Events at and around the MCML

In this workshop, we aim to bring together students and young researchers working in mathematics, computer science, the humanities and cultural history, to learn about state-of-the-art tools, both those readily designed with a specific use in the humanities in mind, and novel research, which could have a potential usecase in the digital humanities.

In spite of the tremendous success of machine learning, especially in its modern guise of deep learning, purely statistical, data-driven approaches to artificial intelligence (AI) are still lacking the commonsense abilities that distinguishes humans from machines. Neurosymbolic AI is a new branch of AI that seeks to combine the best of both worlds, data-driven learning and inductive inference on the one side, and more human-like symbolic reasoning and knowledge processing on the other. This AI double-feature workshop discusses the latest developments in the field of neuro-symbolic AI as well as the impact of AI on sustainable development.

Abstract: Results of simulation studies evaluating the performance of statistical methods are often considered actionable and thus can have a major impact on the way empirical research is implemented. However, so far there is limited evidence about the reproducibility and replicability of statistical simulation studies. In the RepliSims project, we assessed replicability of eight highly cited statistical simulation studies in teams of replicators based on reported information in the original publications. The talk will go over the lessons learned from this replication endeavor as well as the findings on replicability of the evaluated studies. The aim is to stimulate a group discussion on the concepts of reproducibility and replicability for statistical simulation studies. The talk will take place as a hybrid, however we welcome attendance.

Abstract: How could machines learn as efficiently as humans and animals? How could machines learn to reason and plan? How could machines learn representations of percepts and action plans at multiple levels of abstraction, enabling them to reason, predict, and plan at multiple time horizons? Prof. Dr. Yann LeCun, Chief AI Scientist for Meta AI Research and Silver Professor at the Courant Institute of Mathematical Sciences at New York University will propose a possible path towards autonomous intelligent agents, based on a new modular cognitive architecture and a somewhat new self-supervised training paradigm. The centerpiece of the proposed architecture is a configurable predictive world model that allows the agent to plan. Behavior and learning are driven by a set of differentiable intrinsic cost functions. The world model uses a new type of energy-based model architecture called H-JEPA (Hierarchical Joint Embedding Predictive Architecture). H-JEPA learns hierarchical abstract representations of the world that are simultaneously maximally informative and maximally predictable.

The DSSGx 2023 Munich program has been an incredible journey, throughout which our teams have worked tirelessly to address real-world challenges, leveraging the power of data science and technology to drive meaningful change. At the closing ceremony, our outstanding fellows will showcase the impressive projects and outcomes they achieved over the course of the program. Our keynote speaker will be Prof. Leah J. Welty of the Institute for Public Health & Medicine (IPHAM) and the Northwestern University Clinical and Translational Sciences Institute (NUCATS). We truly hope that you can join us in celebrating the accomplishments of the DSSGx Munich 2023 program. To confirm your attendance and reserve your spot, please RSVP by Sep 15th.

Featuring a keynote presentation by Anne Köpken from the German Aerospace Center. The session will delve into the Surface Avatar Project, a joint initiative by the German Aerospace Center and the European Space Agency to advance space telerobotics for Mars exploration and habitat establishment. Utilizing the International Space Station as a command hub, the project aims to orchestrate remotely piloted robots on Earth, tackling issues such as hazardous environments and communication delays. Attendees will have the opportunity for networking and discussion post-presentation.

The LMU Open Science Center is inviting masters and doctoral researchers to join a 5-day summer school so they can gain more trust in their research, and make it as credible as it can be in the eyes of their peers, the public, and funding agencies. Application to attend the whole summer school from September 11 to 15, 2023 is open until July 17, 2023, 12 noon. Registration to attend the public lectures online is unlimited.

Abstract: As MCML's Open Science Transfer Coordinator Moritz Herrmann is committed to making machine learning research more accessible and transparent to everyone. In addition to providing advice and support to individual researchers and projects on key issues such as reproducibility and replicability, his aim is to develop and implement policies and best practices to promote open science and foster collaborations in the scientific community to advance research and application of machine learning in an ethical, transparent, and reproducible manner.

Abstract: Kit Rodolfa is the Research Director at the Regulation, Evaluation, and Governance Lab (RegLab) at Stanford, working at the intersection of machine learning and public policy on using novel computational tools to modernize government and benefit society. His research interests include the bias, fairness, and interpretability of machine learning methods, as well as understanding and filling gaps between the theory and practice of machine learning.

