Chalmers Open Digital Repository
Not a member yet
26247 research outputs found
Sort by
Security-Aware Scheduling of Real-Time Tasks on Multi-core Processors
No full textModern real-time systems are increasingly exposed to timing-based security threats due to their predictable task scheduling. When scheduling tasks for real-time execution, a predictable execution pattern is needed to ensure all tasks will meet their deadlines. A common practice is to employ a fixed-priority scheduler, a deterministic scheduling algorithm always choosing the same task to execute every time it’s given the same conditions. Schedule-based attacks exploit this determinism, enabling adversaries to manipulate or extract sensitive information by aligning their execution with critical tasks. To counter this, schedule randomization has emerged as a potential solution, introducing controlled unpredictability into task execution.
This thesis investigates the application of schedule randomization in multi-core realtime systems, particularly when tasks are pre-allocated to specific cores. The study builds upon TaskShuffler, an already existing algorithm that introduces randomness into the previously deterministic fixed priority scheduler. This algorithm, designed for single-core systems, is now extended for multi-core use. Further, we examine techniques to mitigate or circumvent schedule-based attacks targeting multi-core systems. We also extend the concept of schedule entropy, a “randomness” metric, to better suit multi-core systems, as well as introduce new security-aware metric to capture the risk of common types of targeted attacks. We evaluate the security and performance impact of our methods by by simulating tasks execution on multi-core processors under different task sets and configurations. This provides insights into how core assignment and priority relations affect the system’s exposure to schedulebased attacks. Such insights may help the system designer to strengthen the security of the systems by allocating or not allocating certain tasks to certain cores at design time
A Self-Trained Engine for a Chess Variant
No full textChess has long been a benchmark for artificial intelligence (AI) research due to its complexity and well-defined rules. Recent advances, such as AlphaZero, introduced self-learning AI through reinforcement learning and self-play, achieving superhuman performance without prior strategic knowledge, relying solely on the rules of the
game. AlphaZero defeated the world-champion chess engine Stockfish after only four hours of training, leveraging large-scale computational resources to rapidly learn and
refine its strategies. This thesis presents the development of a chess engine for the chess variant Atomic Chess. The engine was developed in C++ and trained through self-play and reinforcement learning, taking inspiration from AlphaZero’s approach. This project explores the extent to which a chess engine with this approach is feasible for the average enthusiast. Cost-effective cloud-based virtual machine instances with powerful hardware were
rented to manage training workloads. Given limited computational resources, we
opted for a data-centric approach, focusing on refining the training pipeline to maximize the training data that could be produced, rather than hyperparameter tuning
and experimenting with neural network architectures. The final engine was trained on approximately 450,000 self-play games in roughly 150 hours.
The final engine was deployed on the chess platform Lichess and achieved an ELOrating of 1,729, which corresponded to the top 10th percentile of Atomic Chess
players on Lichess. These results demonstrate that it is possible to achieve a competitive Atomic Chess engine within a budget of 3,000 SEK for cloud computation.
