Winter School

NLDL 2025 Winter School

The NLDL Winter School consists of tutorials by experts in the field and is co-hosted by NORA as part of the NORA Research School.

For registration, please use https://www.nldl.org/attend/registration


Getting formal ECTS Credits: UiT The Arctic University of Norway will award 5 ECTS for the Winter School to students who register formally for the course (the number of spots is limited to 40 students).  


EUGLOH Mobility Scholarships: There are a total of 16 mobility scholarships exclusively for EUGLOH students from partner institutions under the EUGLOH alliance.  For more details, please see https://www.eugloh.eu/courses-trainings/activities/northern-lights-deep-learning-winter-school-2025/  Deadline: 6th November 2024.

Program

Venue: Auditorium 1 (Teorifagbygget Hus 1, Floor U1) at the UiT Campus

Monday, January 6th:

08:00 - 08:30 Registration

08:30 - 08:45 Welcome

08:45 - 11:30 Tutorial 1: Aleatoric and Epistemic Uncertainty in Statistics and Machine Learning 

11:30 - 12:30 Lunch

12:30 - 13:30 Winter School Poster Session (Poster Session Chair: Catuscia Palamidessi)

13:30 - 16:30 Tutorial 2: Structure-preserving machine learning for physical systems

Tuesday-Thursday, January 7-9th: 

Main Conference Program

Friday, January 10th:

09:00 - 09:15 Welcome

09:15 - 12:00 Tutorial 3: Responsible and Explainable Artificial Intelligence

12:00 - 13:00 Lunch 

13:00 - 16:00 Tutorial 4: Large Language Models Under the Hood

16:00 - 16:15 Closing

Winter School Tutorial and Speakers


The NLDL Winter School will consist of tutorials from leading experts in the machine learning field. Below we present each tutorial, followed by the tutorial speakers. 

Tutorial 1: Aleatoric and Epistemic Uncertainty in Statistics and Machine Learning

Without any doubt, the notion of uncertainty is of major importance in machine learning and constitutes a key element of modern machine learning methodology. In recent years, it has gained in importance due to the increasing relevance of machine learning for practical applications, many of which are coming with safety requirements. In this regard, new problems and challenges have been identified by machine learning scholars, many of which call for novel methodological developments. Indeed, while uncertainty has a long tradition in statistics, and a broad range of useful concepts for representing and quantifying uncertainty have been developed on the basis of probability theory, recent research has gone beyond traditional approaches and also leverages more general formalism and uncertainty calculi.

This tutorial aims to provide an overview of uncertainty quantification in machine learning, a topic that has received increasing attention in the recent past. Starting with a recapitulation of classical statistical concepts, we specifically focus on novel approaches for distinguishing and representing so-called aleatoric and epistemic uncertainty. By the end of the tutorial, attendees will have a comprehensive understanding of the fundamental concepts and recent advances in this field.

The tutorial will be presented by Willem Waegeman from Ghent University.

Tutorial 2: Structure-preserving machine learning for physical systems

Physics-informed machine learning is an innovative and rapidly evolving field that integrates informatics, mathematics, and physics to create hybrid models for physical systems. This approach leverages data from real-world measurements, typically captured through sensors, and combines it with physics-based knowledge to enhance model accuracy and interpretability. A popular method in this field is the so-called physics-informed neural networks (PINN), which involves embedding a differential equation in the loss function. This is an example of a soft-constrained method, in which a standard method is used to fit data but penalised if the model does not satisfy a given prior assumption.

Although a brief introduction to PINN will be given with a general overview on physics-informed machine learning, the primary focus of the tutorial will be on hard-constrained methods. Specifically, we will introduce and work with (pseudo-)Hamiltonian neural networks for learning ordinary and differential equations. These methods involve designing neural network architectures in such a way that the models respect underlying structures of the system, like energy-preservation. In the hands-on session the attendees will use these methods to train models for springs and waves.

The tutorial will be presented by Sølve Eidnes from SINTEF Digital Oslo

Tutorial 3: Responsible and Explainable Artificial Intelligence

This tutorial presents ongoing research in the field of Responsible Artificial Intelligence (RAI) by introducing core concepts and means of operationalising AI ethics. Attendees will—through both lectures and problem-based learning exercises—get experience in implementing and testing systems for policy compliance.

Here, we introduce the fundamental aspects of RAI by providing a holistic multidisciplinary view. The course structure is such as to introduce to the attendees to the impact intelligent and autonomous systems have on societies and individuals and ongoing state-of-the-art discussions related to ethical, legal, and social aspects of AI. This introduction will be followed by a critical discussion of where accountability and responsibility lie for ethical, legal, and social impacts of these systems, considering decision points throughout the development and deployment pipeline. With this knowledge in mind, students will be introduced to socio-technical approaches the governance, monitoring and control of intelligent systems as tools for incorporating constraints into intelligent system design. Finally, participants apply these skills on a simulated responsible design problem.


