SEARCH
ALL NEWS

Attacking privacy leaks in virtual backgrounds
Peeking through the virtual curtain: A new study by the BIFOLD MLSEC group reveals that current virtual backgrounds in video calls can leak enough pixels from the environment to reconstruct objects in the background.

DEEM Lab Contributes to RecSys 2025
At RecSys 2025 in Prague, DEEM Lab will present a spotlight oral paper at the main conference on scalable data debugging using Data Shapley Values, as well as a workshop paper on realistic benchmarks for unlearning in recommender systems.

Dr. Angela Holzer on Open Science and AI
In our interview, DFG’s Dr. Angela Holzer explains why openness and transparency are key to making AI research trustworthy — and how Open Science principles can guide responsible innovation.

Adrian Michalke Receives Software Campus Grant
BIFOLD’s / TU Berlin doctoral researcher Adrian Michalke received a prestigious Software Campus Grant to develop “REPLAY,” a system for the realistic replay of data extraction processes. In collaboration with Celonis, the project aims to enhance data reliability and debugging.

Rethinking AI for the Humanities
Matteo Valleriani and Oliver Eberle are organizing the BIFOLD workshop, "AI-based Methods for the Humanities." In an interview, they discuss the workshop's key aims. They ask what major debate defines AI in the humanities and how the workshop helps clarify a complex academic mission.

Berlin AI Square: Driving Research and Collaboration
Berlin’s Senator for Economic Affairs, Franziska Giffey, visited BIFOLD to learn about key research, spin-offs, and industry partnerships. The visit also highlighted the new developing Berlin AI Square as a leading AI research, innovation, and business hub.

Simulations of large biomolecules with quantum accuracy
An international team of researchers from BIFOLD, the University of Luxembourg, and Google DeepMind has developed a new machine learning foundation model capable of simulating a wide variety of molecular systems – for example, large and complex biological molecules – with quantum-mechanical accuracy.

VLDB 2025 Conference Contributions
Researchers from five BIFOLD research groups will participate in the 51st International Conference on Very Large Data Bases (VLDB 2025), presenting a range of papers, tutorials, demos, and workshops. The conference will run from September 1 to 5, 2025, in London, United Kingdom.

Researcher Spotlight: Dr. Mahdi Esmailoghli
Dr. Mahdi Esmailoghli earned his PhD at BIFOLD/TU Berlin in 2024 with a thesis on efficient data discovery in large-scale data lakes. Now a Postdoctoral researcher at Humboldt-Universität zu Berlin, his work expands from data discovery to intelligent analysis pipelines, aiming to make Big Data more accessible and effective for machine learning and scientific research.

USENIX 2025 Conference Contributions
BIFOLD researchers from the MLSec group will present two papers at the 34th USENIX Security Symposium (Aug 13–15, 2025, Seattle). One paper shows that virtual backgrounds in video calls can unintentionally reveal parts of a user’s real surroundings, exposing them to privacy risks.