Machine Learning Talks on Campus - Past events

Workshop on Avant-Garde AI Research

Obviously, the development of artificially intelligent systems is not driven by one discipline alone, but rather by an interplay of several different sciences and disciplines. This is why we are delighted to present the Avenues of Intelligence 2024 workshop – a workshop showcasing some of the most intriguing academic branches currently working on artificial intelligence.

Turning the Aesopian fable of the fox and the stork, who served each other food in dishes they couldn't access, on its head, we’ve designed this workshop to ensure that all talks are accessible to a broad academic audience.

We will feature presentations on the latest developments and theories of intelligent systems from multiple perspectives: neuroscience, philosophy, computer science, psychology, engineering, and cognitive science.

Interested? Join us by https://www.kip.uni-heidelberg.de/aoi2024/register for the workshop! And download the workshop program https://antonio.bikic.io/documents/workshop_schedule_aoi_2024.pdf.

Journalism and AI - how do both get along with each other?
Dr. Felicitas Mokler • Journalist in Residence 2024 at the Heidelberg Institute for Theoretical Studies (HITS)

The recent development of AI-based software to create texts, images or even videos has an increasing impact on the work of journalists. This talk will outline benefits as well as concerns and risks of this emerging technology for journalism.

IWR Colloquium Summer Semester 2024

“Robust Optimization Approaches for PDE-constrained Optimization under Uncertainty"
Prof. Stefan Ulbrich Department of Mathematics, Technical University of Darmstadt

The consideration of uncertainty in optimization problems is an important aspect in many practical applications. In this talk we consider optimization problems with partial differential equation (PDE) constraints, where uncertainty occurs in parameters (for example material parameters, usage scenarios) as well as in the optimization variables (e.g. manufacturing tolerances). There exist several approaches to handle uncertainty in optimization, in particular stochastic optimization, probabilistic constraints and (distributionally) robust optimization. We will focus on robust and distributionally robust optimization approaches for problems with expensive PDE-constraints. In robust optimization, relevant realizations of the uncertain variables are described by an uncertainty set and it is required that the optimization variables satisfy critical constraints for all realizations of the uncertain variables in the uncertainty set. In distributionally robust optimization one considers an ambiguity set of relevant distributions for uncertain parameters and requires that the constraints are satisfied in expectation for all distributions in the ambiguity set. The resulting robust formulation is a semi-infinite optimization problem and can be written in a min-max form. We present various techniques how to approximate and reformulate the robust problem by using optimality conditions and duality theory in order to obtain a tractable problem. Depending on the approach, the reformulation contains conic constraints, complementarity conditions and conjugate functions. We will sketch suitable solution methods. Moreover, we will illustrate, how adjoint techniques as well as model order reduction and error estimation can be used to solve these problems efficiently in the PDE constraint case. We present several application examples, in particular the design optimization of electric machines under uncertainty.

Mathematikon • Conference Room / 5th Floor

In addition it will be streamed via Zoom
Zoom meeting: https://zoom.us/j/98119222049

Machine learning galore!

Lab presentations
Roland Herzog
Georgia Koppe
Anna Kreshuk

Rocket science
Evelyn Herberg (Herzog lab)
Optimization in Machine Learning

Manuel Brenner (Koppe lab)
Creating digital twins of social Interaction partners

Peter Park (Kreshuk lab)
Seeing biology beyond the physics of light

To help plan the catering, please register for free by June 24th! 

"Meet Your Scientific Hero Seminar Series`" AIH-ELLIS
AI Visionaries Up Close
David Van Laere 
Neonatologist and Founder University of Antwerp Innocens BV

Charlie Beirnaert
Chief Technology Officer and Founder Innocens BV

“Development of a Machine Learning Model for Late-onset Sepsis
Prediction in Preterm Infants: A Journey of Interprofessional
Collaboration from the Idea towards a Licenced Medical Product”

Christoph.Schwarz@med.uni-heidelberg.de

Hörsaal D
Kinder- Frauen- und Hautklinik
Im Neuenheimer Feld 440
69120 Heidelberg 

Registration required 

Cohomology classes in the RNA transcriptome
Kelly Spry Maggs (EPFL)

Single-cell sequencing data consists of a point cloud where the points 
are cells, with coordinates RNA expression levels in each gene. Since 
the tissue is destroyed by the sequencing procedure, the dynamics of 
gene expression must be inferred from the structure and geometry of the 
point cloud. In this talk, we will build a biological interpretation of the 
one-dimensional cohomology classes in hallmark gene subsets as 
models for transient biological processes. Such processes include the 
cell-cycle, but more generally model homeostatic negative feedback 
loops. Our procedure uses persistent cohomology to identify features, 
and integration of differential forms to estimate the cascade of genes 
associated with the underlying dynamics of gene expression.

Structures Cluster of Excellence

Probabilistic Machine Learning and Latent Variable Models for Molecular Data
Junior Prof. Britta Velten, Centre for Organismal Studies (COS) & Interdisciplinary Center for Scientific Computing (IWR)

Nowadays researchers can obtain quantitative measurements of the abundances and activities of different types of molecules across experimental conditions and spatio-temporal resolutions. In this talk, I will discuss how latent variable models can help us to extract major underlying patterns from the noisy, heterogeneous and high-dimensional data and how to accommodate the statistical properties of molecular data in such models. Case studies will demonstrate how this can help us to highlight core biological processes along development, in health and disease and pinpoint underlying molecular drivers.

Mathematikon conference room (INF 205, 5th floor, 5/104)

Zoom meeting:
https://zoom.us/j/98119222049

Transformation des Kunden- und Service-Managements durch künstliche Intelligenz

Der Einsatz von generativer KI ist ein „Game Changer“ für viele Arbeitsbereiche – so auch im Kunden- und Service-Management. Doch wo macht der Einsatz bereits heute Sinn und wo sind die Grenzen? Die performio GmbH aus Brühl unterstützt hierbei über 500 Geschäftskunden in Deutschland – vom kleinen SMB bis hin zu Enterprise Kunden. Nico Hemker, Geschäftsführer, verdeutlicht den Einsatz und Mehrwert von KI anhand von Kundenbeispielen und zeigt Ihnen Einstiegschancen in diesem Bereich auf.

Die Veranstaltung findet online statt.

Anmeldung auch per E-Mail an:  simone.lasser@uni-heidelberg.de
Betreff: Performio

Gesellschaftliche Perspektiven der Künstlichen Intelligenz
Vortrag in der Reihe "Unsupervised Thinking / Unüberwachtes Denken"

Prof. Dr. Vincent Heuveline erläutert, wie ein Algorithmus lernt und welche Entscheidungen man ihm übertragen kann. Barbara Klauß (RNZ Wirtschaftsredaktion) moderiert den Abend.

