Sign in

BIFOLD Berlin Institute for the Foundations of Learning and Data

@bifold.berlin
1.5K followers 6.3K following 454 posts

Groundbreaking foundational research in Big Data Management, Machine Learning, and their intersection. #AI #Research www.bifold.berlin 📰News: www.bifold.berlin/news-events/news 🔑Data Privacy: www.bifold.berlin/data-privacy

PostsRepliesMedia
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 28/09/2026
Let's stay in touch. Are you joining the BEADS Summit on October 21, 2026? 👉 events.tu-berlin.de/de/events/01... #Data #Systems #Analytics #📈🤖 @tuberlin.bsky.social @databricksinc.bsky.social
events.tu-berlin.de
000
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 28/09/2026
What a Kick-Off for BEADS (Berlin Ecosystem for Analytics and Data Systems): The very first Meetup was already a great success. With a full house and a lively discussion, it offered a glimpse of what this new format brings to Berlin's data community. www.bifold.berlin/news-events/...
bifold.berlin
Photo-Recap: BEADS Kick-Off
BEADS launched its Tech Meetup Series with a kick-off event at BIFOLD, drawing around 60 participants from Berlin's data community for three talks from Snowflake, Databricks, and BIFOLD, followed by a...
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 28/09/2026
We thank her for sharing her work with our community. More BIFOLD Colloquia will follow soon. Photos: BIFOLD/Gereon Dusella
000
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 28/09/2026
, and early work on foundation models and agents for predictive reasoning over tables. Together, these threads point toward smaller, specialized models that make insights from structured data more accessible and trustworthy.
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 28/09/2026
assessing when a dataset is sensitive (a notion she argued goes beyond personally identifiable information and depends on context), grounding text-to-SQL models in real-world database schemas through the new SQaLe dataset
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 28/09/2026
Many thanks to Madelon Hulsebos for her talk at the BIFOLD Colloquium. @madelonhulsebos.bsky.social @tuberlin.bsky.social In "On the Different Flavors of Tabular Reasoning," she showed that reasoning over tabular data spans several distinct challenges:
230
Reposted by BIFOLD Berlin Institute for the Foundations of Learning and Data
Alexander von Humboldt Institut für Internet und Gesellschaft @hiigberlin.bsky.social · 23/09/2026
Rechts vor links lernt jedes Kind. Was mit unseren #Daten im Internet passiert, weiß kaum jemand. Brauchen wir einen Netz-Führerschein? Im #DigitalerSalon diskutieren wir über #Datenkompetenz, #Bildung & selbstbestimmte Entscheidungen. 👉 Am HIIG oder im Livestream: www.hiig.de/events/digit...
124
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 18/09/2026
BIFOLD thanks all participants, speakers, and organizers for four engaging days. See the full photo recap on our website: www.bifold.berlin/news-events/... @weizenbauminstitut.bsky.social @tuberlin.bsky.social #ResponsibleAI #PhysicalAI #BIFOLD #weizenbaum #LIS #AI #Summerschool
bifold.berlin
Photo Recap: Berlin Summer School on AI and Society
Four days focused on the responsible design of AI-driven autonomous systems, robotics, and physical AI. BIFOLD Co-Director Klaus-Robert Müller opened the summer school with his reflections on safety, ...
010
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 18/09/2026
BIFOLD Co-Director Klaus-Robert Müller opened with a call for Safety by Design: safety must be built into AI systems from the outset, not added later. Keynotes by Judith Simon, Marc Toussaint, and Iyad Rahwan, plus a visit to Google AI Center Berlin.
110
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 18/09/2026
Photo Recap: Berlin Summer School on AI and Society. From Sept 14-17, 50 students from 10 countries met in Berlin for four days on the responsible design of AI-driven autonomous systems, robotics, and physical AI. Organized by BIFOLD, the Weizenbaum Institute, and TU Berlin's LIS Lab.
121
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 17/09/2026
The workshop brings together experts from artificial intelligence, geospatial analytics, climate and weather science, remote sensing, and space mission operations o discuss the emerging role of agentic AI in Earth Observation (EO). @tuberlin.bsky.social @esa.int
000
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 17/09/2026
Join us in Berlin at the scientific workshop "Agentic AI for Earth Observation," organized by BIFOLD and the ESA Φ-lab from October 19 to 21, 2026. Begüm Demir shares the workshop's aims and invites researchers, practitioners, and decision-makers. agentic-eo.berlin #gischat #EO #ESA #GeoAI
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 17/09/2026
Madelon Hulsebos, a tenured senior researcher at Centrum Wiskunde & Informatica (CWI) in Amsterdam, will explore how to judge the sensitivity of tabular data beyond personally identifiable information, and introduce SQaLe, a large-scale text-to-SQL dataset grounded in real-world database schemas.
