dataninja

Lecturer: Dr. Viktor Bengs is a postdoctoral researcher at LMU Munich. His research focuses on the development of theoretically sound algorithms for sequential decision tasks with weakly supervised feedback, e.g., preference information or censored observations, with automated algorithm configuration being the main application area. His recent research covers bandit algorithms, explainable AI, and the study of fundamental properties for quantifying uncertainty.

Summary: This tutorial provides an overview of multi-armed bandits, an intensively studied problem class in the realm of machine learning, in which a learner is facing a sequential decision-making problem under uncertainty. A decision (action) corresponds to making a choice between a finite set of specific choice alternatives (objects, items, etc.), also called arms in reference to the metaphor of gambling machines in casinos. After each decision to choose a particular arm, the learner receives some form of feedback – typically a numerical reward – determined by a feedback mechanism of the chosen arm. The learner is not aware of the arms’ feedback mechanisms and consequently tries to learn these in the course of time by performing actions according to its learning strategy. The concrete design of its learning strategy depends essentially on two main components of the learning setting: the assumptions on the feedback mechanisms and the learning task.

In this tutorial, we will have a special focus on the classical learning setting, where the feedback mechanism is of stochastic nature and the learning task is to minimize regret. Accordingly, the tutorial begins by introducing the basic terminology and concepts of multi-armed bandits, including the exploration-exploitation tradeoff and regret minimization. Moreover, different examples of real-world applications of multi-armed bandits, such as online advertising, personalized recommendations, and clinical trials will be provided. In the next step, the different types of bandit algorithms and the idea for their design will be presented. In addition to the theory, we will develop small code examples for the algorithms in the tutorial. As an outlook, we will consider further assumptions about feedback mechanisms and the learning task.

By the end of the tutorial, the participants will have a solid understanding of the theory and practice of multi-armed bandits and be able to apply these techniques to their own problems.

Lecturer: Jonas Peters is Professor of Statistics at the University of Copenhagen. His work focuses mainly on causal inference, trying to learn causal structures either from purely observational data or from a combination of observational and interventional data. His work covers both theoretical and methodological contributions, relating to areas like high-dimensional statistics, computational statistics or graphical models.

Summary: In science, we often want to understand how a system reacts under interventions (e.g., under gene knock-out experiments or a change of policy). These questions go beyond statistical dependences and can therefore not be answered by standard regression or classification techniques. In this tutorial we will learn about the powerful language of causality and recent developments in the field. No prior knowledge about causality is required.

Lecturer: Dr. Andrew Melnik is an AI Researcher at Bielefeld University, focusing on integrating Artificial Intelligence, Machine Learning,and Robotics. In 2021, Andrew received one of the coveted AI-Starter grants for his project “Learning to Plan with Deep Neural Networks“.

Summary: Reinforcement learning is a powerful technique used in various fields of artificial intelligence, including natural language processing, robotics, and agent-based systems. I will introduce the basic concepts of reinforcement learning, including the Q-learning algorithm, discuss how reinforcement learning is used in ChatGPT, applications in game playing, explore the role of reinforcement learning in robotics for navigation, control, and manipulation. Overall, I will provide a comprehensive overview of the current state-of-the-art in reinforcement learning and highlight future research directions in this exciting field.

Lecturer: Ann Nowé, director of the AI-Lab at Vrije Universiteit Brussels (VUB), is a professor both in the Computer Science Department of the faculty of Sciences as well as in the Computer Science group of the Engineering faculty at VUB. Prof. Nowé is well-known for her work in reinforcement learning, supporting the development of smarter and safer intelligent agents.

Lecturer: Christian Igel is a professor at the Department of Computer Science at the University of Copenhagen, and director of the SCIENCE AI Centre. His main research area is Machine Learning, with particular focus on the theory, application, and efficient implementation of deep neural networks, kernel-based methods, multi-objective optimization, reinforcement learning, and ensemble methods. With his strong commitment towards cross-disciplinary collaborations, he develops applications of machine learning that help achieve the sustainable development goals.

Summary: Machine learning can help to reach sustainable development goals. In this talk, we will consider deep learning for semantic segmentation in the areas “Good Health and Well-being”, “Life on Land”, and “Climate Action”. 

