Skip to main content




Three more CS MSc Thesis Presentations 4 April 2024

From: 2024-04-04 11:15 to 15:00 Föreläsning

Three additional Computer Science MSc theses to be presented on 4 April

Thursday, 4 April there will be three more master thesis presentations in Computer Science at Lund University, Faculty of Engineering. See also this page for one more presentation:

The presentations will take place in E:4130 (Lucas).

Note to potential opponents: Register as an opponent to the presentation of your choice by sending an email to the examiner for that presentation ( 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.

11:15-12:00 in E:4130 (Lucas)

Presenters: Dalia Saleh Abbas, Mustafa Al-Jailawi
Title: Data-Driven Strategies for Improving Email Campaign Engagement : A Send Time Optimization Approach
Examiner: Jacek Malec
Supervisors: Pontus Giselsson (LTH), Fredrik Danielsson (IKEA)

Email direct marketing is one of the cornerstone pillars of digital marketing playing a pivotal role in engaging target audiences and achieving campaign success.

This collaborative research between IKEA and Lund University explores optimizing and tailoring email send times to better correspond with customers’ behavioral patterns. The study delves into the realm of send time optimization (STO), a methodology marketers adopt to enhance the impact of email campaigns by increasing customer engagement. Furthermore, this study builds upon prior research by incorporating web engagements alongside Recency, Frequency, and Monetary (RFM) values as additional features. Web engagements not only facilitate tailored send times to a wider range of customers but also enhance the scope of data collected about each customer.

Through the use of various regression algorithms, the model was trained on data from 90 days of customer interactions. The evaluation results indicated XGBoost to be the best-performing algorithm on the dataset, leading to its selection for A/B testing in a production environment. The A/B test, conducted over two weeks, was designed to benchmark the newly proposed model against the current model at IKEA. The comparative analysis evaluates the efficacy of each model by examining the click-through, open, and opt-out rates. The findings suggest marginally superior performance exhibited by the proposed model compared to IKEA's current model. Notably, both models outperform fallback send times, indicating that optimizing send times positively influences email engagement.

Link to popular science summary:

13:15-14:00 in E:4130 (Lucas)

Presenters: Samil Kapetanovic, Mohammad Raja
Title: Enhancing System Documentation Accessibility through Question-Answering
Examiner: Jacek Malec
Supervisors: Marcus Klang (LTH), Eltayeb Bayomi (Consafe Logistics AB)

In recent years, the challenge of quickly finding answers to questions has increased, with individuals often spending a lot of time searching for relevant information. This has led to an increasing interest in developing efficient systems to enhance accessibility. This thesis investigates the usage of open-retrieval question-answering systems and their potential to enhance the accessibility of system documentation. Through a comprehensive exploration, we investigate various open-retrieval question-answering configurations, including sparse and dense retrievers, extractive and generative readers, and additional experiments across three different datasets.

We present the system performances on both benchmark datasets and a private dataset created in collaboration with Consafe Logistics AB. These systems exhibit strong performance across the benchmark datasets. In ConsafeQA, we developed a system configuration that achieves an F1 score of up to 79.68 and another configuration that reaches an F1 score of up to 81.96 on benchmark datasets. Our results also indicate that different combinations of components and hyperparameters excel in different contexts. Therefore, no fixed pipeline consistently achieves optimal performance on all datasets, emphasizing the significance of tuning the system components and parameters according to the domain.

Link to popular science summary: To be uploaded

14:15-15:00 in E:4130 (Lucas)

Presenter: Sulthan Suresh Fazeela
Title: Skill-Based Multi-Objective Reinforcement Learning for Dual-Arm Robotic Assembly
Examiner: Jacek Malec
Supervisors: Matthias Mayr (LTH), Faseeh Ahmad (LTH)

Cooperative robotic assembly tasks are complex industrial operations that require multiple robotic arms to work together to assemble parts or components. These tasks often involve multiple conflicting objectives, such as accuracy, speed, and safety which makes it challenging to optimize. In this thesis, we propose a Skill-based Multi-Objective Reinforcement Learning (MORL) approach to train dual-arm robots for cooperative assembly tasks, specifically focusing on the peg-in-block insertion task. Furthermore, we demonstrate a transfer learning approach using prior knowledge injection to accelerate the learning process for task variations. We leverage SkiROS2 platform for skill-based programming and SkiREIL framework for multi-objective reinforcement learning. We conduct our experiments in a simulated environment using DART simulator, but the presented approach and findings can be applied to real-world scenarios as well. Our results demonstrate that skill-based MORL is a powerful approach for training industrial robots in cooperative assembly tasks.

Link to popular science summary:


Om händelsen
From: 2024-04-04 11:15 to 15:00

E:4130 (Lucas)

birger [dot] swahn [at] cs [dot] lth [dot] se