This shows that strong self-play reinforcement learning agents for niche games can be developed without requiring large-scale computing infrastructure. These results highlight the viability of accessible, low-budget AI research for underexplored game variants
AI-driven Single Image Super Resolution for Improved Neuron Segmentation
No full textConnectomics research relies heavily on high-resolution imaging of neurons, allowing for the segmentation and tracing of nerve structures. Typically, high-resolution images cover only limited areas, while broader overviews are captured at lower resolutions. As segmentation techniques continue to improve, the analysis of these low-resolution regions has become of increasing interest. AI-driven super resolution models offer the potential to upsample low-resolution neural images, enabling automated segmentation in regions that were previously unusable for analysis and increasing precision in areas with high feature density. As this specific application of super resolution is previously unexplored, and given the growing variety of model architectures available, this work investigates three representative models of different architectures, Real-ESRGAN, SR3, and EDT, and variety of training loss functions. The goal is to compare the strengths and limitations of these architectures when fine-tuned on serial block-face electron microscopy images. This task demands a high degree of structural consistency between low and high resolution outputs. REAL-ESRGAN and SR3 were found to be prone to hallucinations and artifacts, which can hinder downstream applications. In contrast, the fine-tuned EDT models tended to produce overly smooth outputs and in this removed small features. Some improvement was achieved with a task-specific EDT model trained from the ground up and the use of structural similarity–based loss functions
Domain-Aware Reasoning with Lightweight Models via Knowledge Distillation
No full textLarge Language Models (LLMs) offer powerful reasoning capabilities but their computational demands often hinder deployment in domain-specific applications like cybersecurity. This thesis investigates the efficacy of knowledge distillation for transferring advanced reasoning from a large teacher model (DeepSeek-R1) to a lightweight student model (Meta-Llama-3.1-8B-Instruct) within a Retrieval-Augmented Generation (RAG) framework for cybersecurity intelligence. Using a dataset of real-world queries and RAG-retrieved context, the student model was fine-tuned via Supervised Fine-Tuning (SFT) on the teacher’s generated reasoning chains, employing Parameter-Efficient Fine-Tuning (PEFT) with Low-Rank Adaptation (LoRA). Comprehensive evaluation, incorporating both AI-assisted analysis and blind domainexpert assessments, demonstrated that the distilled model significantly outperformed both the existing production RAG system at Recorded Future and its base Meta- Llama-3.1-8B-Instruct model. The distilled model exhibited superior contextual accuracy, a marked reduction in hallucinations, and higher overall response quality. This research successfully validates knowledge distillation as a potent strategy for creating computationally efficient, yet highly capable, domain-aware reasoning models, offering a practical pathway to enhance AI-driven solutions in specialized fields
Towards More Sustainable Dishwashing Behaviour. A User Research Study with Design Strategies to Modify Unsustainable Practices
No full textDishwashers have become part of the modern home and when used correctly can lead to significantly less water and energy consumption than the alternative of manual dishwashing. Manufacturers are constantly improving the efficiency of these appliances with better mechanical components, lower cycle temperatures and increases in capacity. However, an aspect which has received less attention is how the behaviours of users impact the efficiency. This project investigated unsustainable behaviours in automatic dishwashing by studying previous research as well as conducting new research, through user studies in a lab environment complemented by a diary study in a home context. Six unsustainable behaviours were identified: (1) inefficient programme choice, where users choose to run a dishwashing programme which is either too intensive or too gentle, resulting in increased water and energy consumption or unclean dishes.
(2) inefficient loading, not utilizing the full capacity of the dishwasher or overloading the dishwasher resulting in unclean dishes;
(3) pretreatment of dishes, rinsing dishes under running water before placing them in the dishwasher;
(4) inefficient detergent use, using more detergent than necessary for a good result;
(5) manual dishwashing, households with dishwashers washing some dishes manually and
(6) poor maintenance, not doing the maintenance needed for effective cleaning and along lifespan of the appliance.