SCIENCE SPARKS STARTUPS
ARTIFICIAL INTELLIGENCE EDITION
Conference | Heidelberg

*/What to Expect:/*
• Exchange with leading AI experts from the Heidelberg ecosystem

• Inspiring talks on trends and developments

• Exclusive networking opportunities with other innovative startups and established companies

Business Development Center Heidelberg (Heidelberg BDC), Carl-Friedrich-Gauß-Ring 5, 69124 Heidelberg <https://maps.app.goo.gl/NBxkDrMpkZWYctTG6>

Community Forum Data-Intensive Computing (DIC)
The special topics for this meeting are machine learning (ML) and artificial intelligence (AI)

Agenda
* News about our services
* Presentation of individual aspects related to ML/AI
* Short contributions from users related to ML/AI:
 * KI-Morph - A Platform for petabyte-scale image analysis (Alexander Zeilmann, Engineering Mathematics and Computing Lab, Heidelberg University)
 * Medieval chatbot ParzivAI - challenges and learnings on the road from concept to prototype (Thomas Renkert, Heidelberg School of Education & Heidelberg Center for Digital Humanities)
 * Application of bwVisu in teaching (Max Seitz, Institut für Medizinische Informatik & Artificial Intelligence in Cardiovascular Medicine)
* Discussion / Exchange

https://audimax.heiconf.uni-heidelberg.de/jc4e-wurm-a4hr-nccp

Chemical Compound Space Conference (CCSC2024)

CCSC2024 will bring together the vibrant community of chemists, physicists, and data scientists who employ quantum machine learning to gain physics-informed understanding of the vast chemical compound space. We have assembled a star-studded lineup of keynote and invited speakers, and hope to see many 

ELLIS Life / NCT Data Science seminar / Heidelberg.ai 
Artificial intelligence-based biomarkers in precision oncology
Jakob Nikolas Kather
TUD Dresden University of Technology

Precision oncology requires complex biomarkers which are often based on molecular and genetic tests of tumor tissue. For many of these tests, universal implementation in clinical practice is limited. However, for virtually every cancer patient, pathology tissue slides stained with hematoxylin and eosin (H&E) are available. Artificial intelligence (AI) can extract biomarkers for better treatment decisions from these images. This talk will summarize the state of the art of AI in oncology for precision oncology biomarkers. It will cover the technical foundations, emerging use cases and established applications which are already available for clinical use.

Oliver Stegle, DKFZ

Organization:
Doreen Heckmann-Nötzel (DKFZ)

Helmholz Imaging Conference 

Keynote Speaker:
Zeynep Akata is a Liesel Beckmann Distinguished professor of Computer Science at Technical University of Munich and the director of the Institute for Explainable Machine Learning at Helmholtz Munich.

Joost Batenburg is a professor at Computer Science institute of Leiden University (LIACS) and his chair is Imaging and Visualization. He is also affiliated with the national research institute for mathematics and computer science in the Netherlands (CWI) and is program director of the interdisciplinary research program Society, Artificial Intelligence and Life Sciences (SAILS).

The call for abstracts is open.

Registration is still open.

Live Q&A with OpenAI: AI and the Future of Humanity

Public Viewing 

Are you interested in frontier AI systems, their astonishing capabilities and risks for humanity? Then join us for a thought-provoking deep dive and OpenAI Live Q&A on AI safety.

Agenda:
19:00 - 19:05: Doors open
19:05 - 19:30: Introduction to AI Safety
19:30 - 20:15: Presentation & Live Q&A with OpenAI researcher Jan H. Kirchner (online), co-author of weak-to-strong generalization paper
20:15 - 20:30: Closing talk – What can we do?
20:30 - onward: Optional socializing and small group discussions with free drinks and snacks.

Confirm your spot by filling external form and prepare your most burning questions.

LARGE LANGUAGE MODELS (LLMS) IN DER PRAXIS: ANWENDUNGSBEISPIELE UND ERFOLGSSTRATEGIEN

In diesem Vortrag erfahren Sie, wie generative künstliche Intelligenz (KI) bereits heute verschiedene Industriebereiche revolutioniert. Mithilfe konkreter Anwendungsbeispiele von Large Language Models (LLMs) werden die  vielfältigen Einsatzmöglichkeiten dieser Technologie in der deutschen Industrie verdeutlicht. 

Neben der Erläuterung grundlegender Konzepte erhalten Sie wertvolle Ratschläge für eine erfolgreiche Implementierung. Gemeinsam werden die Chancen und Herausforderungen, die sich aus der Nutzung von LLMs ergebe, diskutiert. 

Die Veranstaltung findet online statt.

Der Anmeldelink ist auf der Website.

## Cancelled ##

STRUCTURES JOUR FIXE
Tensors and deep neural networks for multidimensional function approximations
Prof. Ivan Oseledets, ITWM Kaiserslautern

In this talk, I will discuss the idea of tensor decompositions and their applications for efficient function approximation. Another popular approach are neural networks and neural operators. I will discuss interesting problems and algorithms that arise while using them.

Pretalk by Jakob Zech starts at 1 pm

Philosophenweg 12, Großer Hörsaal
or via Zoom

## Cancelled ##

Zoom-Link:
https://zoom.us/j/93565493662?pwd=TWJFeTVFanVqeHVlWmFQMSt2UzdVZz09

ZOOM: Meeting ID: 935 6549 3662, Code: 928036

Machine learning galore!

Lab presentations
Tobias Buck
Holger Fröning
Denis Schapiro

Rocket science
William Oliver (Buck Lab)
Machine Learning for cosmological simulations

Hendrik Borras (Fröning lab)
Probabilistic Photonic Computing with Chaotic Light

Miguel A. Ibarra Arellano (Schapiro lab)
Finding broken cells with AI and computer vision

To help plan the catering, please register for free by April 23th! 

HITS Colloquium “AI at HITS” – Symposium in honor of Wilfried Juling

Speakers at the symposium are:
 Carsten Könneker, HITS Stiftung
 Pascal Friederich, Karlsruhe Institute of Technology (KIT)
 Ganna (Anya) Gryn´ova, HITS and Heidelberg University
 Sebastian Lerch, Karlsruhe Institute of Technology (KIT) and HITS

REGISTRATION: 
You can find details about registration on the website.

Studio Villa Bosch

Schloss-Wolfsbrunnenweg 33
Heidelberg, 69118 Germany

From subcellular mapping to modeling the human cell
Prof. Emma Lundberg

Biological systems are functionally defined by the nature, amount and spatial location of the totality of their proteins. We have generated an image-based map of the subcellular distribution of the human proteome and showed that there is great complexity to the subcellular organization of the cell giving rise to potential pleiotropic effects. As much as half of all proteins localize to multiple compartments and around 20% of the human proteome shows temporal variability. Our temporal mapping results shows that cell cycle progression explains less than half of all temporal protein variability, and that most cycling proteins are regulated post-translationally, rather than by transcriptomic cycling. 
This work is critically dependent on computational image analysis, and I will discuss machine learning approaches embedding of spatial subcellular patterns and how such embeddings as well as generative AI can be used to build multi-scale models of cell architecture. 
In summary, I will demonstrate the importance of spatial proteomics data for improved single cell biology and present how the freely available Human Protein Atlas database (http://www.proteinatlas.org) can be used as a resource for life science.

