Seminar Series of the Cluster for Excellence "Machine Learning"

-------------------------
Wednesday, 2:00 - 3:00 pm, followed by get-together

Lecture hall, AI Research Building, Maria von Linden-Str. 6 (ground floor), 72076 Tübingen
-------------------------

PROGRAM

• 01.12.2021   Beyond the First Portrait of a Black Hole  
                              Katherine L. (Katie) Bouman
                              Assistant Professor of Computing and Mathematical Sciences, Electrical Engineering and Astronomy at
                             California Institute of Technology Pasadena, CA, USA.  Website
  As imaging requirements become more demanding, we must rely on increasingly sparse and/or noisy measurements that fail to paint a complete picture. Computational imaging pipelines, which replace optics with computation, have enabled image formation in situations that are impossible for conventional optical imaging. For instance, the first black hole image, published in 2019, was only made possible through the development of computational imaging pipelines that worked alongside an Earth-sized distributed telescope. However, remaining scientific questions motivate us to improve this computational telescope to see black hole phenomena still invisible to us and to meaningfully interpret the collected data. This talk will discuss how we are leveraging and building upon recent advances in machine learning in order to achieve more efficient uncertainty quantification of reconstructed images as well as to develop techniques that allow us to extract the evolving structure of our own Milky Way's black hole over the course of a night.

• 06.10.2021   Employing Immersive Virtual Reality and Machine Learning to Reveal Geometric Principles
                              of Individual and Collective Decision-Making   
                              Iain Couzin
                              Director of the Max Planck Institute of Animal Behavior, Department of Collective Behaviour
                              and the Chair of Biodiversity and Collective Behaviour at the University of Konstanz  Webseite
  I will discuss our development and application of new machine learning technologies for automated tracking and identification of unmarked individuals, computational sensory reconstruction, bio-mimetic robotics, and ‘holographic’ virtual reality (VR) for animals, and demonstrate how these quantitative methodologies provide new insights into individual and collective sensing and decision-making. For example, they allow us to reconstruct (automatically) the dynamic, time-varying sensory networks in large animal collectives, to identify at any instant in time the most socially-influential individuals within groups, and to predict cascades of social contagion (behavioural change) before they actually occur. Furthermore, such methodologies enable us to study phenomena across multiple scales of biological organisation, from neural interactions to individual and collectives decision-making. Specifically, we have recently investigated how animals choose among spatially-discrete options, a central challenge in their lives. Employing an integrated theoretical and experimental approach (using immersive VR), with both invertebrate and vertebrate models—the fruit fly, desert locust and zebrafish—we find that the brain spontaneously reduces multi-choice decisions into a series of abrupt (critical) binary decisions in space-time, a process that repeats until only one option—the one ultimately selected by the individual—remains. This mechanism facilitates highly effective decision-making, and is shown to be robust both to the number of options available, and to context, such as whether options are static (e.g. refuges) or mobile (e.g. other animals). In addition, we find evidence that the same geometric principles of decision-making occur across scales of biological organisation, from neural dynamics to animal collectives, suggesting they are fundamental features of spatiotemporal computation.

• 07.07.2021     Beyond Expectations & Better than Average: Foundations of Machine Learning Systems
                              Robert Williamson
                              Professorship for 'Foundations of Machine Learning Systems'
                              at our Cluster of Excellence  "Machine Learning"
  Machine Learning (ML) technology provides value by its integration into a system, either technological or socio-technical. This requires an external outward-looking view of ML, rather than the usual internal inward-looking view of much ML research. I will illustrate the external view by considering some foundational concerns regarding  ML which motivates the research I propose to conduct in Tübingen.  The starting point turns out to be a more careful analysis of expectation or averages. Somewhat astonishingly, there is a lot to be said, and a lot to be yet discovered, about this apparently elementary building block of ML systems, and about its natural generalisations. I will argue why one should consider moving beyond expectations, and explore some of the intriguing technical problems that naturally arise. I will also explain how replacing expectation by something else offers an attractive alternate approach to questions of fairness in ML, but how this choice, like other approaches,  is inherently value-laden in a strong sense.
   

