̽ѡ

Norwegian version of this page

dScience Lunch Seminar: PhD Special

The PhD special event highlights a diverse spectrum of research topics, ranging from statistical methods for change detection and methane emission analysis using drones, to the role of electric vehicles in renewable energy integration, trustworthy AI through explainability and fairness, and advances in data-constrained language modeling.

Join us at the Science Library on May 8, 2025, from 12:00 PM to 1:15 PM.

Image may contain: Azure, Organism, Font, Aqua, Electric blue.

Presentations

Change and Anomaly Detection: Challenges and Statistical Methods
Per August Jarval Moen

Change and anomaly detection are active areas of research at the University of ̽ѡ, as well as within the broader fields of statistics and machine learning. This presentation will provide a brief overview of key challenges encountered in these domains, as well as showcasing some recent advancements and practical applications.

Navigating methane plumes: A journey from drone observations to flux estimates
Alouette van Hove

In this presentation, we will discuss how drone observations and atmospheric models can be combined to estimate methane fluxes within a Bayesian framework. Additionally, we will examine the potential of reinforcement learning to improve drone sampling strategies for more informative data collection.

Driving towards net-zero: Can electric vehicles support the integration of variable renewable energy?
Tobias Verheugen Hvidsten

With the transition to electric vehicles vast amounts of batteries are distributed in the energy system. This could become a valuable resource of flexibility as electricity demand increase and new generation from variable renewable sources is added. This presentation will address the question if electric vehicles could provide flexibility to balance supply and demand in a future net-zero electricity system, with a focus on Norway.

More Towards Trustworthy AI: Integrative Approaches to Explainability, Accuracy, Privacy, and Fairness
Poushali Sengupta

This presentation explores a comprehensive study of trustworthy AI, focusing on four key themes: enhancing model explainability without compromising accuracy, securing private data while preserving utility, and ensuring fairness across diverse applications. These themes are crucial in developing AI systems that are not only technically proficient but also ethically sound and socially responsible.

Furthermore, we extend the ExCIR framework to models with dependent features, demonstrating its effectiveness in simulated energy data environments and ensuring model fidelity even in resource-constrained settings. We also discuss the practical application of the Hierarchical XAI (HXAI) framework, which is designed for Iot-based smart home energy management. This framework enhances data privacy and utility, providing the detailed insights individual households and energy providers need without compromising their privacy.

Lastly, I will address our work on vehicular traffic systems, which focuses on balancing privacy, utility, and fairness in handling geographic data. Here, we employ differential privacy techniques to safeguard sensitive data while ensuring equitable traffic management across regions. These studies not only provide robust methodologies for developing AI systems but also underscore the importance of ethical considerations in our work, fostering trust and broader acceptance in critical sectors.

Data-constrained language modeling
David Samuel

Large language models have taken the world of NLP by storm, substantially outperforming older solutions on a large number of tasks. However, this comes at a cost — they are notoriously “data hungry”, requiring training on trillions of words to achieve such performance. Getting this data in English is quite simple, but it is virtually impossible for languages such as Norwegian. This presentation will investigate what happens when we constrain the amount of data a language model is trained on, and what can be done to train well-performing language models in this constrained scenario.

Program

12:00 – Doors open and lunch is served

12:15 – Presentations by PhD Candidates Alouette van Hove (Department of Geosciences), David Samuel (Department of Informatics), Per August Jarval Moen (Department of Mathematics), Poushali Sengupta (Department of Informatics) and Tobias Verheugen Hvidsten (Department of Technology Systems)

13:15 – Mingling (and goodbye)

To participate, please fill out the registration form. This way, we will not be short on food and drinks! (Registration is not binding and you are welcome to join us anyway!)

Follow the event online here

 

About the seminar series

Once a month, dScience will invite you to join us for lunch and professional talks at the Science Library. In addition to these, we will serve lunch in our lounge in Kristine Bonnevies house every Thursday. Due to limited space (40 people), this will be first come, first served. .

Our lounge can also be booked by PhDs and Postdocs on a regular basis, whether it is for a meeting or just to hang out – we have fresh coffee all day long!

Organizer

dScience and
Tags: dscience, lunch seminar, PhD, postdoc
Published Apr. 14, 2025 11:27 AM - Last modified Apr. 30, 2025 1:58 PM