DKFZ Communication Center 
Im Neuenheimer Feld 280
69120 Heidelberg

Registration
The talk will be preceded by an opportunity to meet the speaker and ask questions about her work and career in an informal setting. 
Spaces are limited - so please register early. We look forward to your registrations by March 13.

“Meet Your Scientific Hero" Seminar Series: AIH-ELLIS
Prof. Emma Lundberg

The AIH-ELLIS “Meet Your Scientific Hero” Research Seminar Series called upon regional researchers to nominate their favorite AI visionaries from around the world to give a keynote presentation. Through this program, five AI experts will give a talk in the Heidelberg – Mannheim AI community. 
In the first talk of the series, join Dr. Loan Vulliard (DKFZ postdoc) in welcoming his scientific hero, Prof. Emma Lundberg. We look forward to an afternoon of exciting scientific exchanges!

AIH Mathematikon B
Berliner Str. 41
69120 Heidelberg

Registration
The talk will be preceded by an opportunity to meet the speaker and ask questions about her work and career in an informal setting. 
Spaces are limited - so please register early. We look forward to your registrations by March 13.

NCT Data Science seminar / ELLIS Life / Heidelberg.ai 
Mihaela van der Schaar, University of Cambridge,UK 
BEYOND CAUSALITY:
DISCOVERING AND ANALYZING THE GOVERNING EQUATIONS OF MEDICINE

In this talk, I will present several cutting-edge machine learning methods developed in our lab which enable us to discover and analyze the governing equations of medicine, moving beyond traditional causal discovery methods. The focus will be on how this new way of modeling medical and biological processes as dynamical systems offers unprecedented insights into disease progression and the efficacy of treatment strategies over time. Through real-world examples, I will illustrate the transformative impact which I believe this new strand of machine learning can play in deciphering complex medical phenomena and improving patient care.

Hosts:
Lena Maier-Hein (Chair)
Klaus Maier-Hein
Oliver Stegle

Organization:
Doreen Heckmann-Nötzel (DKFZ)

Live event

Communication Center, DKFZ, INF 280, Lecture hall

Recording planned

NCT Data Science Seminar
NEURAL AND SPECTRAL OPERATOR SURROGATES
Jakob Zech, Interdisciplinary Center for Scientific Computing at Heidelberg University

In this talk we discuss expression rates for neural network-based operator surrogates, which are employed to approximate smooth maps between infinite-dimensional spaces. Such surrogates have a wide range of applications and can be used in uncertainty quantification and parameter estimation problems in fields such as classical mechanics, fluid mechanics, electrodynamics, earth sciences etc. In this case, the operator input represents the problem configuration and models initial conditions, material properties, forcing terms and/or the domain of a partial differential equation (PDE) describing the underlying physics. The output of the operator is the corresponding PDE solution. Our analysis is based on representing the operator in- and outputs in stable bases and exploiting the resulting sparsity created by a separation in high- and low-frequency features. We show that algebraic convergence rates that are free from the curse of dimension can be achieved.

Hosts:
Lena Maier-Hein (Chair)
Klaus Maier-Hein
Oliver Stegle

Organization:
Doreen Heckmann-Nötzel (DKFZ)

Hybrid event

INF267, Bioquant, lecture hall SR41 & Zoom Recording planned

Multiscale exploration of single cell data with geometric harmonic analysis
Prof. Guy Wolf, Université de Montréal, Canada

High-throughput data collection technologies are becoming increasingly common in many fields, especially in biomedical applications involving single cell data (e.g., scRNA-seq and CyTOF). These introduce a rising need for exploratory analysis to reveal and understand hidden structure in the collected (high-dimensional) Big Data. A crucial aspect in such analysis is the separation of intrinsic data geometry from data distribution, as (a) the latter is typically biased by collection artifacts and data availability, and (b) rare subpopulations and sparse transitions between meta-stable states are often of great interest in biomedical data analysis. In this talk, I will show several tools that leverage manifold learning, graph signal processing, and harmonic analysis for biomedical (in particular, genomic/proteomic) data exploration, with emphasis on visualization, data generation/augmentation, and nonlinear feature extraction. A common thread in the presented tools is the construction of a data-driven diffusion geometry that both captures intrinsic structure in data and provides a generalization of Fourier harmonics on it. These, in turn, are used to process data features along the data geometry for interpretability, denoising, and generative purposes. Finally, I will demonstrate the application of the resulting tools in biomedical applications, such as early embryoid body development and COVID 19 mortality. 
 

ELLIS Life, NCT Data Science Seminar, Heidelberg.ai
Fred Hamprecht

Mathematikon
Conference Room / 5th Floor 
Im Neuenheimer Feld 205
69120 Heidelberg

Hybrid event:
join online or live 

Workshop in der Reihe "KI in Anwendung" 
"Custom GPTs für die eigene Lehrveranstaltung - 'Bring your own stuff' - Hands-on Workshop"

Generative KI revolutioniert Bildungssysteme weltweit. Das Potenzial großer Sprachmodelle, Wissensvermittlung individualisiert zu unterstützen, ist enorm, genauso wie ihre Risiken. Und doch steht diese Entwicklung erst ganz am Anfang - und die nächste Generation von GPT-Assistenten schon in den Startlöchern!

Anstatt die Debatte abstrakt und spekulierend zu führen, möchten wir praktisch werden. Aus diesem Grund haben sich die Heidelberg School of Education (HSE) und das Heidelberg Center for Digital Humanities (HCDH) zusammengeschlossen, um mit den Forschenden und Lehrenden der Universität Heidelberg und der Pädagogischen Hochschule Heidelberg zusammen Nutzungsszenarien für KI in der Hochschullehre (u.a. mit Fokus auf das Lehramt) zu entwickeln und zu erproben. Es sollen dazu Veranstaltungen für grundlegende Verständnisfragen auf Einsteigerniveau sowie praktische Hands-On-Workshops stattfinden, zu denen jeweils gesondert eingeladen wird und die expliziten Werkstattcharakter haben werden. Es werden zu bestimmten Veranstaltungen Expert*innen eingeladen, um bestimmte Thematiken fokussiert besprechen zu können. Die Teilnehmendenzahlen werden voraussichtlich auf ein Höchstmaß an 25 Teilnehmenden für Praxisworkshops und bis zu 50 für themenzentrierte Veranstaltungen begrenzt werden, um produktiv arbeiten zu können.

Compact course: Python Packaging
Dr. Liam Keegan, Research Software Engineer, Scientific Software Center

In this course we will learn how to package a Python library, how to publish it on PyPI and on conda-forge, as well as look at more advanced topics like building pre-compiled wheels including c++ extensions using pybind11, and automatically publishing new releases using continuous integration and cibuildwheel.

Geometric-harmonic data exploration
Prof. Guy Wolf,
CIFAR AI Chair at MILA, Montreal, and Associate Professor at Université de Montréal.
The pretalk starts at 1 pm and will be held by Prof. Fred Hamprecht. 