• 09.06.2021      Learning from Trees - Modeling, Simulation, and Inference of Microbial Genome Evolution
                              Franz Baumdicker
                              Head of the Independent Research Group 'Mathematical and Computational Population
                              Genetics', a joint group of Tübingen's Excellence Clusters "Controlling Microbes to Fight
                              Infections " and "Machine Learning" 
  Population genetics aims to understand how the observed genetic diversity emerged. In population genetics, many theoretical results have been developed in times where not much genomic and genetic data were available. These theory-driven results are still essential for our research, but data-driven discoveries have meanwhile dramatically changed our view of evolution and ecology, in particular for bacteria.
  The vast amount of newly sequenced genetic data leads to a multitude of interesting applications in the emerging field of machine learning in population genetics. The main challenge is that sequence data are not independent of one another, but rather are linked by their phylogenetic relationship, often represented by a tree sequence. Thus independent training data generation relies heavily on simulation procedures. I will present some of our approaches to develop, analyze, and apply supervised machine learning tools that can use this phylogenetic relationship to improve our understanding of bacterial genome evolution.
   

• 14.04.2021  Machine learning for medical image analysis and why clinicians are not using it
                             Christian Baumgartner   
                            Head of the Independent Research Group 'Machine Learning in Medical Image Analysis' at our  
                            Cluster of Excellence  "Machine Learning"
   

• 03.03.2021    Challenges in Renewable Energy Systems: A (mostly) Probabilistic Perspective
                             Nicole Ludwig
                            Head of the Early Career Research Group 'Machine Learning in Sustainable Energy Systems
                            (MSES)' at our Cluster of Excellence  "Machine Learning"
   

• 03.02.2021   Scalable Bayesian Inference: New Tools for New Challenges
                            Robert Bamler
                            Professorship for "Data Science and Machine Learning"
                            at our Cluster of Excellence  "Machine Learning"
                            Abstract
   

• 13.01.2021    Generalizing from sparse data and learning from other people - Charley Wu
                            -ONLINE-
                            Head of the Independent Research Group "Human and Machine Cognition"
                            at our Cluster of Excellence  "Machine Learning"
                            Abstract
   

• 02.12.2020     Morals and Methodology - Konstantin Genin
                            -ONLINE-
                            Head of the Independent Research Group "Epistemology and Ethics of Machine Learning"
                            at our Cluster of Excellence  "Machine Learning"
   

• 14.10.2020      CANCELED
   

• 01.04.2020    CANCELLED DUE TO CORONA CRISIS
                         How to be fair - The concept of fairness from a Computational Social Choice perspective
                         Britta Dorn (Host: Fabian Sinz)
                         Mathematical Structures in Computer Science, Department of Computer Science,
                         University of Tübingen.  Webpage
                         Abstract
   

• 06.05.2020      CANCELLED DUE TO CORONA CRISIS
                        Bob Williamson (Host: Ulrike von Luxburg)
                        Distinguished Researcher in Data61 and Professor at Research School of Computer Science,
                        Australian National University, Canberra
   

• 04.03.2020    Ian Couzin  -- CANCELLED
   

• 05.02.2020    Learning and artificial intelligence in the quantum domain
                         Hans Briegel (Host: Eric Raidl)
                          Institute for Theoretical Physics, University Innsbruck & Department of Philosophy,
                          University Konstanz    Webpage
                          Abstract
   

• 08.01.2020    Mining observations and climate models for detecting and attributing anthropogenic
                              climate change in the world’s water cycle
                              Lukas Gudmundsson (Host: Fabian Sinz)
                         Institute for Atmospheric and Climate Science, ETH Zürich  Webpage
                         Abstract
   

• 04.12.2019    Explaining and Representing Novel Concepts With Minimal Supervision
                              Zeynep Akata (Host: Fabian Sinz)
                        Cluster of Excellence "Machine Learning", Explainable Machine Learning, University Tübingen
                        Webpage
   

• 21.10.2019        Biomarker Discovery in Clinical Time Series
                                Karsten Borgwardt (Host: Fabian Sinz)                     
                        Department of Biosystems Science and Engineering, ETH Zuerich. Webpage
                       Abstract
   

• 06.11.2019       Gaussian Process emulation of tsunami and climate models                            
                               Serge Guillas (Host: Motonobo Kanagawa)                   
                         University College London / The Alan Turing Institute, UK.  Webpage 
                         Abstract
   

• 03.07.2019      Machine Learning, Neuroscience, and Spiking Neural Networks
                               Robert Legenstein (Host: Harald Baayen)
                          Inst. for Theoretical Computer Science, Graz University of Technology, Austria 
                          Webpage
                          Abstract

Workshop "Philosophy of Science Meets Machine Learning"
 

 The Workshop takes place in person in Tübingen

    on Nov.   9 - 10 in the Alte Aula (Münzgasse 30)

    on Nov. 11 - 12 in the Max Planck Institute for Intelligent Systems (Max-Planck-Ring 4)

The workshop is organised by the ‘Ethics and Philosophy Lab’ of the Cluster of Excellence ‘Machine Learning: New Perspectives for Science’ at the University of Tübingen (Workshop Convenors: Thomas Grote, Thilo Hagendorff, Eric Raidl).

