BIFOLD Berlin Institute for the Foundations of Learning and Data

Dr. Maximilian Bandle, founding engineer at @cedardb.com , talks about the real-world experience of transforming Umbra, a TU Munich research project, into CedarDB.

Join us at the BIFOLD Colloquium: t1p.de/zrt13
March 19, 2026 - 4 pm

@tuberlin.bsky.social #datascience #database #ops #data

Tasks:
→ Scalable agentic methods for generating data preprocessing workflows
→ Technologies for cleaning massive dataset repositories
→ Understanding and improving data preparation & extraction pipelines

#RAG #PhDCandidate #CS #CSResearch #PhDOpportunity #PhDLife #PhDChat #DataCleaning #phdsky

BIFOLD is hiring: #Phd Candidate. Focus: Data Science & AI.

Details
💼 D2IP Group, led by Ziawasch Abedjan
🗓 Start date: May 6, 2026
📅 Apply by: March 27, 2026
🔗https://t1p.de/dfhtm

@tuberlin.bsky.social #jobvacancy
#jobalert #ScienceJobs #DataIntegration #DataPreprocessing #jobvacancy

Podcast: Grünflächen - unterschiedliche Sprachen - Magnesium Park(er)leben - Warum Grünflächen für Städte so wichtig sind; Was meine Stimme über mich verrät; Was tun gegen den Gender Pay Gap?; Digitale Souveränität - Für Europa realistisch? Große Fragen in 10 M...

🎙️ WDR 5 Quarks features a segment on AI sovereignty and the #DigitalIndependenceDay movement. BIFOLD researcher Sebastian Schelter (DEEM Lab) shares his perspective on US dominance in the booming AI landscape. (Start: 57:33 min)

t1p.de/8ek6e

@tuberlin.bsky.social @quarkswdr.bsky.social #DIDit

🔔 Application deadline approaching: March 13, 2026! The DIMA Research Group (BIFOLD/ @tuberlin.bsky.social ) is hiring a Doctoral Researcher.

Job Info: t1p.de/w6k50

#jobvacancy
#jobalert
#ScienceJobs
#AdminJobs
#phdsky
#fundedphd
#sciencejobs
#STEMjobs
#SciJobs
#AcademicJobs
#FacultyPosition

#Student Assistant at BIFOLD / @tuberlin.bsky.social. Join the D2IP group, led by Ziawasch Abedjan. Research focus: data lakes & ML pipelines!

🔹 40–80 hrs/month
🔹 Python, pandas, PyTorch
🔹 Until April 2027
📅 Apply by March 16, 2026
👉 t1p.de/t93i6

#jobvacancy
#jobalert
#ScienceJobs
#STEMjobs

Paper: Rank-based Geographical Regularization: Revisiting Contrastive Self-Supervised Learning for Multispectral Remote Sensing Imagery. Tom Burgert, Leonard Hackel, @paolorota.bsky.social , Begüm Demir.

Rank-based Geographical Regularization: Revisiting Contrastive Self-Supervised Learning for Multispectral Remote Sensing Imagery Self-supervised learning (SSL) has become a powerful paradigm for learning from large, unlabeled datasets, particularly in computer vision (CV). However, applying SSL to multispectral remote sensing (...

GeoRank is a novel regularization method for contrastive self-supervised learning ( #SSL ) applied to #multispectral remote sensing imagery.

@arxiv.bsky.social : arxiv.org/abs/2601.02289 #preprint #eo

BIFOLD at #WACV2026 will present #GeoRank.

Meet the researcher: Mar 9, 26; 1:30 pm; Oral Session 5B: Remote Sensing and Sensors.

More: t1p.de/8lcqx

#earthobservation #remotesensing @tuberlin.bsky.social #GIS #geospatial #computervision #gischat #thescienceofwhere #dataviz #openaccess #MLsky

Charité/BIFOLD: W3 Prof in #DataEngineering in Health. Based at: Institute of Medical Informatics. Focus: advancing foundational and applied research in #clinical #dataManagement, #multimodal #machineLearning ...

@prof2prof.bsky.social #FacultyPosition #VacancyEdu #ScienceCareer #ScholarshipAlert

Charité - Universitätsmedizin Berlin & BIFOLD are #hiring. #W3 #Professorship in Data Engineering in Health, permanent tenure.

career.charite.de/calls/aussch...

#AcademicSky #chariteBerlin #STEMjobs #academicjobs @tenuretracker.bsky.social #Medinformatics #ClinicalResearch @tuberlin.bsky.social

1955 tagte der Deutsche Bundestag an der TU Berlin.

Der Physikhörsaal heite.

🏛️ Am 19. Oktober 1955 tagte der Deutsche Bundestag erstmals in Berlin – in einem Hörsaal der #TUBerlin. Die Sitzungswoche war ein Ausdruck der Verbundenheit mit dem geteilten Berlin & der Idee der Wiedervereinigung. Wo sonst studiert wurde, ging es um Wirtschaft, Freiheit & die Zukunft Deutschlands.

🔦 Past Paper Highlight: PRCA & DRSA, methods that disentangle neural network explanations into concept-level components.

