Short Biography

Gintarė is a senior research scientist at Google DeepMind, based in Toronto, an adjunct professor in the McGill University School of Computer Science, and an associate industry member of Mila, the Quebec AI Institute. Prior to joining Google, Gintarė led the Trustworthy AI program at Element AI / ServiceNow, and obtained her Ph.D. in machine learning from the University of Cambridge, under the supervision of Zoubin Ghahramani. Gintarė was recognized as a Rising Stars in Machine Learning by the University of Maryland program in 2019. Dziugaite is known for her work on network and data sparsity, developing algorithms and uncovering effects on generalization and other metrics. Dziugaite coined the term “linear mode connectivity” and carried out the first in depth study connecting it to the existence of lottery tickets, loss landscapes and the mechanism of iterative magnitude pruning. Another major focus of her research is on understanding generalization in deep learning, and more generally the development of information-theoretic methods for studying generalization. Her most recent work looks at removing the influence of data on the model (unlearning).

 

Title

The Dynamics of Memorization and Unlearning

 

Abstract

Deep learning models exhibit a complex interplay between memorization and generalization. This talk will begin by exploring the ubiquitous nature of memorization, drawing on prior work on “”data diets”“, example difficulty, pruning, and other empirical evidence. But is memorization essential for generalization? Our recent theoretical work suggests that eliminating it entirely may not be feasible. Instead, I will discuss strategies to mitigate unwanted memorization by focusing on better data curation and efficient unlearning mechanisms. Additionally, I will examine the potential of pruning techniques to selectively remove memorized examples and explore their impact on factual recall versus in-context learning.

Short Biography

Hardt is a director at the Max Planck Institute for Intelligent Systems, Tübingen. Previously, he was Associate Professor for Electrical Engineering and Computer Sciences at the University of California, Berkeley. His research contributes to the scientific foundations of machine learning and algorithmic decision making with a focus on social questions. He co-authored Fairness and Machine Learning: Limitations and Opportunities (MIT Press) and Patterns, Predictions, and Actions: Foundations of Machine Learning (Princeton University Press).

 

Title

The Emerging Science of Benchmarks

 

Abstract

Benchmarks have played a central role in the progress of machine learning research since the 1980s. Although there’s much researchers have done with them, we still know little about how and why benchmarks work. In this talk, I will trace the rudiments of an emerging science of benchmarks through selected empirical and theoretical observations. Looking back at the ImageNet era, I’ll discuss what we learned about the validity of model rankings and the role of label errors. Looking ahead, I’ll talk about new challenges to benchmarking and evaluation in the era of large language models. The results we’ll encounter challenge conventional wisdom and underscore the benefits of developing a science of benchmarks.

Short Biography

Mounia is a Senior Director of Research at Spotify and the Head of Tech Research in Personalization, where she leads an interdisciplinary team of research scientists. She also holds an honorary professorship at University College London and serves as a Distinguished Research Fellow at the University of Amsterdam. Previously, Mounia was a Director of Research at Yahoo, overseeing a team focused on advertising quality and collaborating on user engagement projects related to news, search, and user-generated content. Before her tenure at Yahoo, Mounia held a Microsoft Research/RAEng Research Chair at the School of Computing Science, University of Glasgow, and before that was a Professor of Information Retrieval at the Department of Computer Science at Queen Mary, University of London. She is a prominent figure in the research community, regularly serving as a senior program committee member at major conferences such as WSDM, KDD, WWW, and SIGIR. She has also been a program co-chair for SIGIR 2015, WWW 2018, WSDM 2020, and CIKM 2023. Mounia is widely recognized for her contributions as a speaker and author, with over 250 published papers and appearances on platforms like ACM ByteCast and the AI Business Podcasts series. She was nominated for the VentureBeat Women in AI Awards for Research in both 2022 and 2023.

 

Title

Enhancing User Experience with AI-Powered Search and Recommendations at Spotify

 

Abstract

This talk will explore the pivotal role of search and recommendation systems in enhancing the Spotify user experience. These systems serve as the gateway to Spotify’s vast audio catalog, helping users navigate millions of music tracks, podcasts, and audiobooks. Effective search functionality allows users to quickly find specific content, whether it is a favorite song, a trending podcast, or an informative audiobook, while also satisfying broader search needs. Meanwhile, recommendation systems suggest new and relevant content that users might not have thought to search for, while ensuring their current needs for familiar content are met. This encourages exploration and discovery of new artists, genres, and shows, enriching the overall listening experience and keeping users engaged with the platform. Achieving this dual objective of precision and discovery requires sophisticated technology. It involves a deep understanding of representation learning, where both content and user preferences are accurately modeled. Advanced AI techniques, including machine learning and generative AI, play a crucial role in this process. These technologies enable the creation of highly personalized recommendations by understanding complex user behaviors and preferences. Generative AI, for instance, allows us to create personalized playlists, thereby enhancing the user experience with innovative features. This presentation is based on the collective research and publications of numerous contributors at Spotify.