Registration: Space is limited. Guests please register via: rebigtimspam prevention@googlemail.com

Machine learning does not only transform businesses and the social sphere, it also fundamentally transforms science and scientific practice. The workshop focuses on that latter issue. It aims to discuss whether and how exactly recent developments in the field of machine learning potentially transform the process of scientific inquiry. For this, it sets out to analyse the field of machine learning through the lenses of philosophy of science, epistemology, research ethics and cognate fields such as sociology of science. The workshop will bring together philosophers from different backgrounds (from formal epistemology to the study of the social dimensions of science) and machine learning researchers. The workshop`s central topics are:

1. A critical reflection on key-concepts, such as ‘learning’, ‘inference’, ‘explanation’ or ‘understanding’.
2. The implications of machine learning for the special sciences, e.g. cognitive science, social science or medicine.
3. The ethics of machine learning-driven science, e.g. the moral responsibilities of researchers.
4. Social aspects of machine learning-driven science, e.g. the impact of funding structures on research.

 Program

DAY 1: November 9 - Alte Aula

DAY 2: November 10 - Alte Aula

DAY 3: November 11 - MPI-IS

DAY 4: November 12 - MPI-IS

| Bandit learning with delays in Non-stationary environments |

Claire Vernade

DeepMind, London, UK
 

WHEN:      Thursday, 30. September 2021, 9:00 am
WHERE:     Hybrid-Event:
Claire will give her talk in the Lecture Hall, AI Building, Maria-von-Linden-Straße 6. Due to the current Covid restrictions, seats are limited. If you are a PI of our Cluster and want to join in person, please contact Elena Sizana to register. You need to be fully vaccinated, tested or recovered.
We invite everyone else to participate via Zoom: https://zoom.us/j/99906352769

ABSTRACT

We consider the problem of learning with delayed bandit feedback, meaning by trial and error, in changing environments. This problem is ubiquitous in many online recommender systems that aim at showing content, which is ultimately evaluated by long-term metrics like a purchase, or a watching time.  Mitigating the effects of delays in stationary environments is well-understood, but the problem becomes much more challenging when the environment changes. In fact, if the timescale of the change is comparable to the delay, it is impossible to learn about the environment, since the available observations are already obsolete. However, the arising issues can be addressed if relevant intermediate signals are available without delay, such that given those signals, the long-term behavior of the system is stationary. To model this situation, we introduce the problem of stochastic, non-stationary and delayed bandits with intermediate observations. We develop a computationally efficient algorithm based on UCRL, and prove sublinear regret guarantees for its performance.

Workshop "Simulation-based Inference for scientific discovery"

This workshop will be taking place online on Zoom.

The workshop is jointly organized by the ML⇌Science Colaboratory, the MLS Chair (Jakob Macke) and Helmholtz AI.

-----------------------------------------------------------------------------------------------------------------------------------------------------------------------

Registration:
Apply for the workshop by filling out the Registration Form  Deadlin for application is Aug 31, 2021.

We can only select 20 participants for the workshop. To come up with that list and make a final decision, we will form a small selection board. We will inform successful applicants subsequently.

-----------------------------------------------------------------------------------------------------------------------------------------------------------------------

You use simulation in physics, economics, archaeology or any other domain of your choice? You want to find the simulator's parameters that best fits the observations? Then simulation-based inference is something for you!

We teach a practical simulation-based inference workshop to help you understand recent machine learning techniques and apply them to your problem.

Apply to learn, have fun, and participate in a supportive and inclusive community. The workshop will combine lectures and practical hands-on sessions by experts in the field. We strive to provide a seamless computing environment for you to focus on the content rather than in import errors.

We plan to cover the following topics:

| Program

Machine Learning meets Environmental Science

Friday, September 25, 2020
 

Meeting Venue
Neue Aula, Audimax, Geschwister-Scholl-Platz

Organizers
Prof. Martin Butz (ML Cluster, Dep. Of Computer Science)
Prof. Christiane Zarfl (Center for Applied Geoscience)

Registration
Registration is required by Email until Sept 22, 2020.

