Academic Lectures
We have invited several high-level specialists all over the world to give lectures and share your knowledge with you. Here is a short overview of their background and abstracts.
Prof. Eyke Hüllermeier - Uncertainty Quantification in Machine Learning: From Aleatoric to Epistemic
University of Munich (Ludwig-Maximilians-Universität München) (Linkedin, Scholar)
Abstract:
Bio: https://en.cs.uni-paderborn.de/is/team/group/huellermeier
Eyke Hüllermeier is a full professor at the Heinz Nicdorf Institute and the Department of Computer Science at Paderborn University, Germany, where he heads the Intelligent Systems and Machine Learning Group. He graduated in mathematics and business computing, received his PhD in computer science from the University of Paderborn in 1997, and a Habilitation degree in 2002. Prior to returning to Paderborn in 2014, he spent two years as a Marie Curie fellow at the Institut de Recherche en Informatique de Toulouse (IRIT) in France (1998-2000) and held professorships at the Universities of Marburg (2002-04), Dortmund (2004), Magdeburg (2005-06) and again Marburg (2007-14).
His research interests are centered around methods and theoretical foundations of artificial intelligence, with a specific focus on machine learning and reasoning under uncertainty. He has published more than 300 articles on these topics in top-tier journals and major international conferences, and several of his contributions have been recognized with scientific awards. Professor Hüllermeier serves on the editorial board of several journals, including Machine Learning, Journal of Machine Learning Research, Data Mining and Knowledge Discovery, IEEE Computational Intelligence Magazine, Artificial Intelligence Review, and the International Journal of Approximate Reasoning. He is a coordinator of the EUSFLAT working group on Machine Learning and Data Mining and head of the IEEE CIS Task Force on Machine Learning.
Prof. Christoph Lampert - Fair and Robust Machine Learning
Institute of Science and Technology Austria (homepage, Scholar)
Abstract:
In times where more and more automatic decision systems are deployed for real-world tasks, machine learning researchers have to make sure that the systems are not only accurate, but also robust and fair in the decision they make. In the talk, I will give an introduction what these terms mean in a scientific context, and I will discuss recent developments for achieving them.
Public lecture: Prof. Christoph Lampert -- Behind the scenes: How does one become a (machine learning) researcher, and what does it mean to be one?
Time and location: Wednesday, August 17, 15:30, Delta Centeer, Tartu
Abstract:
Machine learning (=artificial intelligence) researcher might currently be one of the coolest jobs on the planet. But what does it take to become a machine learning researcher, and what do they actually do all day? The talk shines light on the steps for starting a scientific career (ideally in machine learning, but not only). In particular, how to successfully apply to a graduate school? And how to find a job in academia or industry research afterwards? I will also detail how the scientific community works behind the scenes: How do papers get published? How to build a scientific network and why? And where does the money for all of this come from? After the talk there be time for a Q&A.
Bio: Christoph Lampert received the PhD degree in mathematics from the University of Bonn in 2003. In 2010 he joined the Institute of Science and Technology Austria (ISTA) first as an Assistant Professor and since 2015 as a Professor. There, he leads the research group for Machine Learning and Computer Vision, and since 2019 he is also the head of ISTA's ELLIS unit. His research on computer vision and machine learning has won international and national awards, including a best paper prize at CVPR 2008. In 2012 he was awarded an ERC Starting Grant (consolidator phase) by the European Research Council. He currently is an Editor of the International Journal of Computer Vision (IJCV) and Action Editor of the Journal for Machine Learning Research (JMLR), and he used to be Associate Editor in Chief of the IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI).
Assoc. Prof. Meelis Kull - Calibration and confidence in Machine Learning
Machine Learning, University of Tartu (Scholar)
Abstract: For many purposes, it is important that machine learning models would report their uncertainty also, in addition to just making a prediction. Information about uncertainty allows to reduce the harm of errors, i.e. an autonomous car can slow down if its perception system has high uncertainty, and a doctor can run additional tests if the diagnostic tool is highly uncertain. Unfortunately, the huge success of deep neural nets has been shadowed by over-confidence, e.g. on out-of-distribution instances, or adversarial examples, or just outright over-confidence on standard test instances from the same distribution as training data. In this presentation, we will review the over-confidence problem and look into some methods that can be used for obtaining calibrated uncertainty estimates, particularly in multi-class classification, such as temperature scaling (2017), Dirichlet calibration (2019), and decision calibration (2021).