The tutorial will be presented by Leila Methnani and Virginia Dignum from Umeå University. This tutorial is sponsored by EurAI.


Tutorial 4: Large Language Models Under the Hood

Recently, large generative language models (LLMs) have not only become the backbone of natural language processing (NLP), but also entered daily life of those not interested in deep learning. The names of the models like GPT-4, Gemini, Llama, etc, often make headlines.

However, under the hood these systems are not some apocalyptic "artificial intelligence": they are still statistical language models based on well-known techniques from machine learning and specifically multi-layer (thus "deep") artificial neural networks. This tutorial will introduce the attendants to the foundations of deep learning and then move on to explaining how these approaches are employed in pre-training  models capable of seemingly human-like conversational capabilities. Most recent research problems associated with generative LLMs will also be briefly presented.


The tutorial will include a hands-on session where the participants will be given access to remote computing nodes and will directly interact 

with open-source large language models for Norwegian (of NORA.LLM family https://huggingface.co/norallm) using Python. By the end of the 

tutorial, the students will have a solid understanding of the basics of modern generative language models, and will acquire practical experience of loading, using and evaluating LLMs locally (as opposed to querying black-box API endpoints like ChatGPT). For the hands-on part, at least basic knowledge of Python is a prerequisite, although there will also be a possibility to form teams including students with varying levels of programming skills.

The tutorial will be presented by Andrey Kutuzov, Egil Rønningstad, and David Samuel from the Language Technology Group from the University of Oslo.


Willem Wageman

Willem Waegeman is a professor at Ghent University, and a member of the research unit Knowledge-based Systems (KERMIT) of the Department Data Analysis and Mathematical Modelling. His main interests are machine learning and data science, including theoretical research and various applications in the life sciences. Specific interests include multi-target prediction problems, constructive machine learning, preference learning and time series analysis.

Sølve Eidnes

Sølve Eidnes is a senior research scientist in the Analytics and AI group in SINTEF Digital in Oslo. He has a PhD in structure-preserving numerical integration of differential equations from the Norwegian University of Science and Technology (NTNU). His current research involves developing physics-informed machine learning methods for use in industrial systems. This involves leveraging techniques from numerical analysis, and specifically structure-preserving integration, in the development and analysis of neural networks.


Leila Methnani

Leila is a WASP-HS affiliated PhD student at the Department of Computing Science and member of the Responsible Artificial Intelligence research group at Umeå University. Her research focuses on explainable AI and human-centric approaches to realising explainable AI in industry. She is an organising committee member of the upcoming workshop in Ethics of Game Artificial Intelligence (EGAI) at ECAI 2023.

Virginia Dignum

Virginia is Professor at Umeå University and Directory of WASP-HS, the Wallenberg Program on Humanities and Society for AI, Autonomous Systems and Software, the largest Swedish national research program on fundamental multidisciplinary research on the societal and human impact of AI. She is a member of the Royal Swedish Academy of Engineering Sciences (IVA), and a Fellow of the European Artificial Intelligence Association (EURAI). She is member of the Global Partnership on AI (GPAI), World Economic Forum’s Global Artificial Intelligence Council, Executive Committee of the IEEE Initiative on Ethically Aligned Design, of ALLAI, the Dutch AI Alliance, EU’s High Level Expert Group on Artificial Intelligence, leader of UNICEF’s guidance for AI and children, and member of UNESCO expert group on the implementation of AI recommendations. She is author of “Responsible Artificial Intelligence: developing and using AI in a responsible way”. 

Andrey Kutzov

Andrey Kutuzov is an associate professor at the University of Oslo Language Technology Group (LTG). His main research interest is computational semantics, including language models and diachronic semantic change.

Egil Rønningstad

Egil Rønningstad is a PhD student at the University of Oslo Language Technology Group(LTG). His research focus is Norwegian text analysis with language models, great and small.

David Samuel

David Samuel is a PhD student at the University of Oslo Language Technology Group (LTG). His research focus is pretraining of large and small language models.

Course Description

Deep learning is a rapidly growing segment of machine learning. It is increasingly used to deliver near-human level accuracy for many tasks such as image classification, voice recognition and natural language processing. Applications areas include facial recognition, scene detection, advanced medical and pharmaceutical research, and autonomous, self-driving vehicles. 

This 5-day course is built upon tutorials on specific topics on deep learning. The course further encompasses among others keynote talks, as well as special sessions on industry and diversity in AI as part of the NLDL conference program. 

The Winter School will provide a study of several emerging topics of high relevance within advanced deep learning, from a basic understanding of the techniques to the latest state-of-the-art developments in the field. In addition, the participants will be exposed to the latest advances and applications in deep learning by the oral presentations and poster presentations in the main conference program.


The course will thus consist of 5 full days of the NLDL conference including: tutorials, keynote sessions, oral presentations and poster presentations, as well as practical components.

Evaluation Committee for the Course's exam: 

Contact : 

Winter School Chair:

For other inquiries: