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
Post
Quantifying the greenhouse gas reduction potential of battery electric cars with integrated photovoltaics
(2024) Abedi, Sahar; Simangan Liggayu, Francis Jessy; Chalmers tekniska högskola / Institutionen för rymd-, geo- och miljövetenskap; Chalmers University of Technology / Department of Space, Earth and Environment; Johansson, Daniel; Morfeldt, Johannes
Post
Developing a one-to-many generation LLM for diverse, accurate and efficient retrosynthesis
(2024) Li, Junyong; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Engkvist, Ola; Johansson, Richard
One of the most common applications of deep learning for cheminformatics is retrosynthesis,
which is a task of predicting reactants given a chemical product. After
transformer was invented, it has been widely used for retrosynthesis. Chemformer
is a transformer-based model, which was pre-trained using SMILES of chemical
molecules first and can be fine-tuned for retrosynthesis. The model achieves stateof-
the-art performance on this task. Retrosynthesis task expects multiple predictions
of reactants. Chemformer uses beam search or multinomial search to get multiple
predictions, which results in a lack of diversity, accuracy and efficiency of the model.
In this project, the sphere projection strategy, which is a one-to-many generation
strategy, was applied to Chemformer to enable it to generate multiple predictions.
The sphere projection achieves one-to-many generation by introducing variations of
source embedding of encoder and combining those variations with a single-prediction
sampler, such as greedy search and multinomial search (multinomial size = 1). By
comparing the modified Chemformer with sphere projection strategy to the baseline
Chemformer, it was shown that the strategy can improve diversity, accuracy and
efficiency by 197%, 7% and 4% respectively for beam search, and 101%, 2% and
17% respectively for multinomial search.
Post
AI and ML for Software Product Management: A Framework for Emerging Challenges
(2024) JÖNMARK, JULIA; SÖDERSTRÖM, HANNA; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Knauss, Eric; Bosch, Jan
In the rapidly evolving landscape of software product management (SPM), the integration
of artificial intelligence (AI) and machine learning (ML) presents both
unprecedented opportunities and significant challenges. This thesis investigates the
impact of AI and ML on SPM practices and develops a comprehensive framework
tailored to address the emerging needs of this dynamic field. Utilizing a mixedmethod
approach, the study first conducts a systematic literature review to identify
the current utilization and challenges of AI and ML within SPM. This is followed by
empirical data collection through interviews with professionals in the field, ensuring
a robust foundation for framework development. The research findings reveal that
while AI and ML can significantly enhance decision-making and efficiency in SPM,
they also introduce complexities related to integration, ethics, and management. In
response, this thesis proposes a novel SPM framework that incorporates how SPM
should use AI and ML components and tools effectively, focusing on enhancing SPM
and aligning with digital transformation and digitalization goals. The framework is
validated through a workshop and an interview with experts in the field. This study
aims to bridge a crucial gap in academic literature and to also offer practical insights
for individuals and organizations aiming to leverage AI and ML for enhanced
SPM strategies, ensuring both competitive advantage and alignment with evolving
technological landscapes.
Post
Cybersecurity requirements identification using LLMs - A design science study
(2024) Linde, Filip; Sanner, Oscar; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Horkoff, Jennifer; Fotrousi, Farnaz
Context: Threat analysis and risk assessment (TARA) is a widely used approach
for conducting cybersecurity analysis in the automotive industry. The process is
initiated early in the development process and continuously iterated.
Problems: Automotive systems continue to rely more on software. Additionally,
the National Vulnerability Database (NVD) show that more vulnerabilities are found
each year. As a result, much time has to be spent continuously ensuring that systems
have updated TARA analysis.
Method: We designed a Large Language Model (LLM) based artifact to help security
engineers by automatically identifying attack paths and security requirements.
The artifact achieved this via a combination of prompt engineering and grounding
in both the Common Vulnerabilities and Exposures (CVE) database, and the Automotive
Information Sharing and Analysis Center (Automotive-ISAC) Automotive
Threat Matrix (ATM).
Result: The artifact could define security requirements which met the expected
standards of practitioners and were correct based on the attacks they were generated
to mitigate. However, challenges were identified in the generation of attacks
paths, where the generated output was less consistent in how well it met expectations.
Experts perceived it to be able to generate appropriate requirements for
an initial TARA analysis, however future work is needed to determine how more
complex paths and requirements could be identified automatically.
Post
Parallel and Distributed Motif Discovery in Temporal Networks - A feasibility study applying thread parallelism and community structure
(2024) Marton, Stefan; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Damaschke, Peter; Hassan, Ahmed Ali-Eldin
Temporal networks are used to model complex systems in topics such as epidemiology,
finance, and computer networks. Motifs are subgraphs of a directed graph
that are representative of the structure of that particular graph. Motifs have been
extended to temporal networks. Motif discovery is a computationally hard problem;
in fact sub-problems are NP-hard problems. In this thesis, we explore state-of-theart
temporal network motif discovery algorithms, and how they can be parallelized
on a multi-threaded system and distributed across multiple systems. We select Kovanen’s
definition of temporal network motif. We implement a simple approach to
thread parallelism to demonstrate the potential for parallelism of the algorithm,
and find that the parallelizable proportion p > 0.89, which implies great potential
for parallelism. We utilize community structure of the graph the temporal network
represents to divide the network into work packages for distributed computation. In
doing so, we encounter and report numerous challenges in distribution of the exact
solution approach.