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

Bild (thumbnail)
Exploring the Opportunities for Reuse in Infrastructure Projects; A Case Study of a Leading Construction Company in Sweden
(2025) Dires, Hana; Honarkar, Arian
The construction sector is a significant contributor to resource depletion, greenhouse gas emissions, and construction waste, underscoring the need for more sustainable infrastructure development. This study investigates the feasibility of material reuse in infrastructure by examining its key enablers, barriers, and practical pathways for implementation. A qualitative research design was employed, combining a literature review with a case study of a large-scale infrastructure project in Sweden, augmented by semi-structured interviews with stakeholders. The analysis reveals that integrating reuse considerations into the early design phase is pivotal, reinforced by supporting policies, technical capacity, market readiness, collaborative networks, and cultural acceptance. Notable barriers include the lack of standardized guidelines, inconsistent material quality, and logistical challenges related to storage, transport, and supply-demand coordination. The proposed framework emphasizes early design as the foundation for facilitating reuse, supported by regulatory reform, digital tracking tools, and policy incentives to stimulate adoption. This integrated approach provides a viable route for embedding circular economy practices in infrastructure, with promising environmental and economic benefits for the built environment.
Bild (thumbnail)
Semantically Aware Attacks on Text-based Models: An Extension of Context-aware and Neighbourhood Comparisonbased Membership Inference Attacks
(2025) Glänte, Gabriel
Training deep-learning models requires large amounts of data. When this data is sensitive, e.g., containing personal information, it is important to ensure that no sensitive information can be extracted from the trained models. In a membership inference attack (MIA), an adversary is expected to have access to a trained model θ and a data sample d, sampled from the same distribution as the unknown training data. The objective of the adversary is to construct an algorithm A(θ, d) → {0, 1}, where the binary output guesses if d was part of the unknown training data or not. It is commonly assumed that the attacker can access loss values from θ for different prompts; such loss-based signals are crucial for membership checks, even under black-box conditions. For text, the notion of membership is not clear-cut: distinct strings can share the same semantics. Many MIAs therefore fail when they only test exact strings. Recent work reports near-random performance across models and domains (15). This suggests the need to incorporate semantics, i.e., to probe a text together with semantic neighbours that preserve meaning under small, context-appropriate edits. This thesis explores and strengthens such attacks and evaluates them with the standard metrics area under the ROC curve (AUC) and true positive rate at low false-positive rates (TPR@1%FPR). Building on the context-aware membership inference attack (CAMIA) which uses per-token loss sequences rather than a single average loss to construct signals for membership inference (11), the contributions of this thesis are: (i) a custom reimplementation of CAMIA, (ii) integrating a neighbourhood comparison signal that perturbs a text with its semantic neighbours (16), and (iii) novel signals designed to improve loss-informed neighbour generation. Experiments on Pythia-deduped and GPT-Neo models across six subsets of The Pile (19) (streamed via the MIMIR repository (15)) show that these semantics-aware extensions often increase true positive rates at low false positive rates while keeping AUC stable. Overall, modest, loss-guided semantic edits make MIAs more effective for text under realistic black-box conditions.
Bild (thumbnail)
Multi-Objective Optimization Under the Hood: Engine Calibration via Metaheuristics and Probabilistic Methods
(2025) Borg, Sara; Hui, Victor
The automotive industry faces the complex task of optimizing engine performance across diverse and often competing metrics, including fuel consumption and emissions. To effectively address this challenge and manage the necessary trade-offs, accurate engine calibration is essential. This thesis investigates the application of optimization methods, specifically Genetic Algorithms (GA) and Bayesian Optimization (BO), as a promising solution for engine calibration. Single-objective optimization targets the best solution for one goal, while multi-objective optimization balances trade-offs between conflicting goals to approximate the Pareto front, the set of optimal solutions where no objective can be improved without worsening another. This work explores both single-objective and multi-objective optimization implementations for GA and BO. In addition, a hybrid approach combining BO followed by GA is proposed for multi-objective optimization. The methods were evaluated in an experimental study. In the single-objective case, both GA and BO outperformed established internal benchmark values (provided by Volvo). For multi-objective optimization, GA, BO, and the hybrid method also achieved superior results. Across both single-objective and multi-objective problems, BO consistently delivered the best performance. These findings demonstrate that GA, BO, and the hybrid approach are viable strategies for engine calibration and provide a strong foundation for the development of more specialized calibration methods.
Bild (thumbnail)
Bild (thumbnail)
Extensions of Constant Proportion Portfolio Insurance using the Geometric Ornstein-Uhlenbeck process and the Chan-Karolyi-Longstaff-Sanders process
(2026) Bengtsson, Jonathan
We investigate performance of the Constant Proportion Portfolio Insurance (CPPI) strategy and compare it with two of its extensions: Time Invariant Portfolio Protection (TIPP) and Exponential Proportion Portfolio Insurance (EPPI). In order to do this, we model a risky asset (a stock or an index) using a Geometric Ornstein-Uhlenbeck process, and estimate its parameters using the likelihood ratio method with historical price data. We model a non-risky asset (a zero-coupon bound) using a Chan-Karolyi-Longstaff-Sanders process and estimate its parameters using the maximum likelihood method where we approximate the transition probability density function using a Hermite expansion. We find that both extensions of the CPPI improve performance in different ways. The resulting distribution of simulated portfolio outcomes for the TIPP strategy has a lighter tail compared to the CPPI case, and the risk of loss is lower (this is also true compared to the EPPI strategy, but to a smaller degree). The EPPI strategy translates the distribution of simulated portfolio outcomes to the right, so that EPPI performs better than CPPI (and TIPP) in terms of both mean and median.