Semantic image segmentation algorithms provide a classification of each pixel or voxel of the input, and their basic concepts can be adapted for time series analysis. We will discuss fully convolutional neural networks for segmentation of 1-, 2- and 3-dimensional data with application examples from remote sensing and medical data analysis.

Lecturer: Prof. Virginia Dignum leads the Social and Ethical Artificial Intelligence group at Umeå University. Her research focuses on the complex interconnections and interdependencies between people, organizations, and technology. Thus, she works towards responsible AI through value-sensitive designs of intelligent systems, and multi-agent organisations. As one of the leading experts in AI and its social and ethical impact, she acts as a scientific advisor in several international initiatives on policy and strategy guidelines for AI research and applications.

Summary: Every day we see news about advances and the societal impact of AI. AI is changing the way we work, live, and solve challenges but concerns about fairness, transparency or privacy are also growing.

Ensuring AI ethics is more than designing systems whose result can be trusted. It is about the way we
design them, why we design them, and who is involved in designing them. In order to develop and use
AI responsibly, we need to work towards technical, societal, institutional and legal methods and tools
which provide concrete support to AI practitioners, as well as awareness and training to enable
participation of all, to ensure the alignment of AI systems with our societies’ principles and values.

SAIL Lecture and Networking Event

Lecturer: Tanja Schultz is professor for Cognitive Systems of the Faculty of Mathematics & Computer Science at the University of Bremen. She is the spokesperson of the University Bremen high-profile area “Minds, Media, Machines” and the DFG Research Unit “Lifespan AI: From longitudinal data to lifespan inference in health”. Prof. Schultz is a recognized scholar in the field of multilingual speech recognition and cognitive technical systems, where she combines machine learning methods with innovations in biosignal processing to create technologies such as in “Silent Speech Communication” and “Brain-to-Speech”.

Summary: In my talk, I will describe technical cognitive systems that automatically adapt to users’ needs by interpreting their biosignals. Human behavior includes physical, mental, and social actions that emit a range of biosignals which can be captured by a variety of sensors. The processing and interpretation of such biosignals provides an inside perspective on human physical and mental activities, complementing the traditional approach of merely observing human behavior. As great strides have been made in recent years in integrating sensor technologies into ubiquitous devices and in machine learning methods for processing and learning from data, I argue that the time has come to harness the full spectrum of biosignals to understand user needs. I will present illustrative cases ranging from silent and imagined speech interfaces that convert myographic and neural signals directly into audible speech, to interpretation of human attention and decision making from multimodal biosignals.

Lecturer: Dr. Martin Riedmiller, research scientist and former professor for machine learning, is the team lead at DeepMind. His core scientific interest are intelligent machines learn new things autonomously from scratch. His work particularly focuses on neural networks and their ability to store and generalize information.

Summary: Intelligence is the ability to efficiently and effectively generate knowledge out of experience’ – guided by this hypothesis, we investigate algorithms and agent architectures that can autonomously learn with minimal interactions and from minimal prior knowledge. I will discuss the ‘collect & infer’ principle that provides the blueprint for an agent architecture of interacting learning processes and give concrete examples of their implementation. The learning behavior will be shown on several examples in the field of control and robotics.

Lecturer: Dr. Noémie Jacquier is a postdoctoral researcher in the HighPerformance Humanoid Technologies Lab (H²T) at the Karlsruhe Institute of Technology (KIT). Her work is centered around providing robots with efficient and reliable learning capabilities by combining robot learning and control with differential geometry. Specifically, she develops models that efficiently integrate geometric information encapsulated in data.

Summary: To be deployed in our everyday life, robots must display outstanding learning and adaptation capabilities allowing them to act, react, and continuously learn in unstructured dynamic environments. A key component in both data-driven learning and adaptation is how robots may exploit explicit (e.g., domain knowledge) or implicit (e.g., learned) structures arising in the collected data. Domain knowledge and data structures in robotics can be viewed through the lens of geometry: Different variables have specific geometric characteristics, collected data may lie on curved spaces, and various problems can be naturally formulated from a geometric perspective. In this context, differential geometry, or more specifically Lie groups and Riemannian manifold theories provide appropriate methods to cope with the geometry of non-Euclidean spaces. Although geometric methods have been successfully applied to robotics from early on, they only gained interest recently in robot and machine learning. In this brief tutorial, I will provide an introduction on geometric methods in machine and robot learning, and showcase the benefits of these methods in various robotics applications.