Short Biography

Katharina Morik received her doctorate from the University of Hamburg in 1981 and her habilitation from the TU Berlin in 1988. In 1991, she established the chair of Artificial Intelligence at the TU Dortmund. She retired in 2023. She is a pioneer of bringing machine learning and computing architectures together so that machine learning models may be executed or even trained on resource restricted devices. In 2011, she acquired the Collaborative Research Center CRC 876 “”Providing Information by Resource-Constrained Data Analysis” consisting of 12 projects and a graduate school. After the longest possible funding period of 12 years, the CRC ended with the publication of 3 books on Resource-Constrained Machine Learning (De Gruyter). She has participated in numerous European research projects and has been the coordinator of one. She was a founding member and Program Chair of the conference series IEEE International Conference on Data Mining (ICDM) and is a member of the steering committee of ECML PKDD. She is a co-founder of the Lamarr Institute for Machine Learning and Artificial Intelligence. Prof. Morik is a member of the Academy of Technical Sciences and of the North Rhine-Westphalian Academy of Sciences and Arts. She has been awarded Fellow of the German Society of Computer Science GI e.V. in 2019.

 

Title

Resource-Aware Machine Learning — a User-Oriented Approach

 

Abstract

Machine Learning (ML) has become integrated into several processes, ranging from medicine, manufacturing, logistics, smart cities, sales, recommendations and advertisements to entertainment and many more business and private processes. The applications together consume a considerable amount of energy and emit CO2. ML research investigates how to make models smaller and faster through pruning and quantization. Also the use of more energy- efficient hardware is an encouraging field. Research on ML under resource constraints is an active field proposing novel algorithms and scenarios. The aim is that for each application a variety of implementations is offered from which customers and the different types of users may choose the most thrifty one. This, in turn, would push tech providers to focus on the production of economical systems. However, if the customers, users, stakeholders do not know, which of the models offers the best tradeoff between performance and energy-efficiency, they cannot select the most frugal one. Hence, testing implementations of learning and inference needs to be developed. They should be easy to use, produce visualizations that are mass-tailored for specific user groups. Automatized testing is difficult due to the diversity of models, computing architectures, training and evaluation data, and the fast rate of changes. The talk will illustrate work on resource-aware ML and advocate to pay more attention to the role of users in the development of scenarios, models, and tests.

Short Biography

Patrick Lucey is currently the Chief Scientist at sports data giant Stats Perform, leading the AI team with the goal of maximizing the value of the company’s extensive sports data. He has studied and worked in the fields of machine learning and computer vision for the past 20 years, holding research positions at Disney Research and the Robotics Institute at Carnegie Mellon University, as well as spending time at IBM’s T.J. Watson Research Center while pursuing his Ph.D. Patrick originally hails from Australia, where he received his BEng(EE) from the University of Southern Queensland and his doctorate from Queensland University of Technology, which focused on multimodal speech modeling. He has authored more than 100 peer-reviewed papers and has been a co-author on papers in the MIT Sloan Sports Analytics Conference Best Research Paper Track for 11 of the last 13 years, winning best paper in 2016 and runner-up in 2017 and 2018. Additionally, he has won best paper awards at INTERSPEECH and WACV international conferences. His main research interests are in artificial intelligence and interactive machine learning in sporting domains, as well as AI education. He has recently piloted a course on “”AI in Sport,“” which aims to give students intuition behind AI methods using the interactive and visual nature of sports data.

 

Title

How to Utilize (and Generate) Player Tracking Data in Sport

 

Abstract

Even though player tracking data in sports has been around for 25 years, it still poses as one of the most interesting and challenging datasets in machine learning due to its fine-grained, multi-agent, team-based, and adversarial nature. Despite these challenges, it is also extremely valuable as it is (relatively) low-dimensional, interpretable, and interactive, allowing us to measure performance and answer questions we couldn’t objectively address before. In this talk, I will first give a brief history of tracking data in sports, then highlight the challenges associated with utilizing it. I will then show that by obtaining a permutation invariant representation, we can not only measure aspects of sports that couldn’t be done before, but also interact with and simulate plays akin to a video game via our “visual search” and “ghosting” technology. Finally, I will show how we can use both tracking and event data to create a multimodal foundation model, which enables us to generate player tracking data at scale and achieve our goal of “digitizing every game of professional sport.” Throughout the talk, I will utilize examples from top-tier basketball, soccer, and tennis.