High-throughput data collection technologies are becoming increasingly common in many fields, especially in biomedical applications involving single cell data (e.g., scRNA-seq and CyTOF). These introduce a rising need for exploratory analysis to reveal and understand hidden structure in the collected (high-dimensional) Big Data. A crucial aspect in such analysis is the separation of intrinsic data geometry from data distribution, as (a) the latter is typically biased by collection artifacts and data availability, and (b) rare subpopulations and sparse transitions between meta-stable states are often of great interest in biomedical data analysis. In this talk, I will show several tools that leverage manifold learning, graph signal processing, and harmonic analysis for biomedical (in particular, genomic/proteomic) data exploration, with emphasis on visualization, data generation/augmentation, and nonlinear feature extraction. A common thread in the presented tools is the construction of a data-driven diffusion geometry that both captures intrinsic structure in data and provides a generalization of Fourier harmonics on it. These, in turn, are used to process data features along the data geometry for interpretability, denoising, and generative purposes. Finally, time permitting, I will relate this approach to the geometric scattering transform that generalizes Mallat's scattering to non-Euclidean domains and provides a mathematical framework for theoretical understanding of geometric deep learning. 

Philosophenweg 12, Großer Hörsaal 

You are welcome to join the meeting either in person or via Zoom.

Website: https://zoom.us/join 
Meeting-ID: 935 6549 3662
Password: 928036

Integrated surveillance and novel data streams for infectious disease outbreak prediction
Joacim Rocklöv is a distinguished figure in the field of public health and epidemiology, with a significant focus on the intersection of epidemiology, climate, and data science.
 

Pandemic preparedness and prevention focus on increasing the capacity to detect, manage and prevent infectious disease outbreaks spread between the environment, animals, and humans. To improve the preparedness systems novel surveillance data streams and data integration across sectors are required, as well as the development of decision guiding predictive models coupled with effective response mechanisms. This talk focuses on describing the state-of-the-art within the intersection of pandemic preparedness and climate-sensitive infectious disease and early showcasing a few interesting developments and applications of machine learning for sensors and surveillance. For example, in relation to mosquito smart traps, tick citizen science and IoT sensors for bioacoustics of animals. It will further discuss predictive modelling of emerging infectious diseases and provide example on what model requirements and features are important to consider within this area.

Registration:
 

heidelberg.ai

Joacim Rocklöv, Heidelberg Institute of Global Health (HIGH) & Interdisciplinary Centre of Scientific Computing (IWR)

Machine Learning and AI for the Sciences: toward understanding
Prof. Dr. Klaus-Robert Müller • Director Berlin Institute for the Foundations of Learning and Data • Machine Learning Group, Technische Universität Berlin 

In recent years, machine learning (ML) and artificial intelligence (AI) methods have begun to play a more and more enabling role in the sciences and in industry. In particular, the advent of large and/or complex data corpora has given rise to new technological challenges and possibilities. In his talk, Müller will touch upon the topic of ML applications in the sciences, in particular in chemistry and physics. He will also discuss possibilities for extracting information from machine learning models to further our understanding by explaining nonlinear ML models.  Finally, Müller will briefly discuss  perspectives and limitations.

Mathematikon
Im Neuenheimer Feld 205
Konferenzraum / 5. Stock, Raum 5/104
69210 Heidelberg

The meeting is streamed via Zoom. 

Geometric Learning via PDE-G-CNNs: Training of Association Fields
Remco Duits, Eindhoven University of Technology, the Netherlands

We consider PDE-based Group Convolutional Neural Networks (PDE-G-CNNs) that generalize Group equivariant Convolutional Neural Networks (G-CNNs). In PDE-G-CNNs a network layer is a set of PDE-solvers. The underlying (non)linear PDEs are defined on the homogeneous space M(d) of positions and orientations within the roto-translation group SE(d) and provide a geometric design of the roto-translation equivariant neural network. 
The network consists of morphological convolutions with (approximative) kernels solving nonlinear PDEs (HJB equations for max-pooling over Riemannian balls), and linear convolutions solving linear PDEs (convection, fractional diffusion). Our analytic approximation kernels are accurate in comparison to our recent exact PDE-kernels. Common mystifying (ReLU) nonlinearities are now obsolete andexcluded. We achieve network interpretability as we train sparse association fields (modeling contour perception in our own visual system).
We present several medical imaging applications (tracking and segmentation of complex vasculature) and industrial image analysis applications where each time we show benefits of PDE-G-CNNs compared to state-of-the-art G-CNNs: increase of performance along with a large reduction in network parameters and less training data.

Philosophenweg 12, Großer Hörsaal 

You are welcome to join the meeting either in person or via Zoom.

COFFEE & SNACKS IN ROOM 106

Website: https://zoom.us/join 
Meeting-ID: 935 6549 3662
Password: 928036

25.01.24
7:00 PM

Science Notes „KI und Sprache“

Künstliche Intelligenz hat die Welt auf den Kopf gestellt. Große Sprachmodelle wie ChatGPT scheinen auf alle Fragen Antworten zu haben und erzeugen unterschiedlichste Texte in nur wenigen Augenblicken. Wo geht diese Entwicklung hin? Und wo liegen die Grenzen von Künstlicher Intelligenz?

Bei den Science Notes beleuchten wir, wie Künstliche Intelligenz Sprache erzeugt und fragen, was das für uns bedeutet. Vier Forscher:innen berichten von ihrer Arbeit; mit auf der Bühne ist die Schriftstellerin Kathrin Passig.

Das Duo Ströme begleitet den Abend musikalisch an ihren modularen Synthesizern.

Machine learning galore!

Lab presentations
Tristan Bereau
Fred Hamprecht
Caroline Heneka

Rocket science
Bálint Máté (Bereau Lab)
Generative Modelling and Lattice Field Theories

Roman Remme (Hamprecht Lab)
Machine Learning meets Density Functional Theory

Lara Alegre (Heneka lab)
ML for large radio surveys

To help plan the catering, please register for free by Jan 15th! 

Discretization of diffusions via gradient flows
Błażej Miasojedow (University of Warsaw) 

We consider the non-linear Fokker-Planck equation. We show uniform in time bounds for the error of particle approximation in terms of the second Wasserstein distance. We use the variational formulation of non-linear Fokker-Planck and we show that particle approximation can be seen as an inexact gradient flow of free energy functional. To deal with non-local interaction kernel we use results from propagation of chaos theory. The talk will be based on the joint work with Matej Benko, Iwona Chlebicka and Jorgen Endal

Bayesian Inference Models for Healthcare Resilience: 
Insights and Lessons from Mexico’s COVID-19 Response
Prof. Antonio Capella Kort Instituto de Matemáticas Universidad Nacional Autónoma de México

Navigating the challenges of the global COVID-19 pandemic required strategic decisions tailored to each country’s unique circumstances. In less developed nations, the threat of overwhelming hospital capacity was especially severe. Rather than building scenarios based on mathematical models, our team used a dynamic forecasting approach. We developed a series of models that provided 4-week probabilistic forecasts, complete with uncertainty quantification. These forecasts, crucially informed by real-time data, predicted the demand for hospital beds and ventilators, which served as the backbone for decisions and public policies adopted by federal health authorities in Mexico from April 2020 to January 2022. The journey was a challenging one. An incompletely characterized virus and the unpredictable dynamics of societal behavior made crafting a useful model difficult. 
In this talk, I will present a retrospective of these models and critically review their objectives, successes, and limitations. Our methodology and modeling decisions will be presented, from the intricacies of data used for model fitting to the balance between model complexity and parameter identifiability. The predictive power and performance of our models will also be reviewed. Beyond the algorithms and forecasts, I’ll share our experiences collaborating with health authorities and communicating with the general public. Discover the highs and lows, the challenges faced, and the insights we gained.