010
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 17/09/2026
Madelon Hulsebos will join the BIFOLD Colloquium with a talk titled "On the Different Flavors of Tabular Reasoning." Join us on-site at BIFOLD. 👉Sep 25, 2026 👉11 AM 👉Franklinstr. 28/29 www.bifold.berlin/news-events/... @tuberlin.bsky.social @cwi-amsterdam.bsky.social @madelonhulsebos.bsky.social
bifold.berlin
11/2026 BIFOLD Colloquium
Madelon Hulsebos, Centrum Wiskunde & Informatica (CWI), talks about the latest work of her team around foundation models and agents for predictive reasoning over tabular data, and a reflectio...
120
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 15/09/2026
"I am particularly excited that the funding will allow us to pilot our developed system in a real-world clinical setting," says Jamshidi Idaji. #BIFOLD #Oncology #AI www.bifold.berlin/news-events/...
bifold.berlin
Beyond the Snapshot: New AI takes a dynamic view of Cancer Diagnosis
A consortium led by Dr. Mina Jamshidi Idaji (BIFOLD/TU Berlin), with LMU Munich and UKE Hamburg, is uniting oncologists, pathologists and ML researchers to build AI that combines tissue images with mo...
000
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 15/09/2026
Led by Dr. Mina Jamshidi Idaji (BIFOLD/TU Berlin), with Prof. Dr. Philipp Jurmeister (LMU Munich) and Prof. Dr. Michael Bockmayr (UKE Hamburg).
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 15/09/2026
In reality, a patient's picture develops continuously as new findings arrive. The consortium is building AI models that integrate images, molecular data, and clinical information and adapt at each stage of the diagnostic journey.
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 15/09/2026
When oncologists assess a cancer patient, they draw on histopathological images (tissue samples), molecular analyses of the tumor, and clinical data. Many AI systems examine these sources separately, or treat them as one fixed snapshot of the patient.
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 15/09/2026
BIFOLD researcher Dr. Mina Jamshidi Idaji to lead a €1.5 million research consortium on AI in Oncology. The project brings together oncologists, pathologists, and ml researchers from BIFOLD/TU Berlin, LMU Munich, and UKE Hamburg. 🧵 @tuberlin.bsky.social @lmu.de @xai-berlin.bsky.social
130
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 14/09/2026
The 2026 Berlin Summer School on Artificial Intelligence and Society opened today, September 14, and runs through September 17, 2026, in a new setting: for the first time, the summer school takes place in BIFOLD's own premises. @tuberlin.bsky.social @weizenbauminstitut.bsky.social
032
Reposted by BIFOLD Berlin Institute for the Foundations of Learning and Data
icde-publicity.bsky.social @icde-publicity.bsky.social · 30/08/2026
The #ICDE2027 Call for Tutorials is OPEN! 📚✨ Share your expertise by proposing a tutorial on cutting-edge data engineering topics, core systems, vector/graph data, or AI/LLMs for data. Shaping the program for ICDE 2027! Details: icde2027.github.io/cf-tutorials... #DataEngineering
icde2027.github.io
Tutorials
012
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 28/08/2026
BIFOLD congratulates Dr. Pattarawat Chormai on the successful defense of his doctoral dissertation. On August 27, 2026, Pattarawat Chormai defended his dissertation, "Understanding and Improving Neural Networks with Relevant Subspace Analysis." @tuberlin.bsky.social
030
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 28/08/2026
Save the date: Sep 25, 2026. #BIFOLD #Colloquium. Prof. Dr. Masashi Sugiyama (The University of Tokyo, RIKEN) will present his group's latest research on advanced reward modeling.
000
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 28/08/2026
His doctoral work spent years testing the opposite assumption: building databases that reveal only as much as they must and stay fast while doing so.