Abstract: The talk presents the UBC PLAI group and Inverted AI's novel work on using foundation models and amortized inference for behavior, image, and video processing, with a particular emphasis on ITRA, a model set to become the GPT of driving behavior.

Abstract: This workshop provides you with an overview of different research strains in the field of network data analysis. In a hands-on session, you will learn to analyze a real-world network dataset using existing, readily available software packages.

Abstract: Jürgen Schimdhuber is one of leading AI researchers worldwide. Since the 1980s he pioneered the principle of generative adversarial networks, artificial curiosity, transformers with linearized self-attention, and meta-learning machines that learn to learn. In this talk in the LMU Audimax he will share his thoughts on AI.

Abstract: Stable Diffusion is a text-to-image AI-model capable of generating photo-realistic images given merely a textual description of the desired image content. This makes it possible to generate images quite easily without requiring special artistic skills, computer knowledge, or computer hardware and in a matter of seconds. In the few months since its release, it has become an enabling technology and platform for a large body of further research, numerous start-ups, and industrial applications. Having millions of daily users already and being among the fastest adopted open-source projects, Stable Diffusion has shown a great transformative potential with impact on many practical application areas. The opportunities, challenges, and the changes that the use of Stable Diffusion brings will be discussed in this panel with the automotive industry and image agencies as examples.

Abstract: Learn how to structure, tidy and anonymize raw geolocation data to create activity indicators using the data collected by the IAB-SMART-App.

Abstract: In this webinar, you will get to know the new IAB-SMART data module with activity indicators from smartphone sensor data.

Abstract: In the lecture 'Automation and Artificial Intelligence: Management for Do-It-Yourself-Artists' it will be shown how Artificial Intelligence opens up many new possibilities in various fields of application. Especially in management and the automation of organizational tasks besides the stage, artists can take advantage of these new possibilities. In this talk, Maximilian Blaschke will discuss various use cases, trends and relevance of new technologies in artist management.

Abstract: In this webinar, you will get an overview on how to collect smartphone data ethically and transparently with an Android app.

Abstract: In his talk, Armin Berger explains how knowledge workers, editors and culture enthusiasts can use AI not only to view more data and digital exhibits and make them accessible to more interested parties, but also how the AI-supported storytelling system enables users to create their own 'stories'. This opens up a whole new world of the museum for both scientists and visitors.

Abstract: In this inaugural lecture as a professor at LMU Institute of AI in Management, Stefan Feurriegel will speak in German on the topic of Humans or AI: Who Will Come Out on Top?.

WiDS Munich Conference is an independent event to coincide with the annual Global Women in Data Science (WiDS) Conference held at Stanford University and an estimated 150+ locations worldwide. All genders are invited to attend WiDS regional events, which features outstanding women doing outstanding work.

The ever increasing impact of methods in machine learning and data science urgently calls for the study of their foundations. Important such foundational questions include their mathematical justification, their interpretability, and their consequences for the nature of scientific inference; and these questions are the concern of theorists, methodologists and philosophers alike. This workshop brings together researchers from the Munich Center for Machine Learning (MCML) and the Munich Center for Mathematical Philosophy (MCMP) to exchange ideas on these topics and initiate possible collaborations.

Abstract: Machine learning (ML) methods, combined with large-scale electronic health databases, could enable a personalized approach to healthcare by improving patient-specific diagnosis, prognostic predictions, and treatment decisions. In this talk, I will focus on the problem of predictive inference on the effect of a treatment on individual patients using machine learning models applied to observational data. First, I will describe some Bayesian approaches to tackle this problem using Gaussian processes. Next, I will discuss model-free approaches for predictive inference based on conformal prediction that provide frequentist coverage guarantees. In particular, I will describe new methods for inference on ITEs by applying conformal prediction on top of pseudo-outcome regression models in a post-hoc fashion. Finally, I will discuss exciting avenues for future work.

Abstract: The lecture will provide an overview and insight into the possible uses and applications of AI in cultural institutions. One focus is on target group segmentation using a Self-Organizing Map (SOM) and the playful handling of digital data using Neural Style Transfer (NST). NST will be demonstrated live using a Python showcase.

