AI Fraunhofer HHI
@aifraunhoferhhi.bsky.social
Pushing the boundaries of AI in Explainabability, Efficiency and Application - leading the industry with state-of-the-art methods like LRP, CRP, SpRAy and PCX.
➡ https://www.hhi.fraunhofer.de/en/departments/ai.html
Branche
➡ https://www.hhi.fraunhofer.de/en/departments/ai.html
Branche
Reposted by AI Fraunhofer HHI
Now accepted at #NeurIPS2025 :)
ICL allows LLMs to adapt to new tasks and at the same time enables them to access external knowledge through RAG. How does the latter work?
TL;DR we find that certain attention heads perform various, distinct operations on the input prompt for QA!
arxiv.org/abs/2505.15807
1/
TL;DR we find that certain attention heads perform various, distinct operations on the input prompt for QA!
arxiv.org/abs/2505.15807
1/
The Atlas of In-Context Learning: How Attention Heads Shape In-Context Retrieval Augmentation
Large language models are able to exploit in-context learning to access external knowledge beyond their training data through retrieval-augmentation. While promising, its inner workings remain unclear...
arxiv.org
September 24, 2025 at 12:20 PM
Now accepted at #NeurIPS2025 :)
🔬 Happy Research Appreciation Day!
On July 5, we celebrate the people behind the progress — our researchers. 🌍
Thanks to slapuschkin.bsky.social, Aakriti Jain and Minh An Pham from our team at aifraunhoferhhi.bsky.social for sharing what they value most about research in AI and XAI.
On July 5, we celebrate the people behind the progress — our researchers. 🌍
Thanks to slapuschkin.bsky.social, Aakriti Jain and Minh An Pham from our team at aifraunhoferhhi.bsky.social for sharing what they value most about research in AI and XAI.
July 7, 2025 at 9:27 AM
🔬 Happy Research Appreciation Day!
On July 5, we celebrate the people behind the progress — our researchers. 🌍
Thanks to slapuschkin.bsky.social, Aakriti Jain and Minh An Pham from our team at aifraunhoferhhi.bsky.social for sharing what they value most about research in AI and XAI.
On July 5, we celebrate the people behind the progress — our researchers. 🌍
Thanks to slapuschkin.bsky.social, Aakriti Jain and Minh An Pham from our team at aifraunhoferhhi.bsky.social for sharing what they value most about research in AI and XAI.
Reposted by AI Fraunhofer HHI
Have had enough of the fake "sources" "cited" by ChatGPT? We have the solution in the form of low-cost causal citations for LLMs.
Go check this out! arxiv.org/abs/2505.15807
Thanks to my amazing co-authors
@pkhdipraja.bsky.social,
@reduanachtibat.bsky.social , Thomas Wiegand and Wojciech Samek!
Go check this out! arxiv.org/abs/2505.15807
Thanks to my amazing co-authors
@pkhdipraja.bsky.social,
@reduanachtibat.bsky.social , Thomas Wiegand and Wojciech Samek!
May 28, 2025 at 2:50 PM
Have had enough of the fake "sources" "cited" by ChatGPT? We have the solution in the form of low-cost causal citations for LLMs.
Go check this out! arxiv.org/abs/2505.15807
Thanks to my amazing co-authors
@pkhdipraja.bsky.social,
@reduanachtibat.bsky.social , Thomas Wiegand and Wojciech Samek!
Go check this out! arxiv.org/abs/2505.15807
Thanks to my amazing co-authors
@pkhdipraja.bsky.social,
@reduanachtibat.bsky.social , Thomas Wiegand and Wojciech Samek!
Reposted by AI Fraunhofer HHI
Join Europe’s largest gathering of geoscientists with experts from the Global Initiative on Resilience to Natural Hazards through AI Solutions, chaired by Monique Kuglitsch! 🌍
Look forward to exciting sessions.
📅 Vienna | 27.04. – 02.05. | online or in person
Full program: egu25.eu
Look forward to exciting sessions.
📅 Vienna | 27.04. – 02.05. | online or in person
Full program: egu25.eu
April 28, 2025 at 7:01 AM
Join Europe’s largest gathering of geoscientists with experts from the Global Initiative on Resilience to Natural Hazards through AI Solutions, chaired by Monique Kuglitsch! 🌍
Look forward to exciting sessions.
📅 Vienna | 27.04. – 02.05. | online or in person
Full program: egu25.eu
Look forward to exciting sessions.
📅 Vienna | 27.04. – 02.05. | online or in person
Full program: egu25.eu
Evgenii Kortukov, Sebastian Lapuschkin and Wojciech Samek from our team joined in exploring how to make certain LLM inputs non-executable, in order to defend LLMs from prompt injections. The results are both promising and conceptually interesting. Find Evgenii this week at ICLR to discuss!
I’ll present our 𝗔𝗦𝗜𝗗𝗘 paper as an 𝗢𝗿𝗮𝗹 at the #ICLR2025 BuildTrust workshop! 🚀
✅ ASIDE = architecturally separating instructions and data in LLMs from layer 0
🔍 +12–44 pp↑ separation, no utility loss
📉 lowers prompt‑injection ASR (without safety tuning!)
🚀 Talk: Hall 4 #6, 28 Apr, 4:45
✅ ASIDE = architecturally separating instructions and data in LLMs from layer 0
🔍 +12–44 pp↑ separation, no utility loss
📉 lowers prompt‑injection ASR (without safety tuning!)