000
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 28/08/2026
Most privacy protection systems slow down and become less useful the more data they have to protect. Does it have to be that way? In our new Researcher Spotlight, BIFOLD introduces Dr.-Ing. Rudi Poepsel Lemaitre, who defended his PhD summa cum laude at BIFOLD. t1p.de/8n1sk @tuberlin.bsky.social
120
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 27/08/2026
Full overview, paper links and author lists: www.bifold.berlin/news-events/... #VLDB2026 #DataManagement #BIFOLD
bifold.berlin
VLDB 2026 Conference Contributions
At VLDB 2026, taking place in Boston, USA, from August 31 to September 4, 2026, BIFOLD researchers will present a total of six research papers, three workshop papers, two demonstrations, a workshop keynote, and a panel contribution.
000
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 27/08/2026
Also at VLDB 2026: Aug 31, QDB Workshop keynote "Ad-hoc Data Integration in Data Lakes" by Ziawasch Abedjan DEI Panel with Ziawasch Abedjan Congratulations to Arnab Phani, recipient of the AIDB Workshop Emerging Reviewer Award
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 27/08/2026
Workshop papers: Aug 31, TPCTC Workshop: SECBench, a benchmarking framework for stream processing in the sensor-edge-cloud continuum Aug 31, NOVAS Workshop: ArtiFact, a multi-modal cultural heritage dataset Plus: Be Fair! on fairness constraints for ML engineering agents
110
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 27/08/2026
Demonstrations, Sep 1 (13:45-15:15) and Sep 2 (15:45-17:15), Graph, Systems and Infrastructure: EarthLake, a model lake system for Earth observation foundation models SemPiper, interactive code synthesis for semantic operators in ML pipelines
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 27/08/2026
Research papers, part 2: Sep 2, 10:45-12:15, R19: stratum, infrastructure for agent-centric ML workloads Sep 2, 10:45-12:15, R19: SemBench, a benchmark for semantic query processing Sep 3, 15:45-17:15, R51: Meerkat, failure recovery for the Internet of Things
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 27/08/2026
Research papers, part 1: Sep 1, 10:45-12:15, R3: Matryoshka (feature discovery in data lakes) Sep 1, 10:45-12:15, R3: FedAugment (table augmentation over decentralized data) Sep 1, 17:15-18:30, Poster Session 1: Distilled Documentation for Dialect-Specific SQL Generation
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 27/08/2026
Meet BIFOLD researchers at VLDB 2026! Aug 31-Sep 4 in Boston, USA: 6 research papers, 3 workshop papers, 2 demos, a workshop keynote, and a panel, plus an Emerging Reviewer Award. Here is when and where to catch us on site (thread):
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 26/08/2026
The #ELLIS Unit Berlin hosted its first #VLDB & #ECCV Fest, bringing together around 50 researchers from Berlin and Potsdam ahead of VLDB 2026 and ECCV 2026. Thank you to everyone who presented and joined! Wishing all presenters great success at both conferences. @tuberlin.bsky.social
000
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 11/08/2026
The study, with TU Berlin, Inria & Ruhr University Bochum, finds current defenses aren't built for this kind of automated threat. Catch the presentation this week at #USENIX Security Symposium 2026 in Baltimore.
000
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 11/08/2026
AI-generated phishing emails are 2.4x more effective than generic ones, and cost only $0.03 to produce. BIFOLD field study (7,741 participants) shows how cheap automated spear phishing has become. Full story on our website 🧵 t1p.de/10n32 @tuberlin.bsky.social @ruhr-uni-bochum.de @rieck.mlsec.org
t1p.de
AI Goes Phishing
BIFOLD study, co-authored by members of the Machine Learning and Security Group, shows AI-generated phishing emails are three times as effective as generic ones and cost next to nothing. The field exp...
151
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 11/08/2026
📄 Paper: arxiv.org/abs/2606.24716 Authors: Jonas Klotz, Cassio F. Dantas, Pallavi Jain, Diego Marcos, Begüm Demir 🌐 Project page: jonasklotz.com/sae-metric/ 📂 Code: github.com/JonasKlotz/s...
arxiv.org
Evaluating the Interpretability of Sparse Autoencoders with Concept Annotations
Sparse autoencoders (SAEs) are increasingly used to extract interpretable concepts from vision and vision language models, yet existing evaluation methods largely rely on proxy metrics or qualitative ...
020
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 11/08/2026
Moderate dictionary sizes turned out to be the most genuinely interpretable. For anyone building or evaluating interpretability tools for vision or foundation models, this offers a concrete way to check whether "interpretable" claims hold up, not just assume they do.
110
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 11/08/2026
The result surprised a common assumption in the field: bigger isn't better. Larger SAE dictionaries mostly inflate the proxy scores researchers have been relying on, without producing latents that are any more aligned with real, human-understandable concepts.
