Examensarbeten för masterexamen

Browse

Senast publicerade

Visar 1 - 5 av 1606
  • Post
    Challenges in Specifying Safety-Critical Systems with AI-Components
    (2023) Malleswaran, Iswarya; Dinakaran, Shruthi; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Feldt, Robert; Knauss, Eric; Heyn, Hans-Martin
    Safety is an important feature in automotive industry. Safety critical system such as Advanced Driver Assistance System (ADAS) and Autonomous Driving (AD) follows certain processes and procedures in order to perform the desired function safely. Many ADAS applications relies significantly on Machine Learning and data needed to perform the desired function. Data quality, more specifically the information content of the data, can highly impact the effectiveness of the model and its function. It is important to select the right data to train the model. Furthermore, monitoring the safety critical system during runtime helps to understand the data which the model receives. Such information helps further to create and update machine model. There are uncertainties and challenges in defining the requirements for finding the right information content of the data such that the desired and a safe behaviour of the system is ensured. This case study investigates and explores the challenges experienced in creating the requirements for proper selection of training data. It also analyzes challenges when specifying runtime monitoring and the relation between requirements on runtime monitoring and the training data. This case study follows the approach of qualitative and exploratory research. The analysis for this study is based on ten interviews with experts from different field. Moreover, a workshop has been conducted with academic and industry experts to validate the results from our interview analysis. Based on the qualitative analysis of data, the case study shows that there is lack of clarity in defining requirements, lack of communication, no clear scope of design domain, missing guidelines for data selection and safety requirements, and a lack of metrics for defining the right variety of data and runtime monitors. The results outline challenges experienced by practitioners when specifying data and defining requirements for runtime monitors for safety critical systems.
  • Post
    An Empirical Survey of Bandits in an Industrial Recommender System Setting
    (2023) Brandby, Johan; Schwarz, Tobias; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Dubhashi, Devdatt; Jorge, Emilio
    In this thesis, the effects of incorporating unstructured data—images in the wild—in contextual multi-armed bandits are investigated, when used within a recommender system setting, which focuses on picture-based content suggestion. The idea is to employ image features, extracted by a pre-trained convolutional neural network, and study the resulting bandit behaviors when including respective excluding this information in the typical context creation, which normally relies on structured data sources—such as metadata. The evaluation is made both online, through A/B-testing enabled by the industrial partner YouPic AB, and offline, effectuated by a simulation pipeline that models the online counterpart. The results are compiled as a survey, covering a selection of contextual bandit algorithms, highlighting the differences brought by the unstructured data. The offline result points towards that if the contextual bandit utilizes a joint or hybrid action-value function, with respect to the parameterization, the addition of the image vectors can significantly outperform the instances without it; however, if a disjoint model is instead employed, no noticeable change is observed. In comparison, those from the online trials can be interpreted as supporting the inclusion of convolutional features, but due to meager and unbalanced sample sizes, the outcomes are deemed inconclusive. To summarize, though there is support for incorporating unstructured data, given that the action-value function is joint or hybrid, the online experiments gave too little evidence for any trustworthy findings; in other words, the question is still partially open.
  • Post
    Methods for detecting echo chambers in social media networks
    (2023) Bonafilia, Brian; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Johansson, Moa; Bruinsma, Sebastianus
    This thesis presents an approach to using Natural Language Processing to detect echo chambers in social media networks and to find identifying terms for those echo chambers. A dataset consisting of posts and user information from the micro-blogging service Twitter related to Sweden’s application to join the North Atlantic Treaty Organization was collected for the year leading up to the Swedish national election of 2022. Tight-knit communities of users on the platform were extracted using the Infomap and Leiden Algorithms based on user connections and interactions. From each community found using these methods, the corpus composed of the text postings of the users in that community was used to train a Word2Vec model to recover vector word embeddings for key words related to the subject of the discussion. Semantic change was quantified by assessing the differences in cosine similarity between pairs of words over time and between communities. Changes in the use of terms related to the subject over time were observed, but patterns representing possible echo chambers arose only with the aid of manual annotation of user positions on the issue. Conclusions could not be drawn about how successful the method is from the results alone, as evidence suggests that the issue was insufficiently polarized to generate strong echo chambers.
  • Post
    Autonomous Drug Design with Reinforcement Learning
    (2023) Edvinsson, Filip; Jonsson, Victor; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Schliep, Alexander; Haghir Chehreghani, Morteza
    The drug design process is currently one of manual trial and error, where potential drug candidates are proposed by chemists, synthesized in laboratories, and then tested and analyzed for properties and efficacy. This process, also called the Design- Make-Test-Analyze (DMTA) cycle, is repeated until a satisfying drug candidate is reached. Statistical models to sample the chemical space and generate potential molecules, combined with automated laboratories and machine learning allows for the automatization of the DMTA-cycle. However, there is still a need for improvement and this is where our project comes in. One way to improve the automatization of the DMTA-cycle is to reduce the number of cycles needed, and our aim was to achieve this by improving the selection of compounds. To do this, we developed two deep reinforcement learning algorithms, Deep-Q Network (DQN) and Double Deep-Q Network (DDQN), and compared these to two baseline selection algorithms. This approach was chosen as it translates well into the drug development field. Reinforcement learning in drug discovery works by exploring the proposed molecules to find potential candidates and selecting the most promising ones based on molecular similarity to some predetermined properties. Ultimately, the project was unsuccessful. The baseline selection algorithms using random and greedy selection approaches proved more efficient and accurate than the two algorithms we developed. The involvement of reinforcement learning agents when selecting compounds seemed to cloud the generative model’s understanding of what constitutes a good molecule, and thereby reduced the quality of proposed molecules for both the implemented selection algorithms. However, we found that the DQN algorithm shows some signs of promise and can, with some fine-tuning, potentially be brought up to par with the baseline selection algorithms, and perhaps even surpass them.
  • Post
    Rehab Rush
    (2023) Möller, Mathilda; Kristensson, Maja; Chalmers tekniska högskola / Institutionen för data och informationsteknik; Chalmers University of Technology / Department of Computer Science and Engineering; Björk, Staffan; Belford, Pauline
    In this master’s thesis, a high-fidelity prototype of a body controlled mobile phone game was developed, with the purpose of investigating if a mobile phone game can support and facilitate patients when executing their physiotherapeutic exercises. The targeted user group for the game was young women (16-20 years old) with ACL-injuries. The research was based on the question: How can a body controlled mobile phone game be designed to help young women with ACL-injuries do their exercises in a correct and rewarding way?. To address this research question, an iterative development process was executed, which involved conducting a qualitative study to evaluate the idea and prototype with experienced physiotherapists. Seven guidelines elicited from the research, development process and qualitative study, were formulated to provide inspiration and support for further development of this game or other mobile games for physiotherapeutic purposes.