Mathematikon
Conference Room / 5th Floor 
Im Neuenheimer Feld 205
69120 Heidelberg

The meeting is streamed via Zoom. 

Analog in-memory computing for deep learning
Dr. Abu Sebastian, Distinguished Research Staff Member & Manager, IBM Research - Zurich

Deep neural networks (DNNs) are revolutionizing the field of artificial intelligence and are key drivers of innovation in device technology and computer architecture. While there has been significant progress in the development of specialized hardware for DNN inference, many of the existing architectures physically split the memory and processing units. This means that DNN models are typically stored in a separate memory location, and that computational tasks require constant shuffling of data between the memory and processing units – a process that slows down computation and limits the maximum achievable energy efficiency. Analog in-memory computing (AIMC) is a promising approach that addressing this challenge by borrowing two key features of how biological neural networks are realized. Synaptic weights are physically localized in nanoscale memory elements and the associated computational operations are performed in the analog/mixed-signal domain.
In the first part of the lecture, I will introduce AIMC based on non-volatile memory technology. The focus will be on the key concepts and the associated terminology. Subsequently, a multi-tile mixed-signal AIMC chip for deep learning inference will be presented. This chip fabricated in 14nm CMOS technology comprises 64 AIMC cores/tiles based on phase-change memory technology. It will serve as the basis to delve deeper into the device, circuits, architectural and algorithmic aspects of AIMC. Of particular focus will be achieving floating point-equivalent classification accuracy while performing the bulk of computations in the analog domain with relatively less precision. In the final part, I will present some of the ongoing research towards the next generation of AIMC chips and provide an outlook for the future.

Tee : 4:45 PM
Magistrale des Physikalischen Instituts 

Vortrag:
Hörsaal 1 des Kirchhoff-Instituts, INF 227

An Interdisciplinary Journey of Computational Mathematics in Theoretical Chemistry
Prof. Benjamin Stamm, Institute of Applied Analysis and Numerical Simulation, University of Stuttgart.

Numerical simulations are widely used as a third pillar besides experimental and theoretical investigations in many sciences such as physics and chemistry as well as engineering science. It requires the development of robust and efficient numerical methods for the resolution of the underlying physical laws that arise often in form of partial differential equations (PDEs). In this talk, I will describe two interdisciplinary stories. While illustrating the key-ideas of the theory and methods, I will also highlight the occasions where not only the mathematical tools have been successfully developed and transferred to the application but also where the interdisciplinary interactions raised new mathematical questions and triggered new answers and theories in mathematics. From the application viewpoint, this talk will touch upon implicit solvation models and Born-Oppenheimer molecular dynamics but the methods rely on various and divers mathematical concepts from Grassmann manifolds to descriptors from machine learning and perturbation theory.
 

Mathematikon
Im Neuenheimer Feld 205
Konferenzraum / 5. Stock, Raum 5/104
69210 Heidelberg

The meeting is also streamed via Zoom. 

Recruitment Fair: Analytics in Practice
Business meets Analytics: Praktischer Einsatz von Analytics und KI in den Unternehmen

Folgende Unternehmen stellen Use-Cases für Data Science, Analytics und Künstliche Intelligenz vor:
- Commerzbank
- Fraport AG
-1&1

NCT Data Science Seminar
Integrative data analysis by combining networks, dynamical models and machine learning
Katharina Baum, FU Berlin, Hasso Plattner Institute, Icahn School of Medicine at Mount Sinai

Combining different views from complementary data layers is key for robust predictions in biomedical research. However, it is still challenging to incorporate dependencies and relationships into coherent modeling and prediction frameworks, especially when applying machine learning (ML). I focus on two approaches that I combine with ML: Network-based methods allow for representing and harnessing the interaction of entities and data layers, and dynamical models can represent temporal properties and intricate dependencies of the investigated processes. In this talk, I will present recent examples of our developed methods for integrative data analysis. These range from differential integrated multi-omics networks, over neural networks with (multi-)graph input, to infusing prior knowledge from dynamical models into ML. We tackle problems such as drug response prediction or epidemic time series forecasting in scenarios with sparse data. 

Hosts:
Lena Maier-Hein (Chair)
Klaus Maier-Hein
Oliver Stegle

Organization:
Doreen Heckmann-Nötzel (DKFZ)

Online Seminar:

You can find the Registration Link on:
https://www.dkfz.de/en/datascience/seminar/Baum.html
 

About Vision and Language models: What grounded linguistic phenomena do they understand? How much do they use the image and text modality?
Letitia Parcalabescu, Department of Computational Linguistics, Heidelberg University

In this talk, we will introduce Vision and Language (VL) models which can very well say if an image and text are related and answer questions about images. While performance on these tasks is important, task-centered evaluation does not tell us why they are so good at these tasks, such as what are the fine-grained linguistic capabilities of VL models use when solving them. Therefore, we present our work on the VALSE💃 benchmark to test six specific linguistic phenomena grounded in images. Our zero-shot experiments with five widely-used pretrained VL models suggest that current VL models have considerable difficulty addressing most phenomena. In the second part, we ask how much a VL model uses the image and text modality in each sample or dataset. To measure the contribution of each modality in a VL model, we developed MM-SHAP which we applied in two ways: (1) to compare VL models for their average degree of multimodality, and (2) to measure for individual models the contribution of individual modalities for different tasks and datasets. Experiments with six VL models on four VL tasks highlight that unimodal collapse can occur to different degrees and in different directions, contradicting the wide-spread assumption that unimodal collapse is one-sided.

Advanced AI Workflows in Higher Education - A Hands-On Workshop with Dr. Sarah Schwettman (MIT)

Anstatt die Debatte abstrakt und spekulierend zu führen, möchten wir praktisch werden.

Der Workshop wird in die brandneuen Erweiterungen verfügbarer Sprachmodelle einführen und im Detail erklären, wie deren Möglichkeiten im Kontext Hochschullehre ganz konkret eingesetzt werden können. Mit Dr. Sarah Schwettmann (Massachusetts Institute of Technology/Berkman Klein Center, Harvard) wird uns eine ausgewiesene Expertin im Bereich sprachlicher und visueller KI eine „Tiefenbohrung“ in die Materie ermöglichen, die gerade auch für die hochschulische Ausbildung künftiger Lehrer:innen (und deren späteres Arbeitsfeld Schule) zahlreiche Fragen aufwirft.