Abstract: The 'intelligent.museum' project at the Deutsches Museum Nuremberg and the ZKM | Zentrum für Kunst und Medien Karlsruhe uses the exhibition space as a 'sandbox' to develop its own museum AI prototypes. The goal is to demystify the technologies and promote ethical discourse. In doing so, the project explores whether AI can help make exhibitions more inclusive and accessible. As part of the project, media artist and curator Yannick Hofmann aims to highlight both the opportunities and potentials as well as the risks and side effects of AI in the museum context. In addition, the project aims to show how art and cultural institutions can contribute to the sustainable development of future technologies.

Abstract: This talk discusses uncertainty quantification and inference using ensemble methods.

Abstract: The work Deep.Dance by Hamburg-based choreographer Jascha Viehstädt, created entirely by an artificial intelligence, focuses on a still great unknown in research on human intelligence and artificially intelligent software: creativity. What is creativity and can it be manufactured by machines? Deep.Dance confronts the makers and observers with the questions: What happens when the choreographer is not a human but a software? What does it mean when perhaps the most human of all processes - the creative process - is taken over by a machine? The lecture will address the potential and limitations of using artificial neural networks, probability theory, and big data analysis to replace artistic creative processes. A subsequent exchange puts up for discussion the future role of creative corporeality in the 21st century in the field of tension between code, art, networks and the body.

Abstract: Policy makers typically face the problem of wanting to estimate the long-term effects of novel treatments, while only having historical data of older treatment options. We assume access to a long-term dataset where only past treatments were administered and a short-term dataset where novel treatments have been administered. We propose a surrogate based approach where we assume that the long-term effect is channeled through a multitude of available short-term proxies. Our work combines three major recent techniques in the causal machine learning literature: surrogate indices, dynamic treatment effect estimation and double machine learning, in a unified pipeline. We show that our method is consistent and provides root-n asymptotically normal estimates under a Markovian assumption on the data and the observational policy. We use a data-set from a major corporation that includes customer investments over a three year period to create a semi-synthetic data distribution where the major qualitative properties of the real dataset are preserved. We evaluate the performance of our method and discuss practical challenges of deploying our formal methodology and how to address them.

Abstract: Martin Grünheits lecture will be about an insight into the artistic practice at the interface of theater and technology. Based on his work 'Regie: KI' at the Düsseldorfer Schauspielhaus and his new work 'HAUFEN UFFRUHR FORTSCHRITT' at the Theater Eisleben, it will be about different possible applications, problems and potentials of Machine Learning in the context of theater. The relationship between man and machine will be considered and a look at the challenges of such a project will be taken.

Abstract: In a recent effort to push modern tools from machine learning into several areas of science and engineering, deep learning based methods have emerged as a promising alternative to classical numerical schemes for solving problems in the computational sciences – example applications include fluid dynamics, computational finance, or computational chemistry. Philipp Grohs illuminates the limitations and opportunities of this approach, both on a mathematical and an empirical level.

Abstract: Sven Sören Beyer will give an insight into the project 'chasing waterfalls - An Artificial Intelligence Opera'. This is an opera project of the independent performing arts group phase7 performing.arts from Berlin with the Semperoper Dresden and the New Vision Performing Arts Festival in Hong Kong as co-producers. Together with composer and conductor Angus Lee (Hong Kong) and the creative studio kling klang klong (DEU), the theater collective embarks on a music-theatrical journey about the impact that AI and social media have on our lives. In the process of developing the work, in the composition and also during the performances, an AI is involved as a co-creating element in a so-called deep learning process and therefore becomes a co-author, co-composer and co-performer. In addition to the AI as a co-creating 'object', six singers and nine instrumentalists will interact musically and scenically with each other and explore the questions of what self-determination can look like in the digital age, what digital influences and changes the individual is exposed to, what proportion of our 'conscious' decisions is actually the result of autonomous action, whether AI is capable of creativity and empathy, and to what extent our identity, perceived as unchanging, is not rather a fluid, multifaceted-amorphous one.

AI researchers in Bavaria are ambitious, powerful and internationally renowned. After the successful AI.BAY2023 conference, the baiosphere is launching the new AI FOR US networking series in the 'German AI Month mAI'. In an exclusive setting, excellent AI researchers are invited to exchange and network with their colleagues and the Bavarian AI Council across disciplinary boundaries.The baiosphere guests can expect LLM live demos, four top-class keynote speeches from industry (Microsoft Deutschland GmbH) and research (Ludwig-Maximilians-Universität München, Technische Universität München) as well as an audience discussion on the topic of AI and society.

