15.10.2026
1:15 PM Get Together
2:15 PM Talk | Deep Learning With Missing Data
Prof. Richard Samworth • University of Cambridge (UK)
We are very happy to invite you to a special edition of our Statistical Colloquium on October 15 at 14:00 in the Mathematikon conference room, featuring Richard Samworth from University of Cambridge.
Before the talk, there will be a Get-Together at 13:15 in the common room for a lively exchange and to meet the speaker.
13:15 Get-Together (Common Room)
14:15 Talk (Conference Room)
The talk is an addition to the HGS MathComp Joint Workshop at KIT earlier this week. Accordingly, we want to emphasise that everyone is welcome, and that this is a collaborative event between HGS MathComp, the IWR and the AG Mathematical Statistics.
Abstract:
In the context of multivariate nonparametric regression with missing covariates, we propose Pattern Embedded Neural Networks (PENNs), which can be applied in conjunction with any existing imputation technique. In addition to a neural network trained on the imputed data, PENNs pass the vectors of observation indicators through a second neural network to provide a compact representation. The outputs are then combined in a third neural network to produce final predictions. Our main theoretical result exploits an assumption that the observation patterns can be partitioned into cells on which the Bayes regression function behaves similarly, and belongs to a compositional Hölder class. It provides a finite-sample excess risk bound that holds for an arbitrary missingness mechanism, and in combination with a complementary minimax lower bound, demonstrates that our PENN estimator attains in typical cases the minimax rate of convergence as if the cells of the partition were known in advance, up to a poly-logarithmic factor in the sample size. Numerical experiments on simulated, semi-synthetic and real data confirm that the PENN estimator consistently improves, often dramatically, on standard neural networks without pattern embedding. Code to reproduce our experiments, as well as a tutorial on how to apply our method, is publicly available. | AG Mathematical Statistics & HGS MathComp | Mathematikon • Common Room & Conference Room, 5th Floor • Im Neuenheimer Feld 205, 69120 Heidelberg | |
29:10:2026
4.15 PM - 6.15 PM | Statistical Inference for infinite-dimensional dynamical systems
Prof. Dr. Richard Nickl (Cambridge)
We study optimal statistical inference procedures for the states of time evolution phenomena occurring in `data assimilation’ or filtering problems. There it is a common practice to assign a Gaussian process prior on the initial condition of a dynamical system and to update it to a Bayesian posterior measure in the space of possible trajectories given a discrete sample of the process. In key applications the dynamics are non-linear, such as with 2D Navier-Stokes equations in geophysical sciences or reaction-diffusion equations in biochemistry. While Bayesian posterior distributions are widely computed by filtering or MCMC methods, little is known about the statistical behaviour of these posterior measures in non-linear settings. In this talk we will introduce a theoretical framework for such models and then present recent results, known as `Bernstein-von Mises theorems’, that show that the posterior measures are approximated in function space by the Gaussian laws of solutions to certain SPDEs that involve the inverse Fisher information of the underlying statistical model. | | | |
11.11.2026
4:30 PM - 6:00 PM | Machine learning galore!
Lab presentations
Lena Maier-Hein
Claudia Strauch
Alexander Zipf
Rocket science
Leon Mayer (Maier-Hein Lab)
How Naturally Adversarial Medical Images Challenge Vision-Language Models
Marta Gentiloni Silveri (Strauch Lab)
How Reliably Can Generative Models Reconstruct Data from Pure Noise?
Gefei Kong (Zipf Lab)
From bird's eyes to human eyes: mapping street-view greenness by DL model from satellite imagery and geo foundation embeddings
To help plan the catering, please register for free by November 9th! | Scientific Machine Learning
Organizer: Barbara Quintel | INF 205, Mathematikon, 5th floor
| |
17.11.2026
11:15 AM - 12:15 PM | ELLIS AI in Life Seminar Series Heidelberg:
Foundation Models for the Data we Actually Have in Medicine: TabPFN and Beyond
Frank Hutter
Most data in cancer research is tabular: patient records, biomarker panels, clinical trial results, and features derived from omics. For decades, gradient-boosted trees have dominated this domain, and deep learning has struggled, especially in the small-sample regime typical of clinical studies. TabPFN is a tabular foundation model that changes this picture. Pre-trained on millions of synthetic datasets, it performs in-context learning: given a new dataset, it yields well-calibrated predictions in seconds, without hyperparameter tuning, and natively handles missing values, outliers, and mixed data types. In this talk, I will explain the ideas behind TabPFN, our Nature 2025 paper, and the fast-paced developments since then. TabPFN has already been used in hundreds of medical prediction use cases, and we are keen to work with partners to identify & solve the hardest challenges of tabular predictions whose solutions would bring the greatest benefit to society.
| Oliver Stegle
Moritz Gerstung | Kommunikationszentrum (KoZ) des DKFZ
and via Stream
Zoom-Link:
https://dkfz-de.zoom.us/meeting/register/mvbrjkzxSfyE8wurevnCAg#/registration | |
| HITS Symposium
“Future of Sustainable Computing”
Due to the advent of AI and the steadily growing number of data centers, computation accounts for a continuously increasing fraction of the global electricity, water, and natural resource consumption.
Join us for the “Future of Sustainable Computing Symposium” at the Heidelberg Institute for Theoretical Studies (HITS) to explore and discuss the future of sustainable computing research ranging from low-level hardware measures over sustainable AI to policy related approaches. The symposium features a panel of early career as well as established researchers in the broader area of sustainable computing from around the world
Registration for the event is now open. | | Studio Villa Bosch Heidelberg,
Schloss-Wolfsbrunnenweg 33, 69118 Heidelberg | |
09.12.2026
11:00 AM - 12:00 PM | ELLIS AI in Life Seminar Series Heidelberg:
Nationwide disease prediction from electronic health records and genetics
Andrea Ganna
The talk will present recent advances in disease prediction through the integration of nationwide health data and genomic information. It will describe a foundation model trained on Finnish nationwide health data, covering more than 7 million individuals and 3 billion health events, to model and predict health trajectories. The talk will examine the current strengths and limitations of this approach, including challenges related to fairness and generalizability. It will then show how polygenic scores complement electronic-health-record-based prediction, with the two approaches providing distinct strengths across disease domains. Finally, the talk will discuss how integrating genomic and EHR data can improve trial emulation, strengthen causal inference, and improve biomarker discovery. | | Kommunikationszentrum (KoZ) des DKFZ and via Stream Zoom-Link: https://dkfz-de.zoom.us/meeting/register/e6USSBqVQxO4ZH1ZmZe1WA#/registration | |