Bio: Meelis Kull is an associate professor of machine learning at the Institute of Computer Science, University of Tartu. He is leading a research group developing machine learning methods that know the limits of their capabilities and report uncertainty if needed, thus making the AI systems more trustworthy.
Prof. Raul Vicente - Complexity Science for Computer Scientists
Computational Neuroscience, University of Tartu (Scholar)
Abstract:
Bio:
Prof. Mark Fišel - "Learning to Write Text (Artificially)
NLP Tartu, University of Tartu (Scholar)
Abstract: The lecture will cover the topic of decoders -- the technology behind GPT-3, Google Translate and other famous solutions that allows neural networks to generate text. We will talk about standalone as well as conditioned decoders, their training data, pre-training and fine-tuning. We will discuss GPT-3-like models and what they can and cannot do -- why they are so good, what tasks they perform without specifically being taught and why they are not general AI despite the hype around them. We will compare encoders and decoders and will also discuss the challenges in evaluating text generation solutions.
Bio: Mark Fishel is heading the chair NLP at UTartu. His and his group's research currently focuses on neural networks and covers machine translation, text-to-speech synthesis and several other NLP tasks. A particular area of interest for the group is low-resource settings and making neural networks work without the abundance of training material they usually rely on.
Assoc. Prof. Jaan Aru - The neural idea factory: How new ideas come about"
AI and Neuroscience, University of Tartu (Scholar)
Abstract: In our work in academia or industry, we need to be able to generate new ideas. Creativity has been studied for decades; I will try to distill some of this research into practical tips about how to come to novel solutions. We will also discuss the neural mechanisms underlying the generation of novel thoughts. How can we take advantage of these mechanisms in our work and everyday life?
Bio: Jaan Aru is an associate professor of artificial intelligence and computational neuroscience at the Institute of Computer Science at the University of Tartu. He did his PhD in the Max Planck Institute for Brain Research and obtained the Barbara-Wengeler Prize for it. In addition to more than 35 peer-reviewed articles in international journals, he has published three books in Estonian. For his efforts in popularizing science, he has received twice the national science communication prize in Estonia. In 2019 he received the Young Scientist Prize from the President of Estonia.
Marek Rei, PhD - Encoders: The Art of Packing Text into Vectors
Lecturer, Imperial College London (Homepage, Scholar)
Abstract: Encoders are the backbone of nearly all modern NLP systems. They allow us to take a text of variable length and turn it into a vector (or a sequence of vectors) that machine learning models can use as informative features. In this talk, we will investigate neural encoder models, why they are useful and how exactly do they work. We will go into more detail on BERT, the most well-known pre-trained encoder, but we will also discuss other variations for different applications. The talk will also go over the most important points of building the best possible encoder, which can be applied to other types of data beyond text and natural language.
Bio: Marek Rei is a Lecturer of Machine Learning at Imperial College London. His research is focussed on improving neural machine learning architectures for understanding and modelling natural language. Much of his work extends the areas of representation learning, transfer learning and multi-task optimization. These models then get applied on practical applications in the areas of education, healthcare and business analytics. Marek did his PhD and post-doc at the University of Cambridge. He has also worked at industry, researching neural language models for text prediction at SwiftKey and co-founding Transformative AI for the prediction of cardiac arrhythmias in critical care patients.
Prof. Tanel Tammet - On Common Sense
Applied Artificial Intelligence group, Tallinn University of Technology ( ETIS CV, Scholar)
Abstract:
The talk investigates the current state of the art and the main research directions for common sense reasoning on natural language texts: one of the holy grails of A.I. We will start with a brief history of the area and continue with the approaches and results using end-to-end machine learning: the successes, failures and the issue of explainability. We'll briefly consider why rule based reasoning has not been successful for the common sense problem. Next we will investigate the current hybrid systems combining machine learning with rule-based reasoning and describe both the approach of adding rules to machine learning systems and the opposite approach of adding machine learning to rule-based systems. The last part of the talk goes into the examples and details of the hybrid common sense system our research group is working on, along with the practical open questions on the path.
Bio: Tanel Tammet is a full professor of applied artificial intelligence at the Tallinn Uni of Technology. He got his MSc from the Uni of Tartu and PhD from the Chalmers Uni of Technology. His core research area is automated reasoning: developing the theory and implementing several leading provers for both classical first order logic and different nonclassical logics. He has also worked on industrial IT systems, government information systems, cyber security, machine learning, data integration and visualization for both large corporate databases and crowd-sourced geodata. His current line of work is building hybrid - machine learning plus automated reasoning - systems for explainable commonsense reasoning over natural language texts.