From: 2024-06-12 10:15 to 16:00
Place: E:2116 and E:4130 (Lucas)
Contact: birger [dot] swahn [at] cs [dot] lth [dot] se

Three Computer Science MSc theses to be presented on 12 June

Wednesday, 12 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.

10:15-11:00 in E:2116 N.B. Presentation added

Presenter: Ida Levison
Title: Accelerating indirect array accesses using slice compiler optimization
Examiner: Flavius Gruian
Supervisor: Jonas Skeppstedt (LTH)

This master thesis presents an implementation of the slice compiler optimization, targeting nested array accesses in loops, and based on a producer-consumer approach integrated with a hardware queue. The compiler optimization is evaluated on the lmpcc compiler and asim simulator, and compared to other prefetch techniques. Evaluation on an application characterized by nested array accesses showed an overall speedup by 1.13× and the targeted loop was halved in execution time. The optimization also achieved a decrease of L1 cache misses. The implementation proves that the approach of slicing an application into a producer and consumer thread, combined with a hardware queue produces an improvement in execution time and a decrease in cache misses.

Link to popular science summary: To be uploaded

15:15-16:00 in E:2116 N.B. Change of room

Presenters: Abdulrahman Husari, Sepehr Taherpour
Title: Enhanced Techniques for Detecting Performance Abnormalities in Software Quality Assurance Processes
Examiner: Björn Regnell
Supervisors: Per Runeson (LTH), Filip Nilsson & Jacob Lindquist (Axis Communications AB)

Analyzing performance test data for network camera devices presents significant challenges due to the vast amount of data generated. The lack of efficient methods for detecting performance abnormalities necessitates studying this area. The analysis often requires manual work by quality engineers, which becomes increasingly complicated with the expansion of test variety and product variants, creating a bottleneck in the development and Quality Assurance (QA) processes. This study aimed to develop effective and adaptive techniques for detecting abnormal behavior in performance test results. We investigated approaches based on statistical thresholds, unsupervised learning, and supervised learning. In addition, we implemented a feedback mechanism used for improving the detection over time. The proposed solution was designed to assist quality engineers and reduce their workload. The efficiency of the solution was evaluated using machine learning metrics and validated through live testing sessions. The results indicated that supervised learning was the most suitable approach in this study, with the RandomForest model achieving an average AUC score of 0.94 and an average recall value of 0.98 across three selected products. The evaluation of the proposed solution demonstrated the effectiveness of the detection in the QA process at the case company.

Link to popular science summary: To be uploaded

15:15-16:00 in E:4130 (Lucas)

Presenter: Elna Seyer
Title: Mortality prediction using federated learning framework NVIDIA FLARE
Examiner: Elin A. Topp
Supervisors: Maj Stenmark (LTH), Lena Mondrejevski (Getinge)

The Intensive Care Unit handles the most critical patients, necessitating effective triage processes to predict mortality risk and improve patient outcomes. Traditional centralized machine learning approaches, which store and train data locally, face limitations in data collection and disease detection. Federated Learning offers a solution by enabling secure, collaborative model training without data centralization. This thesis explores the practical implementation of FL using NVIDIA FLARE, a versatile framework developed by NVIDIA, within a lab environment for ICU mortality prediction. Despite the popularity of FL in medical data contexts, detailed insights into the NVIDIA FLARE application are limited. This research investigates the performance of a FL system compared to both centralized and local systems for mortality prediction using the MIMIC-III dataset. The findings suggest that FL models with NVIDIA FLARE can achieve competitive performance.

Link to popular science summary: To be uploaded