FYS-8603 Winter School NLDL 2025 - 5 stp

This course 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.

The course will consist of 5 full days of the NLDL conference including: tutorials, keynote sessions, and practical components.

Core concepts:

• Uncertainty in Machine Learning
  • Importance of uncertainty in machine learning
  • Relevance for practical applications with safety requirements
  • Challenges and problems
• Uncertainty Representation in Statistics
  • Concepts of uncertainty in classical statistics
  • Probability theory for uncertainty quantification
• Aleatoric and Epistemic Uncertainty
  • Definition and differences between aleatoric and epistemic uncertainty
  • Examples and implications in real-world machine learning tasks
  • Novel Approaches for Uncertainty Quantification
  • Recent advancements beyond traditional statistical methods
  • New formalisms and uncertainty calculi in machine learning
• Quantifying Uncertainty in Machine Learning
  • Practical scenarios where uncertainty quantification is critical
  • Case studies showcasing uncertainty management in ML models
  • Application of uncertainty quantification techniques on sample datasets
  • Open problems and future directions in uncertainty quantification
• Introduction to Responsible AI (RAI)
  • Overview of Responsible Artificial Intelligence
  • Ethical, legal, and social impacts of AI systems
  • Foundations of AI Ethics
  • Key concepts in AI ethics: fairness, transparency, accountability
  • Introduction to AI policy and governance frameworks
• Operationalizing AI Ethics
  • Methods for embedding ethical principles in AI systems
  • Tools and techniques for monitoring and controlling AI behavior
• Accountability and Responsibility in AI
  • Critical points in AI development where responsibility lies
  • Legal and societal implications of autonomous systems
  • Strategies for incorporating constraints into AI system design
  • Ongoing state-of-the-art discussions on AI regulation
• Simulated scenario for responsible AI system design
  • Group exercises focusing on ethical compliance testing
  • Future research and challenges in RAI
• Introduction to Large Language Models (LLMs)
  • Role of LLMs in Natural Language Processing (NLP)
  • Overview of popular LLMs: GPT-4, Gemini, Llama
  • Foundations of Deep Learning for LLMs
• Pre-Training and Fine-Tuning LLMs
  • Understanding pre-training objectives (e.g., masked language modeling)
  • How LLMs learn human-like conversation capabilities
• Recent Research Challenges in LLMs
  • Ethical concerns: bias, privacy, and environmental impact
  • Technical challenges: scalability, interpretability
• Hands-On Session with Open-Source LLMs
  • Accessing and interacting with NORA.LLM for Norwegian LLMs
  • Evaluating LLMs locally: performance metrics and customization
  • Collaborative exercises using LLMs for practical tasks
  • Summary & Future Directions

Relevance of course in program of study: Deep learning has revolutionized technology and science by enabling information to be extracted with much better precision and at much larger scales compared to only a few years ago from data sources such as images, text, speech, biological material, chemical components, and sensory measurements in general. Most study programs offer courses on the fundamental theory and applications of neural networks and machine learning systems in general, which provide the backbone of deep learning. This course treats topics within deep learning that are not covered in standard courses. The course consists of topics such as in-depth exploration of uncertainty in machine learning (ML), covering its significance, challenges, and representation in statistics. It distinguishes between aleatoric and epistemic uncertainty, emphasizing their implications in practical ML tasks. The course introduces novel approaches for uncertainty quantification and examines case studies demonstrating its importance in real-world applications. In addition, this course also explores Responsible AI (RAI), discussing ethical, legal, and societal impacts, and methods to embed ethical principles into AI systems. Within RAI, socio-technical approaches to the governance, monitoring and control of intelligent systems are discussed with the help of a simulated responsible design problem. In addition, the course’s curriculum includes large language models (LLMs), focusing on their role in NLP, pre-training, and fine-tuning, with hands-on sessions using open-source LLMs for practical applications. In this session, state-of-the-art research problems associated with generative LLMs are discussed.

PhD students or holders of a Norwegian master’s degree of five years (300 ECTS) or 3 (180 ECTS) + 2 years (120 ECTS) or equivalent may be admitted. PhD students do not have to prove English proficiency and are exempt from a semester fee. Holders of a master’s degree must upload a master´s diploma with diploma supplement in English. PhD students at UiT The Arctic University of Norway can register for the course through StudentWeb. External applicants apply for admission through SøknadsWeb. All external applicants must attach a confirmation of their status as a PhD student from their home institution. Please note that students who hold a master of science degree but are not yet enrolled as a PhD student must attach a copy of their master's degree diploma. These students are also required to pay the semester fee.

Recommended prerequisites: Programming skills in Python and hands-on knowledge of Python programming for deep learning. Knowledge of machine learning at master’s level from study programs in computer science, physics and technology, mathematics and statistics, or equivalent.

Application code: 9316, application deadline: 15 November 2024.

The course is limited to 40 participants. Qualified applicants are ranked on the basis of a lottery if there are more applicants than available places.

Knowledge - The student is able to

• describe aleatoric and epistemic uncertainty in ML.
• discuss novel approaches for uncertainty quantification.
• discuss uncertainty quantification techniques on sample datasets.
• describe Responsible AI (RAI).
• discuss ethical, legal, and societal impacts, and methods to embed ethical principles into AI systems.
• describe large language models (LLMs) and its foundations.
• discuss pretraining and fine tuning LLMs.
• discuss accessing, interacting, working, and evaluation of LLMs.

Skills - The student is able to

• explain the fundamental ideas behind aleatoric and epistemic uncertainty in ML..
• explain uncertainty quantification techniques on sample datasets.
• explain Responsible AI (RAI).
• explain ethical, legal, and societal impacts, and methods to embed ethical principles into AI systems.
• apply the learned material to a new application or problem setting.
• use uncertainty quantification techniques and Responsible AI (RAI) for research and industrial settings using software libraries such as e.g., Pytorch or TensorFlow.
• explain large language models (LLMs), its foundations.
• explain pretraining and fine tuning LLMs.
• work towards accessing, interacting, and evaluation of LLMs for a practical task.

General competence - The student is able to

• give an interpretation of recent developments and provide an intuition of the open questions in the field of aleatoric and epistemic uncertainty in ML, .
• show an understanding of Responsible AI (RAI), and how to understand ethical, legal, and societal impacts, and methods to embed ethical principles into AI systems.

work towards understating LLMs and pretraining and fine tuning LLMs, interacting, working, and evaluation of LLMs for a practical task.

Lectures: 40 hours (full 5 days of the winter school)

Self-study sessions: 40 hours

Project work: spread over 8 weeks - net time 50 hours

Hands-on sessions: 10 hours

Net effort (~120 hours)