Heidelberg Centre for Asian and Transcultural Studies (CATS)
Heidelberg Centre for Transcultural Studies
Voßstraße 2 | Gebäude 4400
69115 Heidelberg
Deutschland

Raum R.400.02.12 | 2. OG

Anmeldung bei Dr. Thomas Renkert bis
23. November 2023

Predict to control
Talk by Dr. Patrick van der Smagt (Machine Learning Research Lab at VW Group) 

Mutual predictability is the key to interaction. Or in simpler terms: "experience makes teamwork".
Of course, prediction isn't all that simple. We'll look at "generative AI", but without the hype, and see how we use that to learn the dynamics of complex stochastic systems, use that to predict -- and control. And give some examples, in robotics and beyond, of where this can be used.

ZITI

Lorenzo Masia

Colloquium Laura Maria Sangalli: 
Physics-Informed Spatial and Functional Data Analysis over non-Euclidean Domains

Recent years have seen an explosive growth in the recording of increasingly complex and high-dimensional data, whose analysis calls for the definition of new methods, merging ideas and approaches from statistics and applied mathematics. My talk will focus on spatial and functional data observed over non-Euclidean domains, such as linear networks, two-dimensional Riemannian manifolds and non-convex volumes. I will present an innovative class of methods, based on regularizing terms involving Partial Differential Equations (PDEs), defined over the complex domains being considered. These Physics-Informed statistical learning methods enable the inclusion of the available problem specific information, suitably encoded in the regularizing PDE. Illustrative applications from environmental and life sciences will be presented.

Studio Villa Bosch

Schloss-Wolfsbrunnenweg 33
Heidelberg, 69118 Germany

REGISTRATION:
The talk will be hybrid.If you would like to participate online, please register in advance.

In case you are not able to attend, you can watch the talk afterwards on the HITS YouTube channel.

You can find both links on: 
https://www.h-its.org/event/colloquium-laura-maria-sangalli/

Machine learning galore!

Lab presentations
Christoph Dieterich
Sandy Engelhardt
Britta Velten

Rocket science
Adrian Chan (Dieterich lab)
Fingerprinting your RNA - one molecule at a time

Sven Köhler (Engelhardt lab)
4D Self-Supervised Deep Cardiac Motion Modelling

Florin Walter (Britta Velten lab)
Probabilistic Factor Models for Subcellular Spatial Transcriptomics

To help plan the catering, please register for free by clicking here.

KoMSO Academy TorchPhysics: Deep Learning for partial differential equations

TorchPhysics has been jointly developed by Bosch and the University of Bremen. It aims at providing an ‘as simple as possible’ platform for testing AI concepts for solving PDEs. This workshop aims at providing an overview of the state of the art AI concepts for solving PDEs and related parametric studies/parameter identication problems. We include an introduction in using TorchPhysics as well as hands on exercises using this toolbox. Furthermore we highlight recent advances in modeling injection molding processes in industrial applications via deep learning based surrogate models.

Keynote speakers:
Prof. Dr. Uwe Iben, Robert Bosch GmbH
Prof. Dr. Dr. h.c. Peter Maaß, ZeTeM, University of Bremen
Manuel Wenzel, Robert Bosch GmbH
TorchPhysics Team University of Bremen (Janek Gödeke, Nick Heilenkötter, Tom Freudenberg)

Akademie der Wissenschaften in Heidelberg

Online participation is possible for a limited number of participants.

Registration: Registration fee for industry: 600 € (online 350 €) Registration fee for academia: 250 € (online 150 €) (Registration fee on site incl. conference dinner, 07.11.)

For registration please send an e-mail to the organization comittee: komso-academy@math.uni-bremen.de

STRUCTURES Jour Fixe
Holger Fröning with Pretalk by Guido Kanschat
On Accelerating Deep and Bayesian Neural Architectures

Deep artificial neural networks are a prominent approach for decision-making in scenarios involving uncertainty. These networks have significantly enhanced performance in various prediction tasks, such as image recognition, speech processing, and signal analysis. However, their utilization demands substantial computational resources and memory. On the other hand, there is a growing need to implement machine learning techniques on resource-constrained devices, including Internet of Things (IoT) devices, edge devices, and mobile platforms. In this talk, we will start by examining prior research focused on accelerating Deep Neural Networks (DNNs) through compression techniques, particularly quantization, pruning, and architecture optimization. While DNNs excel at operating under uncertainty, they are incapable of reasoning about uncertainty itself. Detecting situations where a neural architecture cannot provide a well-founded prediction is crucial. Consequently, probabilistic models have recently garnered significant interest. We will provide a brief overview of these models and discuss potential avenues to address their substantially increased computational demands.

Philosophenweg 12, Großer Hörsaal
and online via Zoom 
Zoom link: https://zoom.us/j/93565493662?pwd=TWJFeTVFanVqeHVlWmFQMSt2UzdVZz09

Meeting-ID: 935 6549 3662
Password: 928036

I4L Lecture by Karsten Borgwardt - Director MPI München

Department of Machine Learning and Systems Biology

AI InScide Out Unconference

This event offers the opportunity for leading scientists in AI applied to health, life, and natural sciences to meet and to have interdisciplinary exchanges. Join us for a vibrant community meeting that will bring together AI scientists in the areas of bioinformatics, imaging, structural biology, physics (and more), to stimulate new collaborations and spark boundary-pushing discussions. AI InScide Out will feature keynote talks from distinguished scientists, presentations from researchers associated with AI Health Innovation Cluster and ELLIS-Life Heidelberg, and flash talks from submitted abstracts. The unconference session will be distinct from a regular conference, providing ample opportunities for scientific exchange and informal discussions among all participants.  We invite researchers from all career stages to register. You will have the opportunity to present your own work, either focused on a scientific advance, failure and challenge, or novel opportunities for scientific questions that are ready to be tackled using AI. The event is free of charge but registration is mandatory. Please register online by September 15: https://indico.dkfz.de/e/ai. Admission will be on a first come first-serve basis with priority given to submissions that include a (short) scientific abstract. 

ELLIS Unit Heidelberg, AI Health Innovation Cluster

The program committee: Oliver Stegle, Klaus Maier-Hein, Britta Velten, Julio Saez-Rodriguez, and Ullrich Köthe

ELLIS Life, Heidelberg.ai, NCT Data Science Seminar

Image Denoising and the Generative Accumulation of Photons
Speaker: Alex Krull, Lecturer University Birmingham 

"Shot noise is a fundamental property of many imaging applications, especially in fluorescence microscopy. Removing this noise is an ill-defined problem since many clean solutions exist for a single noisy image. Traditional approaches aiming to map a noisy image to its clean counterpart usually find the minimum mean square error (MMSE) solution,i.e. , they predict the expected value of the posterior distribution of possible clean images. We present a fresh perspective on shot noise corrupted images and noise removal. By viewing image formation as the sequential accumulation of photons on a detector grid, we show that a network trained to predict the where the next photon could arrive is in fact solving the traditional MMSE denoising task. This new perspective allows us to make three contributions:
(i) We present a new strategy for self-supervised denoising. (ii) We present a new method for sampling from the posterior of possible solutions by iteratively sampling and adding small numbers of photons to the image. (iii) We derive a full generative model by starting this process from an empty canvas."

