Examensarbeten för masterexamen


Senast publicerade

Visar 1 - 5 av 1715
  • Post
    An Exploration of Explainability for Internal Stakeholders: A Qualitative Study
    (2024) Admekie, Tedla Bayou; Pravin Karthick, Sushmitha; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Penzenstadler, Birgit; Knauss, Eric
    AI applications are becoming increasingly prevalent across various domains and industries. However, the challenge of comprehending the inner workings of ML/AI systems extends beyond end users and significantly impacts AI/ML developers and testers. This research investigates Explainable Artificial Intelligence (XAI) aspects and challenges concerning internal stakeholders. We conducted a qualitative interview study involving experts and researchers specializing in explainability and engineering AI-based systems. Our findings emphasize the importance of explainability exclusively for internal stakeholders, an aspect that has received limited attention in previous research. We identified research gaps in the following areas: the lack of exploration into how explainability can enhance AI/ML model testing, the need for further investigation into existing XAI evaluation metrics that align with diverse internal stakeholder needs, and the knowledge gap surrounding the concept of XAI and its integration into existing processes. Additionally, we present the challenges that internal stakeholders encounter when incorporating explainability features. The key results show that explainability positively impacts testability, as it can serve as a tool for guiding the test process. There are several noticeable benefits of explainability methods (XAI) for both developers and testers such as explainability aids in debugging the model output, which is an essential aspect of error analysis, and in detecting potential biases in the data. Other benefits are discussed in this paper. Furthermore, there is a need for an accepted set of standardized metrics to assess the trustworthiness of explainability, which would evaluate the effectiveness of the explanations themselves. Our study offers foundational work for future research and underscores critical research gaps. The ability to design explainability for internal stakeholders holds the potential to facilitate the development of complex, AI/ML systems.
  • Post
    Exploring Automated Early Problem Identification Based on Diagnostic Trouble Codes
    (2024) Forsman, Mathias; Yang, Yihan; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Strüber, Daniel; Heyn, Hans-Martin
    In the current automotive industry, problem identification is a reactive process. It starts when the customer experiences a vehicle problem and goes to the workshop. Subsequently, all the problem-related data will be collected from the workshop and forwarded to the vehicle manufacturer. After that, the engineers will start looking into the problem and figuring out the root cause with the cooperation from internal and external departments. It is a case-sensitive procedure and each unforeseen factor may further prolong the process and affect customer satisfaction. This study cooperates with Volvo Cars to explore the possibility of providing a proactive data-driven insight into the problem identification process in the automotive system using the Diagnostic Troubleshooting Code (DTC). The purpose is to identify the most affected group before the problem scales and affects most of the customers. This study involves two case studies and one laboratory experiment. The first-round case study helps to gain a better understanding of the current problem identification process. Also, some challenges and limitations encountered in this process have been identified. Other than these, five cases, including three different car parts: the car part A unit, the climatization system, and an add-on system, have been collected to conduct the following laboratory experiment. In total, four models are constructed and refined using several basic and machine learning techniques, including Group-by, Linear Regression, and K-means Clustering. This process evaluates different models’ capabilities to provide early warnings and the corresponding correctness. It further assesses each technique’s strengths and limitations in predicting the most affected group. The last case study serves as an evaluation action to receive feedback from the industrial experts about model performance and discuss the potential solution to integrate the model construction into the current workflow. In the end, a data-driven approach has been proposed and comprehensively described. The influencing factors, advantages, and limitations of the research have also been discussed, leading to various interesting directions for future research.
  • Post
    Digital Audio Interface Jitter
    (2024) SINKKONEN, FREDRIK; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Fjeld, Morten; Fjeld, Morten
    Jitter is the short-term deviation of a digital signal from its ideal position in time. Some common issues know to produce jitter in currently used digital audio interface formats were examined and multiple implementations of a Universal Serial Bus (USB) audio interface were designed with the intention of creating a device free from interface jitter. Using the three standardized clock synchronization mechanisms in the USB protocol for isochronous transmissions and a selection of suitable clock sources, USB audio class devices were created for which jitter measurements then were performed. The results were compared with jitter audibility thresholds from three studies containing listening tests. While all implementations were functionally acceptable, their jitter results did differ. For the two isochronous synchronization modes of USB that require a continuously adjustable clock source on the receiving side of the interface the jitter issue consists of two parts. Periodic adjustments of the clock signal are in itself a source of jitter and the way in which an adjustable clock source is constructed is another. The initial core idea was that a USB audio interface using isochronous transfers coupled with the asynchronous clock synchronization mode and a fixed frequency clock source would be able to provide an interface in which no additional jitter on top of the inherent jitter level of the source clock would be added by the transfer of data over the interface. The two fixed frequency clocks that were used did however not perform any better than the results of the best adjustable clock source and when they were attached to the test system their jitter levels increased even further. Analysis of the jitter measurements point in the direction of asynchronous mode being preferable for lowest possible jitter levels but the results are not completely unambiguous and jitter levels below the lowest recorded hearing thresholds were also achieved with one of the other synchronization modes for isochronous USB transfers.
  • Post
    Leveraging Data Augmentation for Better Named Entity Recognition in Low-Resource Settings
    (2024) Björnerud, Philip; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Bernardy, Jean-Philippe; Dannélls, Dana; Kokkinakis, Dimitrios
    This thesis investigates the challenges in the field of Natural Language Processing (NLP), with a focus on Named Entity Recognition (NER), a subtask within NLP that involves classifying entities. Addressing the issue of data scarcity, which is particularly critical in non-English languages like Swedish, this study investigates various data augmentation methods by fine-tuning the transformer-based model, KB-BERT. The datasets are simulated as low-resource settings, drawing inspiration from the study X Dai and H Adel (2020) [1] work, using three sets of training data containing 50, 150, and 500 instances respectively. The thesis also explores whether a newly developed state-of-the-art data augmentation method can outperform other data augmentation methods in enhancing an NLP model, centering on three data augmentation methods: Synonym replacement, Mention replacement, and AugGPT, the last being a state-of-the-art method. The findings of this study highlight that synonym replacement emerged as the most effective data augmentation method across various low-resource settings, achieving the highest F1-score increase in all scenarios. AugGPT achieved the second highest average F1-score, while mention replacement achieved the lowest across the tested settings.
  • Post
    Layout Syntax Support in the BNF Converter
    (2023) Burreau, Beata; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Ranta, Aarne; Abel, Andreas
    Many programming languages, such as Haskell and Python, use layout as part of their syntax. We can expect future programming languages to also be layoutsensitive. Therefore, the toolchains for implementing programming languages must support layout-sensitive languages. This thesis presents a declarative approach to describing layout-sensitive languages and parsing programs written in them. We reserve the terminals newline, indent, and dedent for describing layout syntax in BNF grammar and provide an algorithm for representing the layout of a program with these terminals, before parsing it. By verbalising layout syntax this way, mainstream parser generators, and their parsing algorithms, can be used. This approach is successfully implemented in BNF Converter (BNFC), a tool that generates a compiler front-end from a context-free grammar in Labelled BNF (LBNF) form. With a special kind of LBNF rule, called pragma, it is possible to declare global layout syntax rules, such as the offside rule, which affects the insertion of layout terminals by the aforementioned algorithm. The reserved terminals and the pragmas can together describe popular layout syntax. Furthermore, both purely layout-sensitive languages and those mixing layoutsensitive and insensitive syntax are describable in LBNF.