___________________________________________________________________________________________________________________________________________

PROGRAM

09:00    Welcome and Introduction: Cluster of Excellence – Machine Learning for Science
              Philipp Berens
09:10     ML Transfer Center and simulation based inference
              Álvaro Tejero-Cantero & Jakob Macke
09:40     Day-ahead optimization of production schedules for saving electrical energy costs
              Thomas Stüber & Michael Menth
10:00     Uncovering hidden structure in climate data
               Bedartha Goswami
10:20    Coffee Break
10:40    Learning spatiotemporal distributed generative graph neural networks
              Martin Butz
11:00    Modeling environmental processes in rivers
             Christiane Zarfl
11:20    Can plants learn? Coupling models and data in eco-evolutionary research
             Sara Tomiolo & Maximiliane Herberich (from Katja Tielbörger’s group)
11:40    Improving the spatial prediction of soil organic carbon content in two contrasting climatic regions by
             stacking machine learning models and rescanning covariate space
             Ruhollah Taghizadeh-Mehrjardi (from Thomas Scholten’s group)
12:00    The climate situation: Facts and Necessities
             Harald Baayen
12:05    Snack Break
12:30    ML and mobile robots in environmental science
             Andreas Zell
12:50    Turbulent transport of energy, momentum and matter by large data sets
              obtained from airborne probing of the lower atmosphere
             Jens Bange
13:20    Status of the CRC 1253 CAMPOS - Catchments as Reactors:
             Metabolism of Pollutants on the Landscape Scale
             Christiane Zarfl
13:40    Funding options in the ML Cluster of Excellence and beyond
             Tilman Gocht
14:00   End

_______________________________________________________________________________________________________________________________________________________________________________________________

  Aim of the day

Environmental science studies spatio-temporal dynamics of various processes and on different topics, including climate and weather, geology, hydrology, vegetation and agriculture, various forms of pollution (e.g. of organic pollutants), to name just a few. In all these cases, multiple, often interrelated data sources are available at varying degrees of spatial and temporal granularity. Moreover, human activities, such as river dam building, CO2 release, plantations, industry etc., strongly influence the unfolding dynamics. Critical principles – such as basic laws in physics – apply universally in such systems. Environmental science has strong expertise in modeling the underlying processes – typically by systems of partial or ordinary differential equations.

As a result, besides the expertise about the underlying processes, environmental science offers two types of data – real-world data as well as data from the respectively available models of the considered environmental system. This offers essentially the perfect basis for a meaningful, science-driven application of ML algorithms. On the one hand, the parameters of the differential equations may be optimized more effectively by means of state-of-the-art gradient-based approximation approaches from ML. On the other hand, the available models may be augmented or fully substituted by distributed spatio-temporal, generative neural network approaches, such as convolutional networks, graph networks, autoencoders, recurrent neural networks, and combinations thereof.

Seeing that models are available to pre-train and analyze potentially applicable ML architectures, expertise is available to tune these models to the actual underlying processes, and that real world data is available to further train and test the generalizability of these ML architectures, it is time that ML meets Environmental Science! The aim is to foster collaboration with a focus on two main potential strands. First, available models of differential equations and involved prior situation assumptions may be optimized by means of state-of-the-art ML techniques. Second, ML techniques and particularly distributed, generative artificial neural networks may be designed to infer the processes and structures that generate particular data patterns, thus enabling (i) the fast, efficient, and accurate simulation of environmental processes and (ii) the consideration of impacts of human actions, including the potential to derive optimal actions for steering the environmental system towards a desired (stable / homeostatic) direction.

Tangles: from graph minors to identifying political mindsets

Talk by Reinhard Diestel, University of Hamburg, Department of Mathematics

WHEN: Friday, 24.01.2020 at 10:00
WHERE: Lecture hall ground floor, MPI for Intelligent Systems
 

ABSTRACT

Traditional clustering identifies groups of objects that share certain qualities. Tangles do the converse: they identify groups of qualities that often occur together. They can thereby discover, relate, and structure types of phenomena: of behaviour, political views, texts, or bacteria. Tangles can identify key phenomena that allow predictions of others. Tangles also offer a new paradigm for clustering in large data sets. Tangle clusters are, by necessity, fuzzy: they tell us where in a large structure a cluster lies, which key properties of data points identify it, and how the overall data set is structured with respect to these clusters. But they do this without needing, or attempting, to assign individual points to any cluster. Tangles of graphs are central to the theory of graph minors developed by Robertson and Seymour for their celebrated proof of the graph minor theorem. For many years, however, algorithmic applications of graph minor theory were largely confined to applications of tree-decompositions, an overall structure dual to the existence of large tangles. Very recently, tangles have been axiomatised in a way that makes them directly applicable to a much wider range of contexts than graphs, even outside mathematics. This talk will outline how this works, with an emphasis on the basic concepts of abstract tangle theory and how these are applicable in real-world scenarios. No knowledge of graph minor theory will be needed.

Machine Learning meets Social Science

Max Planck Institute for Intelligent Systems, Lecture Hall ground floor
Max-Planck-Ring 4, 72076 Tübingen
 

Registration: If you plan to attend the meeting, please register by sending an email to Sebastian Schwenk.
 

