Three Computer Science MSc theses to be presented on 5 June

Wednesday, 5 June there will be three master thesis presentations in Computer Science at Lund University, Faculty of Engineering.

The presentations will take place in E:2116 and E:4130 (Lucas). See information for each presentation.

Note to potential opponents: Register as an opponent to the presentation of your choice by sending an email to the examiner for that presentation (firstname.lastname@cs.lth.se). Do not forget to specify the presentation you register for! Note that the number of opponents may be limited (often to two), so you might be forced to choose another presentation if you register too late. Registrations are individual, just as the oppositions are! More instructions are found on this page.

10:15-15:00 in E:2116

Presenters: Malin Åstrand, Felix Apell Skjutar
Title: MicroPython Integration for Radar Specific Application: Is it Worth it?
Examiner: Sven Robertz
Supervisors: Jonas Skeppstedt (LTH), Anders Buhl (Acconeer AB)

In contemporary embedded development, speed and ease of prototyping are crucial. However, the prevalent use of C in this domain not only poses challenges for those less acquainted with the language but also increases development time. Python, renowned for its readability and ease-of-use, emerges as a compelling alternative. With libraries like NumPy and SciPy at its disposal, Python offers a rich ecosystem that supports efficient coding practices. MicroPython further extends this accessibility to the embedded realm, enabling swift development cycles similarly to traditional Python. This thesis investigates the feasibility of implementing a radar specific application in MicroPython, lowering the threshold of going from idea to prototype. While previous findings indicate MicroPython's sluggishness when performing computationally heavy tasks, we explore a real-world application leveraging Ulab and a radar API written in C. Our study contrasts the performance of MicroPython against its pure C counterpart, implementing a surface velocity algorithm. Our approach, with optimized algorithm design and data handling strategies, showcases a 40\% improvement in performance compared to the naive implementation, albeit still trailing C by a factor of three. Our findings underscore the significance of thoughtful algorithm design and data management in mitigating MicroPython's performance disparity with C. Finally, we present a practical guide to aid decision-making regarding MicroPython adoption in embedded applications.

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/240605_10AstrandApellSkjutar.pdf

15:15-16:00 in E:2116

Presenter: Fredrik Hessner
Title: Memory-Safe 5G Software Using the CHERI Hardware Architecture
Examiner: Sven Robertz
Supervisors: Jonas Skeppstedt (LTH), Håkan Englund (Ericsson)

As the problem of memory-safety continues to pose a significant threat to the software ecosystem, software developers need to consider strategies to address these vulnerabilities. For high-performing applications built on the C/C++ programming languages, solutions have to be adaptable without requiring too much effort and without compromising too much on performance. In this thesis, high-performing 5G software is evaluated using a hardware architecture designed to significantly improve memory-safety. The research was based on two metrics: How easy it is to port the software and what the overhead of the memory-safe architecture is compared to the baseline. The results show that only around 1% of the total lines of code had to be re-written for the program to successfully run. The overhead of the two measurements of the benchmark were 38% and 53% compared to baseline, and are believed to come from addressable architectural limitations.

Link to popular science summary: To be uploaded

15:15-16:00 in E:4130 (Lucas) (Presentation was added later)

Presenters: Oliver Lövström, Joel Nygren
Title: Real-Time Edge AI Hand Detection for Drone Controls
Examiner: Michael Doggett
Supervisor: Flavius Gruian (LTH)

In recent years, computing and artificial intelligence advancements have enabled complex computing on edge devices. Edge computing can be advantageous for inference time in real-time applications while removing the need for reliable connectivity. We approach a specific control problem to show potential challenges when deploying machine learning models to an edge device and how these challenges can be overcome. The goal is to make a  Bitcraze Crazyflie, a nano-sized drone equipped with a camera that can follow a hand. We conclude that most of the challenges stem from the unique properties of the onboard GAP8 processor, which makes it ultra-low power. These properties restrict the convolutional neural networks that can be deployed. We show how these challenges can be overcome by using suitable neural network architectures and training practices to achieve a tradeoff between precision and inference latency acceptable for our real-time application.

Link to popular science summary: To be uploaded