Hosts:
Lena Maier-Hein (Chair), 
Klaus Maier-Hein, Oliver Stegle

Organization:
Doreen Heckmann-Nötzel (DKFZ)

Online Meeting
Meeting ID 616 0108 8597

Code 11031614

Registration Link:
https://dkfz-de.zoom-x.de/meeting/register/u5Urd-iopjMvE9M_jv-9OVi5zZFbcIaX0LLs

NCT Data Science Seminar: QI DOU Image-Based Robotic Surgery Intelligence  Department of Computer Science & Engineering at The Chinese University of Hong Kong

With rapid advancements in medicine and engineering technologies, the operating room has evolved to a highly complex environment, where surgeons take advantage of computers, endoscopes and robots to perform procedures with more precision while less incision. Intelligence, together with its authorized cognitive assistance, smart data analytics, and automation, is envisaged to be a core fuel to transform next-generation robotic surgery in numerous known or unknown exciting ways. In this talk, I will present ideas, methodologies and applications of image-based robotic surgery intelligence from three perspectives, i.e., AI-enabled surgical situation awareness to improve surgical procedures, AI-powered large-scale data analysis to enhance surgical education, AI-driven multi-sensory perception to achieve surgical subtask automation..

Hosts Veranstaltungsserie: Lena Maier-Hein (Chair), Klaus Maier-Hein, Oliver Stegle

Organization: Doreen Heckmann-Nötzel (DKFZ)

https://eu02web.zoom.us/meeting/register/u5Ytd-mqqzgsEtZKSuQ7EnYWH7wW_z

Big PyData BBQ #5: Large Language Models

If you like cool talks about 🧑‍🔬 Data Science, 🤖Artificial Intelligence, 🐍 coding or 🤗 community, the Big PyData BBQ is the place to be! 🔥
KÖNIGSWEG and hei_INNOVATION invite you to join the 5️⃣th edition of the annual big gathering of the PyData Südwest community. Besides talks there will be a lot of time for networking over a delicious 🥦+🍖 BBQ .

The event will be live streamed and published on https://www.youtube.com/@PyDataTV.

Königsweg
hei_INNOVATION

Organiser:
Alexander C.S.H.

This is a hybrid event.

In Person:
Mathematikon B,
Berliner Str. 41-49, Heidelberg 

DAGM German Conference on Pattern Recognition

The German Conference on Pattern Recognition (GCPR) is the annual symposium of the German Association for Pattern Recognition (DAGM). It is the national venue for recent advances in image processing, pattern recognition, and computer vision and it follows the long tradition of the DAGM conference series, which has been renamed to GCPR in 2013 to reflect its increasing internationalization. In 2023 in Heidelberg, the conference series will celebrate its 45th anniversary.

KI Start-Up Messe
Eintritt frei

Triff Start-Ups & Industrie im Bereich der Künstlichen Intelligenz aus der Region und Deutschland

SYMPOSIUM HEALTHY AI
AI MEETS MEDICINE: PROMISES AND LIMITS

Participation in the event is free of charge. Since the number of participants is limited to 80 people, we ask you to register as soon as possible at
https://informatics4life.org/registration-symposium-healthy-ai-2

Roof Top of the Mathematikon
Im Neuenheimer Feld 205, 69120 Heidelberg

5th Summer School in Medical Physics 2023:
Data Science and Machine Learning in Radiotherapy

This summer school is a hybrid event and is subdivided into an online phase and a hybrid phase (attendance phase or live online phase). participants can decide to follow the course online and on site or 100% virtually. The online phase with pre-recorded lectures introduces the basics in machine learning, focusing on 3D voxelized geometries, its most frequent applications, and methodological as well as ethical aspects. The live online and attendance phases will then expose how machine learning may interfere radiotherapy workflows in detail. The knowledge of deep learning methodologies for image synthesis, organ and target segmentation, and image registration in the context of radiotherapy will be expanded. Furthermore, applications in computational dosimetry and plan optimization will be discussed, ranging from dose prediction, guided plan optimization, and treatment outcome prediction. The summer school is designed for PhD/MD, MSc or BSc students. More information is available on our website: http://www.dkfz.de/summer_school2023_de.

Registration necessary:Y es
Registration deadline: 06/15/2023
Registration: https://www.dkfz.de/en/medphys/education_and_training/summer_school_2023_de.html

Carl-Zeiss-Stiftung-Summer-School 2023
Scientific Machine Learning for Astrophysics

A conference for young, motivated and interdiscplinary researches in natural science with a focus on astrophysics and machine learning

Invited Speakers: 
Nathan Kutz (U. of Washington, Seattle)
Torsten Enßlin (MPI for Astrophysics, Munich)
Patrick Kidger (Google X)
Laurence Perreault Levasseur (U. of Montreal)
Kaze Wong (Flatiron Institute, NY)
Andreea Deac (Mila, Montreal)

PiCR lecture (online): Interactive machine learning @ DKFZ
Dr. Paul Jäger, E290

This lecture is part of the Progress in Cancer Research (PiCR) lecture series April - July 2023 organized by the DKFZ International PhD Program.
Lectures will be held online via Zoom. Please log into Zoom with your full name so we can register your attendance.

DKFZ members please register via the DKFZ Training Portal: https://pseminar.inet.dkfz-heidelberg.de/LS/2046671416/INF
External participants please register online: https://indico.dkfz.de/e/picr

IWR Colloquium Summer 2023
“On Signal Processing Flaws in Convolutional Neural Networks"
Prof. Dr.-Ing. Janis Keuper • Head of the Institute for Machine Learning and Analytics (IMLA), Offenburg University, Germany

Convolutional Neural Networks (CNN) are the current backbone of deep learning architectures in a wide range of applications which process array data, like 2D/3D images. Despite this overwhelming success of the application of CNNs in terms of qualitative (visual) results and classification test set accuracies, this talk will point out some servere inherent problems of CNNs regarding their insufficient signal processing capabilities. While these “signal processing flaws” have been largely ignored as test accuracies on many problems have been increasing over many years, recent research showed that current models are highly vulnerable to even the slightest changes in input distributions. In our latest works [1-5], we showed that the missing “robustness” of CNNs is not only due to insufficient training data, but to a large extend is also caused by faulty operators and architectures which are failing to adhere to basic signal processing demands. The aim of this talk is to give an overview of the dominant problems in the context of image processing and analysis and discuss possible counter measures. A

INF 205 Mathematikon, 5th floor, Conference Room, Room 5/104

Link Zoom Meeting:
https://zoom.us/j/98119222049

AGI Summer Talks

Aleph Alpha, Europe’s leading AGI research company, is inviting you to their AGI Summer Talks, bringing together the region's most brilliant AI researchers and institutions in the picturesque city of Heidelberg to talk about the latest developments and the future of AGI.

The AGI Summer Talks will feature three talks and Q&A with AI professors including distinguished names such as Matthias Bethge and Kristian Kersting. I addition, we are honored to present an inspiring keynote titled “Are Large Language Models the last invention we need to make?“ by Joscha Bach, an influential figure in the field of AGI. The event will conclude with a panel discussion that will bring all the speakers back on stage to explore the implications and future of AGI.