  09:00 - 10:30  Session "ML Cluster" contenteditable="false"
----------------------------------------------------------

09:00 - 09:20    Introduction  &  Fairness in Machine Learning
                          Ulrike von Luxburg
09:20 - 09:40   On the Integration of Machine Learning into Healthcare
                          Thomas Grote
09:40 - 10:00   Explaining Neural Network Decisions with
                          Minimal Supervision         
                          Zeynep Akata
10:00 - 10:20   Explainability & Explanation
                          Eric Raidl

10:30 - 11:00 Coffee Break

  11:00 - 12:30  Session "Sociology"
--------------------------------------------------------

11:00 - 11:20   What does ML do? Key questions from a sociology of
                          technology and science perspective
                          Renate Baumgartner
11:20 - 11:40    (How) does ML affect the shop floor and the
                          sociology of work?
                          Werner Schmidt
11:40 - 12:00   Three Worlds of AI – How political strategies differ
                          Daniel Buhr
12:00 - 12:20   Classifying scientific texts with supervised learning algorithms
                          Steffen Hillmert

12:30 Discussion and End

Cluster Conference "Machine Learning in Science"

Monday, July 22, 2019 | 09:00 am - 07:00 pm | Alte Aula

Tuesday, July 23, 2019 | 09:00 am - 04:00 pm | Pfleghofsaal

Registration: If you would like to attend the meeting, please register by July 15 latest by sending an email to Sebastian Schwenk. Please indicate on which day(s) you would like to participate.

Programm

Monady, July 22, 2019

Alte Aula, Münzgasse 30, 72070 Tübingen

Tuesday, July 23, 2019

Pfleghofsaal, Schulberg 2 (Pfleghof), 72070 Tübingen

Machine Learning meets Physics

AI Building, Lecture Hall (ground floor)

Maria von Linden Str. 6, 72076 Tübingen
 

9:00 – 9:25:   Frank Schreiber et al: “Analysis of X-ray Scattering Data Using Artificial Neural Networks”
9:25 – 9:50:   Hendrik Lensch: “Deep Learning on Unstructured Point Clouds”
9:50 – 10:15: Martin Oettel: “Density functionals from machine learning”

10:15 – 10:45: Coffee break

10:45 – 11:10: Miriam Klopotek: “Variational autoencoders put up to the test in learning a statistical-mechanical model system”
11:10 – 11:35: Georg Martius: “Machine Learning for Equation Identification”
11:35 – 12:00: Andreas Zell: “ML, Physics and Robotics”

12:00 – 13:00: Discussion

Machine Learning meets Law, Neue Aula
 

9:00 Stefan Thomas: Algorithms and Antitrust: How can the law make sure that machine learning does not impede competitive freedom?

9:15 Thilo Hagendorff: Regularory Needs in the Field of AI - From Ethics to Policies

9:30 Thomas Grote: The ethics of (expert-level) algorithmic decision-making

9:45 Isabel Valera: Fairness in Machine Learning

10:00 Oliver Kohlbacher: Legal issues related to AI in medicine

10:15 Discussion as long as we want

Internal Meeting of the Cluster "Machine Learning in Science": November 12-13, 2018

Meeting location: Ground floor lecture hall at the Max-Planck Institute for Intelligent Systems (directions)

Preliminary schedule:

Nov 12th
=========
9:00-10:00   Welcome, information & organisation
                     Ulrike von Luxburg and Philipp Berens
10:00-12:15   Short introductory talks of new group leaders
10:00-10:15   Jörg Stückler
10:15-10:30   Falk Lieder

10:30-11:00   Coffee break

11:00-11:15   Georg Martius
11:15-11:30   Britta Dorn
11:30-11:45   Fabian Sinz
11:45-12:00   Zhaoping Li
12:00-12:15   Gabriele Schweikert
12:15-12:30   Augustin Kelava
12:30-12:45   Michael Krone

12.45 -14:00   Lunch

14:00-15:00   Spotlights for open questions
                       (all PIs: please prepare exactly 1 slide (3 minutes) and send it to Alla at latest Nov 11)
15:00-15:30   Coffee break
15:30-18:00   Work phase for project teams

18:30   Dinner at Hofgut Rosenau

Nov 13th
========
9:00-10:30   Discussion of open questions, directions, ideas for how
                     the Excellence Cluster should start and work

10:30-11:00   Coffee break

11:00-12:00   Discussion and work phase

12:00-14:00   Lunch

14:00-15:00   Presentations of project ideas and discussion

15:00-15:30   Coffee break

Download program