmers Open Digital Repository
Välkommen till Chalmers öppna digitala arkiv!
Här hittar du:
- Studentarbeten utgivna på lärosätet, såväl kandidatarbeten som examensarbeten på grund- och masternivå
- Digitala specialsamlingar, som t ex Chalmers modellkammare
- Utvalda projektrapporter
Forskningspublikationer, rapporter och avhandlingar hittar du i research.chalmers.se
Enheter i Chalmers ODR
Välj en enhet för att se alla samlingar.
Senast inlagda
Analysis and Generation of Wikidata Descriptions Focusing on Bangla Language
(2025) Rakib Imtiaz, Mohammad
We present a Grammatical Framework (GF)-based resource grammar for Bangla designed to automatically generate structured natural language descriptions for Wikidata entities. The system covers multiple entity types including cities, universities, islands, lakes, and humans. Unlike statistical or black-box models, our approach uses a rule-based grammar that guarantees grammatical correctness and structural consistency. Evaluations on more than 76,000 entities demonstrate high coverage (over 99%) and strong alignment with source descriptions, as shown by multilingual embedding similarity. Our results show that the generated Bangla descriptions not only complement existing entries but often exceed them in semantic consistency. This work offers a practical solution for enhancing low-resource language content in multilingual knowledge bases.
Multi-Agent Large Language Model as AD/ADAS System Engineer
(2025) Alkhaled, Ali; Malla, Ali
Recent advancements in generative AI, particularly in Large Language Models (LLMs) have sparked a major revolution and a qualitative shift in various fields, including software code generation and unit test generation, offering new opportunities to automate various aspects of the software development process. At the same time, the demand of sophisticated software in the automotive industry has grown rapidly. This trend motivates the exploration of the potential of LLMs in supporting the development of AD/ADAS functions. A pipeline, CoTeGen, for code generation, test case generation, and the automation of virtual simulation-based testing in Esmini is designed following three iterative development cycles. The pipeline is designed to address four AD/ADAS functions. The first two are constrained to relatively elementary maneuvers, namely simple braking and lane changing, whereas the latter are dedicated to more sophisticated control tasks, specifically Adaptive Cruise Control and Collision Avoidance. Across these iterative cycles, the pipeline progressed from generating non-compilable software components to providing compilable and functional software. Based on a multi-run experimental evaluation involving five open-source LLMs, Codellama:7B, Mistral:7B, DeepSeek-Coder-v2:7B, Gemma3:4B, and Qwen2.5-Coder:7B, the pipeline shows a clear ability to generate correct source code for the simpler functions, while proving far less effective for the more advanced functions. Finally, we discuss the challenges and limitations of applying LLMs to code and unit test generation within the proposed pipeline.
Identification of Loads Based on Historical Data
(2025) Lin, Meixi; Supakkeittikul, Pirapon
To tackle the forthcoming electricity grid tension in Gothenburg, Sweden, this thesis introduces a two-step machine learning approach for managing electricity demand. Initially, consumers are grouped using K-Means clustering according to their past patterns of usage in order to identify categories with the most fluctuating behaviour. Subsequently, several Long Short-Term Memory (LSTMs)—Vanilla, Stacked, Bidirectional, and CNN-LSTM—are trained for predicting electricity demand for such high-impact consumer groups in response to real-time, varying price signals. These models are evaluated using mean absolute error (MAE), root mean square error (RMSE), and loss measures. Among these examined architectures, CNN-LSTM exhibits the most consistent and stable performance across test and prediction datasets. This approach minimises the data and computation needed for deep learning but
allows for more customised forecasting. The proposed solution provides a resourceefficient and scalable solution for energy suppliers who wish to monitor changes in demand in response to price changes.
Vehicle Design Optimization Using AI/ML Methods
(2025) Venugopal Chetan Acharya, Puthige
Energy efficiency and performance are important attributes when developing and designing vehicles. With the transition of the automotive industry towards energy efficiency and sustainability, it is ever more important to save computational resources. Traditional vehicle design optimization heavily relies on computationally expensive simulations. The simulations carried out in this project particularly focus on the powertrain of vehicles. This work focuses on developing a surrogate-assisted, multiobjective optimization framework that efficiently finds the optimal values for the given variables. Surrogate modeling is an engineering method used when the outcome of an experiment cannot be easily computed, so a mathematical approximation is applied. In this case, we use machine learning models to predict the outcome of expensive simulations. These trained model(s) is then used for optimization instead of running optimization on the simulation model directly. First, we generate 4096 Sobol-sampled configurations spanning different parameters like gear ratios and electric motors. We train and compare different surrogate models like Random Forest, XGBoost, and LightGBM on these data, achieving test R2 scores up to 0.96 with Random Forest. Next, we employ the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) to explore trade-offs among various conflicting objectives, extracting a Pareto front of optimal designs. A weighted-sum post-processing step or a constrained method later selects a single best-trade-off configuration, which full simulation validates. This framework slashes computational cost and empowers rapid, data-driven vehicle powertrain design.
The Borsuk-Ulam Theorem in Synthetic Stone Duality
(2025) RASHID, ARMAAN
The recently introduced synthetic Stone duality [1] is an extension of homotopy type theory (HoTT), which is conjecturally modeled in the higher topos of light condensed anima as recently introduced by Clausen and Scholze [2]. As it turns out, synthetic Stone duality turns out to be an appropriate setting in which to develop a restricted form of point-set topology. We survey the development of point-set topology within synthetic Stone duality, recovering a working theory of second countable compact Hausdorff spaces and, in particular, a working interval whose topology is the usual metric topology. Using the interval, we are able to define topological paths and loops, as well as the topological fundamental group, in the standard way. We then turn to the theory of higher modalities in HoTT to relate topological spaces, as developed within synthetic Stone duality, to their homotopy types. After recalling the basics of the theory of higher modalities we are able to define the shape modality as the nullification of the interval. The shape modality, which is conjectured to more generally convert (topological) finite CW complexes to their homotopy types, converts the topological circle 𝕊1 into its higher inductive counterpart 𝑆1. Using shape, we are able to pass back and forth between reasoning about the topological and homotopical circle as convenient. As an application, we use the shape modality to provide a characterization of ℝ, long known as folklore, in terms of the modal unit 𝜂 : 𝕊1 → 𝑆1 for the topological circle. Similarly using the shape modality, we are able to show that topological paths in 𝕊1 lift through the standard covering map ℝ ↠ 𝕊1 while bypassing the standard covering space theory, and prove the two-dimensional Borsuk-Ulam theorem as an application.