Abstract: The cooperation project 'Meistersinger reloaded' of the Singakademie Dresden with the AI startup Aleph Alpha from Heidelberg and students of the TU Darmstadt resulted in a machine-generated new composition based on music by Richard Wagner. Project initiator Michael Käppler talks about the hardships of acquiring training data, unexpected moments of fascination, and the collaboration between humans and machines.

Abstract: The lecture 'Improvising Machines and Listening Humans' develops along the question of how the interaction of humans and machines in aesthetic systems can be depicted and developed as an artistic practice. Prof. Nicola L. Hein will develop the question on the basis of his own artistic works. These juxtapose human musicians with virtual musicians who, with the help of Machine Learning applications, they are able to make music together and connected in difference.

The MCML partnered up with Women in AI & Robotics to invite young and motivated talents to this year’s Girls Day at LMU. On April 27th, 2023, 12 preselected female talents will have the opportunity to dive into the fascinating world of artificial intelligence for a day. The MCML is excited to welcome the new female generation in AI!

Abstract: This talk will first explain how Creative AI works. Through a comparison of creative AI and human creativity, the differences as well as the strengths and weaknesses of creative AI will be examined in more detail. Furthermore, we will explore the question of how this potential can be used for artistic purposes and why collaboration between humans and AI is essential for artistic projects and how the interaction and communication between art creators and AI could look like.

Abstract: This lecture will present some of her recent interactive works involving real-time mutual interaction between human musicians and interactive computer music systems. This compositional approach is based on a distributed concept of creativity that includes both human and non-human actors (computer music system) and makes use of a high degree of interpretive freedom and machine autonomy in these works. In addition, Artemi-Maria Gioti will discuss compositional challenges of this approach and the potential of machine learning as a tool for creative ideation.

The weekend marks the annual Data Fest in Munich, promising to be an exciting event for data enthusiasts. The DataFest Germany is a 48-hour competition where teams of Bachelor and Master students from different disciplines come together to analyze a large and complex data set provided by an organization or company. At the end, students are only allowed to present a few slides to convince the panel of expert judges that they have developed the best data-based insight or visualization. The MCML is looking forward to seeing you there!

Abstract: The introductory lecture will cover the basics of the 'black box' AI. Based on the 'milestones' of the last years and currently practiced application examples, the cultural use of the technology will be presented. The discussion will deepen the understanding of the potentials (and dangers) as well as the possible further development in the future.

Abstract: The talk introduces a regression model termed ”extended GB2 model”, which is designed to analyze ratios of biomarkers in epidemiological and medical research.

The GDSD bring together data scientists from various fields with the aim to identify and discuss new methods, explore new fields of application, and enhance the professional image of data scientists.

Abstract: There is a significant interest in developing ML algorithms whose final predictions can be explained in terms understandable to a human. To address this challenge, we develop a method for constructing high performance ML algorithms which are explainable by design.

Happy and proud to be a part of this at Munich Center for Machine Learning - we will be present on stage with our directors, with talks by PIs, a stand, and a lot more.

Are you interested in the latest advancements and research trends in human augmentation? Join us for the “Workshop on Interaction with Technologies for Human Augmentation” organized by LMU Munich and HumaneAI, co-organized by MCML.

Abstract: Networks are often used to represent complex dependencies in data, and network models can aid the understanding of such dependencies. This talk will present network models. We shall introduce a kernelized goodness of fit test (which is based on Stein’s method), give performance guarantees, and illustrate its use.

Abstract: We propose and demonstrate a novel approach to training image classification models based on large collections of images with limited labels. We take advantage of availability of radiology reports to construct joint multimodal embedding that serves as a basis for classification. We demonstrate the advantages of this approach in application to assessment of pulmonary edema severity in congestive heart failure that motivated the development of the method.

Abstract: The current developments relating to the energy transition (decarbonization, decentralization, electrification of heat and mobility) pose particular challenges for the distribution grids and make accurate load forecasts increasingly important. Novel modeling approaches like the Transformer approach have proven to be promising architectures of neural networks for sequence data. The temporal fusion transformer (TFT) architecture seems particularly suitable for time series data to obtain high quality forecasts. To date, there are only a handful of studies that apply this approach to the electricity load forecasting problem, mostly in combination with other analytical approaches. Therefore, examine the performance of the TFT on short-term forecasting problems in the distribution grid.