110
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 11/08/2026
which matches SAE latents to human-annotated attributes; TAPAScore, a functional check on whether a concept behaves as it should; and two new benchmarks, synCUB and synCOCO, built specifically for this kind of evaluation.
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 11/08/2026
or trust the results. A group of researchers from BIFOLD/TU Berlin, Inria, and UMR TETIS has built a way to test this properly, and the paper has been selected for an oral presentation (spotlight) at ECCV 2026 in Malmö. Their approach has three parts: FBMP,
110
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 11/08/2026
tools for making vision models interpretable. Right now, "interpretable" is mostly a judgment call: researchers eyeball a handful of activating images and decide whether a latent "seems to" represent a concept. There's no shared way to measure it, which makes it hard to compare methods
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 11/08/2026
A model claims to have found a human-understandable concept inside its neurons. How would you actually check that's true? Beyond squinting at a few example images and deciding it "looks right"? That's the quiet problem behind sparse autoencoders (SAEs), one of the most talked-about
100
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 11/08/2026
Everyone calls sparse autoencoders "interpretable," almost no one measures it. A new benchmark changes that. /// #ECCV2026 #OralPresentation #Paper #BIFOLD @tuberlin.bsky.social #compVision #EO #gischat
120
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 02/07/2026
🛠️ Call for Hands-on Sessions: Lead an engaging session where participants actively work with tools, datasets, models, or workflows through practical exercises. Submit your abstract, demo, or hands-on session proposal now: agentic-eo.berlin
agentic-eo.berlin
Agentic AI for Earth Observation
The workshop for Agentic AI and Earth Observation in Berlin, Germany, 19-21 October 2026
030
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 02/07/2026
💻 Call for Demos: Showcase interactive tools, prototypes, or applications that demonstrate technical capabilities and user interaction.
130
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 02/07/2026
We welcome contributions from academia and industry through three dedicated calls: 📄 Call for Abstracts: Present your latest research and development in the context of AI Agents through oral presentations or posters.
110
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 02/07/2026
⏳➕⏰ Submission deadline extended: 8 July 2026! Contribute to "The Agentic AI for EO Workshop" 📍Berlin, 🗓️ 19–21 October 2026 Submit your abstract, demo, or hands-on session proposal now: agentic-eo.berlin #earthobservation #eochat #remotesensing #cfp #cfd #gischat #mlsky 🌍 🛰️ @tuberlin.bsky.social
agentic-eo.berlin
Agentic AI for Earth Observation
The workshop for Agentic AI and Earth Observation in Berlin, Germany, 19-21 October 2026
121
Reposted by BIFOLD Berlin Institute for the Foundations of Learning and Data
Einstein Foundation Berlin @einsteinberlin.bsky.social · 02/07/2026
PhD in Berlin? The city offers a broad spectrum of doctoral programs w/ funding opportunities & collaborations w/ research institutions. www.doctoral-programs.de/welcome/ @freieuniversitaet.bsky.social, @humboldtuni.bsky.social, @tuberlin.bsky.social, @unipotsdam.bsky.social, @udkberlin.bsky.social
182
Reposted by BIFOLD Berlin Institute for the Foundations of Learning and Data
Healthcare in Europe (DE) @health-europe-de.bsky.social · 02/07/2026
💡🐴 Eine neue Studie von @bifold.berlin an der @tuberlin.bsky.social, mit dem KI-Unternehmen Aignostics, @lmu.de und @nkinl.bsky.social zeigt, wie der „Kluge Hans“-Effekt #KI Modelle in der #Pathologie dazu bringen kann, #Krebs zu diagnostizieren, wo keiner ist – und wie sich das verhindern lässt.
healthcare-in-europe.com
KI als „Kluger Hans“: Maßstab für zuverlässigere Pathologie-Diagnostik
Eine neue Studie von BIFOLD Forschenden an der TU Berlin gemeinsam mit dem Berliner KI-Unternehmen Aignostics, der Ludwig-Maximilians-Universität (LMU) München und dem Netherlands Cancer Institute (NK...
021
BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · 30/06/2026
BIFOLD congratulates Dr. Tom Burgert! 🎓🎉 Tom has successfully defended his PhD thesis with distinction - Summa Cum Laude - at @tuberlin.bsky.social Tom is part of the RSiM research group, led by Begüm Demir, at BIFOLD.
020