HIDA Lecture: Making ML Research Count
Speaker: Peter Steinbach, Helmholtz-Zentrum Dresden-Rossendorf
Date: 25.11.2026, 02:00 pm
Title: Making ML Research Count - Avoiding the Pitfalls of Spurious Findings
Abstract
The explosion of ML submissions at major conferences and the rapid adoption of LLMs have accelerated the pace of discovery, but have they increased the reliability of our findings? In the rush to innovate, it is easy to mistake a lucky seed or a leaked test set for a scientific breakthrough.
This lecture addresses the "trust gap" in current ML research. We will critically analyze how common ML practices can inadvertently lead to fragile results and discuss the essential safeguards needed to ensure that your research contributes lasting value to the scientific community rather than adding to the noise.
Peter Steinbach
Peter Steinbach received his PhD in Particle Physics in 2012 from the TU Dresden for an experimental study of LHC data at the ATLAS experiment to reduce background contributions to Higgs Particle searches employing multivariate statistical methods.
He continued to industry as a HPC support and software engineer helping scientists push the limits of their applications in a service oriented group. In this role, he become increasingly exposed to Deep Learning applications for computer vision applications in biology.
In 2019, he started to lead a group of AI consultants at HZDR that aims to help scientists from the research field matter in the Helmholtz Association to use machine learning in experiment and theory.