ELLIS live/NCT Data Science Seminar: Tristan Bereau 
The role of multiscale modeling in molecular discovery

Advanced statistical methods are rapidly impregnating many scientific fields, offering new perspectives on long-standing problems. In materials science, data-driven methods are already bearing fruit in various disciplines, such as hard condensed matter or inorganic chemistry, while comparatively little has happened in soft matter. I will describe how we use multiscale simulations to leverage data-driven methods in soft matter. We aim at establishing structure-property relationships for complex thermodynamic processes across the chemical space of small molecules. Akin to screening experiments, we devise a high-throughput coarse-grained simulation framework. Coarse-graining is an appealing screening strategy for two main reasons: it significantly reduces the size of chemical space and it can suggest a low-dimensional representation of the structure-property relationship. I will briefly mention a biological application of our methodology that led to the discovery of in vivo active compounds. Finally, I will mention a number of ways machine learning can help fulfill the promise of connecting models at different scales.
Tristan Bereau is a computational physicist working at the interface between multiscale modeling and machine learning for soft matter and biomolecules. He earned a Ph.D. in Physics at Carnegie Mellon University. After a postdoc at the University of Basel, he led an Emmy Noether research group at the Max Planck Institute for Polymer Research. He then moved to the University of Amsterdam as an assistant professor in chemistry and computer science, followed by a role in Industry. Tristan serves on the editorial boards of the journals Machine Learning: Science & Technnology and Computational Science and Engineering. He is currently a professor at the Institute for Theoretical Physics at the University of Heidelberg.

Hosts Veranstaltungsserie:
Lena Maier-Hein (Chair), Klaus Maier-Hein, Oliver Stegle

Host Talk: Fred Hamprecht

Organization:
Doreen Heckmann-Nötzel (DKFZ)

Hybrid event

INF267, Bioquant, lecture hall SR41 

Link Zoom Meeting:
https://eu02web.zoom.us/meeting/register/u5Ytd-mqqzgsEtZKSuQ7EnYWH7wW_zDhAXcw#/registration

Colloquium Rafal Weron: Electricity price forecasting in the 2020s

Registration:
The talk will be hybrid.If you would like to participate in person, please register in advance at christina.blach@h-its.org.If you would like to participate online, please register in advance https://kta-email.zoom.us/meeting/register/tJIpc-mqrDwpEtzcAH5UxlAdcjKtne7QkPP2. After registering, you will receive a confirmation email containing information about joining the meeting.
In case you are not able to attend, you can watch the talk afterwards on the HITS YouTube channel: https://www.youtube.com/user/TheHITSters.

Studio Villa Bosch
Schloss-Wolfsbrunnenweg 33
Heidelberg, 69118 Germany

Hybrid

GIS Colloquium – Talks (Summer Term 2023)

GeoAI Research and Technology Transfer for National Mapping
Dr. Samantha T. Arundel, Center of Excellence for Geospatial Information Science, US Geological Survey, USA

GeoAI Research and Technology Transfer for National Mapping" highlights the application of Artificial Intelligence (AI) and Geographic Information Systems (GIS) for national mapping. The presentation emphasizes the use of GeoAI technology for efficient and accurate data acquisition, processing, and analysis in the GIS, Cartography and Mapping fields. Dr. Arundel will discuss the potential benefits of AI in mapping, such as reduced costs, increased accuracy, and faster mapping processes. The presentation also discusses various applications of GeoAI technology, such as image recognition, object detection, and optical character recognition (OCR). Of particular emphasis is the importance of partnerships between research institutions and government agencies to promote the adoption of AI technology in mapping. Finally, the presentation showcases some examples of successful implementation of GeoAI technology in national mapping, including the use of AI for feature extraction from various types of imagery, noise reduction in point-clouds, and OCR for knowledge extraction from historical maps.

Interpretable Representations and Neuro-symbolic Methods in Deep Learning
Jan Stuehmer (Heidelberg Institute for Theoretical Studies)

Current state-of-the-art machine learning methods impress with their capabilities for prediction, classification, or when solving even complex analytical tasks in the case of large language models. However, these methods often appear as a “black box” from the outside which makes it hard to understand how a result was achieved. In this talk, I will discuss several approaches to interpretability in machine learning. First, I will describe a method for representation learning which leads to an interpretable latent representation. Second, I will present our work at the interface between symbolic and sub-symbolic representations, so-called neuro-symbolic methods, which enable a direct interpretation of a model’s intermediate output. The talk concludes with a discussion of the relationship between interpretability and causality.

HEGL(Heidelberg Experimental Geometry Lab)

Organizer:
Diaaeldin Taha

Workshop: “Data Science and Dependence”

Registration to attend talks: by email to Prof. Dahlhaus dahlhaus@statlab.uni-heidelberg.de by July 1st,

IWH 

(International Academic Forum Heidelberg), Hauptstrasse 242,
69117 Heidelberg

Structures Jour Fixe

The role of multiscale modeling in molecular discovery
Tristan Bereau, Uni Heidelberg                                                                                     

Advanced statistical methods are rapidly impregnating many scientific fields, offering new perspectives on long-standing problems. In materials science, data-driven methods are already bearing fruit in various disciplines, such as hard condensed matter or inorganic chemistry, while comparatively little has happened in soft matter. I will describe how we use multiscale simulations to leverage data-driven methods in soft matter. We aim at establishing structure-property relationships for complex thermodynamic processes across the chemical space 
of small molecules. Akin to screening experiments, we devise a high-throughput coarse-grained simulation framework. Coarse-graining is an appealing screening strategy for two main reasons: it significantly reduces the size of chemical space and it can suggest a low-dimensional representation of the structure-property relationship. I will briefly mention a biological application of our methodology that led to the discovery of in vivo active compounds. Finally, I will mention a number of ways machine learning can help fulfill the promise of connecting models at different scales.

Philosophenweg 12, 69120 Heidelberg, Großer Hörsaal                                                                              

Zoom link: https://zoom.us/j/93565493662?pwd=TWJFeTVFanVqeHVlWmFQMSt2UzdVZz09
Website: https://zoom.us/join
Meeting-ID: 935 6549 3662
Password: 928036

GIS Colloquium – Talks (Summer Term 2023)

Spatial Optimization, Significance and Evolving GIScience
Prof. Dr. Alan T. Murray, Department of Geography, University of California at Santa Barbara, CA 93106, USA

Spatial optimization is introduced and reviewed in historical terms. The significance of spatial optimization is demonstrated through current analysis, management, planning and policy contexts focused on emergency response, food production, wildfire risk mitigation and public health monitoring.
Further, mathematical formalization in spatial optimization offers a theoretical framework to establish findings as significant. The ways in which GIScience perspectives can directly and indirectly support spatial optimization are highlighted.

Machine learning galore!

Lab presentations
Jan Budczies
Daniel Durstewitz
Carsten Rother

Rocket Science
Iordanis Ourailidis (Budczies lab)
AI-based detection of tumor budding in head and neck cancer

Florian Hess (Durstewitz lab)
Learning generative models of dynamical systems from time series data

Felix Draxler (Rother lab)
Generative Models: Applications and Guarantees

To help plan the catering, please register for free by clicking here.