📄 Disentangled Explanations of Neural Network Predictions by Finding Relevant Subspaces
🔗 doi.org/10.1109/TPAM...
💻 github.com/p16i/disenta...

Jobs Current open positions appear here. Junior / Senior AI Research Consultant (2 Positions) – AI.UP The AI Competence Center (AI.UP) at the University of Potsdam invites applications for two AI Resear…

Hiring! Postdoc in Causal Inference & Explainable AI at University of Potsdam (CRC 1294 “Data Assimilation”).
TV-L E13 (100%), until 06/2029. Start summer 2026.
Deadline: March 22, 2026.
Apply (ID 308/2026): lisa.goesel@uni-potsdam.de
Details: causalinferencelab.com/jobs

A huge thank you to #LeonardHackel for such a thought-provoking and forward-looking presentation! 🙌

#FoundationModels #AI #EarthObservation #Geospatial #MixtureOfExperts

💡 The big takeaway: Remote sensing foundation models can be computationally efficient without losing representational capacity. Efficiency and performance are no longer a trade-off; they're achievable together.

🔧 Practical deployment guidelines: Match model size to your compute budget and don't overlook post-hoc slimming as a powerful option for optimizing already-pretrained models.

📐 Flexible scaling without the accuracy cliff: Post-hoc and learned slimmability enable flexible width scaling with only minor accuracy loss making deployment adaptable to real-world resource constraints.

🌍 Smarter training data strategy: A novel stratified sampling approach jointly enforces thematic, climatic, and spatial diversity, proving that what you train on is just as critical as how you train.

⚡ 2x+ efficiency, zero compromise: By integrating Soft Mixture-of-Experts, CSMoE delivers superior or comparable performance to state-of-the-art models while requiring significantly fewer FLOPs across scene #classification, #semantic #segmentation, and #image #retrieval.

🏆 A world first: #CSMoE is the first compute-efficient multi-modal foundation model in remote sensing.

Great talk by Leonard Hackel in our lunch talk series on the future of remote sensing foundation models today.

Here are the key takeaways 👇

#RemoteSensing @tuberlin.bsky.social 🛰️🌍 #eochat #geoai #GIS #space4good #opensource #classification #CSMoE #DeepLearning #MachineLearning #BIFOLD

📋 PhD | Information Integration and Data Preparation (D2IP) (IV-81/26)
TV-L E13 | 5 years
⏳ Deadline: March 27, 2026
www.jobs.tu-berlin.de/en/job-posti...

📋 PhD | Data Management & Scalable Systems (IV-76/26)
Distributed data streams, analytics infrastructure.
TV-L E13 | 5 years
⏳ Deadline: March 14, 2026
www.jobs.tu-berlin.de/en/job-posti...

BIFOLD is hiring — 2 open positions in information integration and data management 🧵

#AcademicJobs #VacancyEdu #doctoral #ScienceCareer #AcademicJobs #Researcher #PhDlife #JobSearch #ScholarshipAlert #fundedphd #AcademicSky #STEMJobs #FacultyPosition @tuberlin.bsky.social

Thomas shared his perspective on the AI coding revolution, drawing on his experience leading GitHub through significant changes in software development.

He also discussed, if the increasing use of AI agents by everybody makes programming skills obsolete.

Thanks a lot to Thomas Dohmke for joining us at the BIFOLD Colloquium. He delivered a keynote on "The New Age of Developer AI."

@thomasdohmke.bsky.social @tuberlin.bsky.social @github.com

#AIResearch #ML #SoftwareDevelopment #AIAgents #TUBerlin #BerlinAI #DeveloperTools

📷Michael Setzpfandt

Researcher Spotlight: Dr. Lukas Muttenthaler What if AI could learn to see the world the way we do? BIFOLD PhD graduate Dr. Lukas Muttenthaler is pushing AI beyond raw performance, exploring representational alignment, where cognitive science meets computer vision to build machines that perceive more like humans.

"My goal is to develop vision foundation models that can handle unfamiliar data, much like how the human brain adapts to unfamiliar situations. This is crucial for real-world applications such as autonomous driving or medical diagnostics..."

t1p.de/pe5z5

#MLSky #Scicomm #AIAlignment #ML A.I.

Researcher Spotlight: Dr. Lukas Muttenthaler

BIFOLD PhD graduate Dr. Lukas Muttenthaler is pushing AI beyond raw performance, exploring #representational #alignment.

More: t1p.de/pe5z5

@lukasmut.bsky.social @eml-munich.bsky.social @tuberlin.bsky.social @mpicbs.bsky.social @xai-berlin.bsky.social

Ever get tired of tiny timesteps bottlenecking your MD simulations?

We show how to train a model for large-timestep Hamiltonian dynamics directly on standard MLFF datasets. 𝗡𝗼 𝗿𝗲𝗳𝗲𝗿𝗲𝗻𝗰𝗲 𝘁𝗿𝗮𝗷𝗲𝗰𝘁𝗼𝗿𝗶𝗲𝘀, 𝗻𝗼 𝘂𝗻𝗿𝗼𝗹𝗹𝗶𝗻𝗴, 𝗻𝗼 𝘁𝗲𝗮𝗰𝗵𝗲𝗿 needed!

🧵👇