The study found that these behaviours depend on several factors among users, where the most critical ones were a lack of information, a lack of trust in the dishwasher and convenience. The next part of the project consisted of investigating how design could be used to address the unsustainable behaviours. Which resulted in the six design recommendations: sustainability sheet, programme costs, Auto Dose text, tips & tricks, scraping off tool and filter warning. These recommendations in addition to the insights from the research are aimed to serve as inspiration for dishwasher
manufacturers on how they can improve the sustainability of their appliance by employing a user behaviour point of view
Analyzing Simulation Dynamics in AGV Systems for Improved Traffic Prediction Using Graph Neural Networks
No full textAutomated Guided Vehicle (AGV) systems play a critical role in modern industrial
automation, but designing efficient AGV layouts is time-consuming due to the
reliance on extensive simulations. This thesis explores the use of Graph Neural Networks
(GNNs) to predict congestion and waiting times within AGV layouts, with
the goal of accelerating the design process. Building on previous research, we use
a hierarchical GNN model that combines classification and regression to estimate
segment-level waiting times. To overcome data scarcity and simulation inconsistency,
we construct a graph-based dataset from real AGV systems and apply targeted
data augmentation focused on traverse time. Simulation consistency and appropriate
simulation durations are carefully analysed to ensure data reliability. Our results
show that while the classifier performs robustly, surpassing variation baselines set
by the augmentation average, the regressor faces challenges in accurately modelling
continuous waiting times. The study highlights key limitations, including the absence
of AGV count and fleet management logic in the input features. Nonetheless,
the findings demonstrate the potential of GNNs for layout evaluation and provide a
foundation for more generalizable, traffic prediction tools in AGV system design
Stokastisk modellering av smittspridning inom sjukvården med nätverksmodeller
No full textStokastisk modellering av sjukhusinfektioner är ett forskningsområde som har fått mer
uppmärksamhet på senaste tiden. Tidigare arbeten har använt kontaktnätverk för att bestämma hur infektioner kan sprida sig i ett sjukhus [1]. Andra har utvecklat simulationsalgoritmer
för sjukhus med in-och utflöde av patienter.
I detta arbete har stokastiska simulationer av Methicillin-resistant Staphylococcus aureus utförts på en nätverksmodell med patientflöde, där patienterna ges realistiska inskrivningsperioder på sjukhuset.
Interventioner har implementerats för att undersöka om det sker en minskningen av antalet infekterade och om reproduktionstalet påverkas. Resultaten visar på att både intervention handhygien och screening lyckas reducera smittsamheten av sjukhusinfektioner. Screeningparametrar som fördröjning av resultat och antalet test undersöktes. Resultatet visar att det är fördelaktigt att utföra ett större antal tester med ett större intervall mellan testningen i jämförelse med att testa färre personer oftare om totala antalet test hålls konstant. Dessutom gav screening baserad på närhet till nyliga fall, kontaktspårning, ökad effektivitet jämfört med att screena slumpmässiga individer
Extraction and Analysis of Leachates from Synthetic Corium Samples
No full textThe purpose of this project was to create synthetic corium samples and subsequently
measure the leaching behaviour of these. The study aimed to see if different compositions affected the leaching behaviour and leaching rate. To accomplish this, corium
samples were prepared using uranium dioxide (UO2), Nd, Sr and Ce (as a surrogate
for plutonium). These samples were divided into two groups: one to be leached
and one to be pressed into pellets. The surface area of the powders was measured,
and the water in which the powders leached was analysed using an ICP-MS. The
pellets were analysed using a SEM and XRD. XRD, SEM, and ICP-MS indicated
that corium was successfully synthesised; however, the leachate concentration was
too high for the ICP-MS to measure, indicating significant elemental release, but
this meant that the exact concentration of the element could not be quantified. In
conclusion, corium was synthesised and leached, but further work is needed to accurately quantify the leaching behaviour and elemental release.
In addition, a leaching apparatus was developed to facilitate low-cost, accessible
leaching experiments. This design, which is stackable and allows for simultaneous
parallel testing, was made using CAD and can be 3D-printed, which removes reliance
on specialised or expensive laboratory equipment
A Study Towards the Removal of the Relay Rod in Volvo Trucks: Exploring Solutions for Steering the Second Front Axle in a Future Double Front Axle Arrangement by Wire
No full textThe study explores the removal of the relay rod in Volvo Trucks, specifically focusing on the front axle arrangement with two nondriven steered front axles (FAA20). The relay rod, a mechanical link between the axles in the steering system, presents structural and manufacturing challenges. The investigation aims to replace the currently in production mechanical system with a system that includes individual steering of the first and second axle. The report details the methodology used, including concept generation, risk analysis, and simulations. Seven concepts were evaluated, with Concept 7 identified as the optimal solution. The study concludes a proposed layout of the steering system and a ratio between the first and the second front axle to facilitate efficient steering