🚀 Talk: Hall 4 #6, 28 Apr, 4:45
April 29, 2025 at 11:51 AM
Evgenii Kortukov, Sebastian Lapuschkin and Wojciech Samek from our team joined in exploring how to make certain LLM inputs non-executable, in order to defend LLMs from prompt injections. The results are both promising and conceptually interesting. Find Evgenii this week at ICLR to discuss!
🚀 New demo! Explore CLIP’s hidden concepts with SemanticLens.
With 16 SAE from ViT Prisma (github.com/soniajoseph/...)
💻: semanticlens.hhi-research-insights.eu
📄: arxiv.org/pdf/2501.05398
@fraunhoferhhi.bsky.social @tuberlin.bsky.social @bifold.berlin
#ViTPrisma #CLIP #SAE #XAI #Interpretability
With 16 SAE from ViT Prisma (github.com/soniajoseph/...)
💻: semanticlens.hhi-research-insights.eu
📄: arxiv.org/pdf/2501.05398
@fraunhoferhhi.bsky.social @tuberlin.bsky.social @bifold.berlin
#ViTPrisma #CLIP #SAE #XAI #Interpretability
April 9, 2025 at 12:04 PM
🚀 New demo! Explore CLIP’s hidden concepts with SemanticLens.
With 16 SAE from ViT Prisma (github.com/soniajoseph/...)
💻: semanticlens.hhi-research-insights.eu
📄: arxiv.org/pdf/2501.05398
@fraunhoferhhi.bsky.social @tuberlin.bsky.social @bifold.berlin
#ViTPrisma #CLIP #SAE #XAI #Interpretability
With 16 SAE from ViT Prisma (github.com/soniajoseph/...)
💻: semanticlens.hhi-research-insights.eu
📄: arxiv.org/pdf/2501.05398
@fraunhoferhhi.bsky.social @tuberlin.bsky.social @bifold.berlin
#ViTPrisma #CLIP #SAE #XAI #Interpretability
🎉 On #GirlsDay 2025, Fraunhofer HHI welcomed schoolgirls to explore STEM careers through hands-on workshops.
🤖 In the AI session, led by Dr. Katharina Weitz they trained models, tackled bias & tested explainable AI tools.
💡 A day full of learning, tech & inspiration!
🤖 In the AI session, led by Dr. Katharina Weitz they trained models, tackled bias & tested explainable AI tools.
💡 A day full of learning, tech & inspiration!
April 8, 2025 at 12:41 PM
🎉 On #GirlsDay 2025, Fraunhofer HHI welcomed schoolgirls to explore STEM careers through hands-on workshops.
🤖 In the AI session, led by Dr. Katharina Weitz they trained models, tackled bias & tested explainable AI tools.
💡 A day full of learning, tech & inspiration!
🤖 In the AI session, led by Dr. Katharina Weitz they trained models, tackled bias & tested explainable AI tools.
💡 A day full of learning, tech & inspiration!
Reposted by AI Fraunhofer HHI
Comment: Explainability can foster trust in the use of AI in geoscience
@jesper.drams.ch @ecmwf.bsky.social @aifraunhoferhhi.bsky.social
www.nature.com/articles/s41...
@jesper.drams.ch @ecmwf.bsky.social @aifraunhoferhhi.bsky.social
www.nature.com/articles/s41...
Explainability can foster trust in artificial intelligence in geoscience - Nature Geoscience
Uptake of explainable artificial intelligence (XAI) methods in geoscience is currently limited. We argue that such methods that reveal the decision processes of AI models can foster trust in their res...
www.nature.com
February 5, 2025 at 4:04 PM
Comment: Explainability can foster trust in the use of AI in geoscience
@jesper.drams.ch @ecmwf.bsky.social @aifraunhoferhhi.bsky.social
www.nature.com/articles/s41...
@jesper.drams.ch @ecmwf.bsky.social @aifraunhoferhhi.bsky.social
www.nature.com/articles/s41...
Reposted by AI Fraunhofer HHI
Chinese New Year brought a delegation from Singapore to Forum Digital Technologies.
It was an honor to demonstrate AI research projects from Fraunhofer HHI, FDT, & BIFOLD to Deputy Prime Minister Heng Swee Keat & Ambassador Chong Hock Lee.
@aifraunhoferhhi.bsky.social @fdtberlin.bsky.social
It was an honor to demonstrate AI research projects from Fraunhofer HHI, FDT, & BIFOLD to Deputy Prime Minister Heng Swee Keat & Ambassador Chong Hock Lee.
@aifraunhoferhhi.bsky.social @fdtberlin.bsky.social
January 29, 2025 at 3:57 PM
Chinese New Year brought a delegation from Singapore to Forum Digital Technologies.
It was an honor to demonstrate AI research projects from Fraunhofer HHI, FDT, & BIFOLD to Deputy Prime Minister Heng Swee Keat & Ambassador Chong Hock Lee.
@aifraunhoferhhi.bsky.social @fdtberlin.bsky.social
It was an honor to demonstrate AI research projects from Fraunhofer HHI, FDT, & BIFOLD to Deputy Prime Minister Heng Swee Keat & Ambassador Chong Hock Lee.
@aifraunhoferhhi.bsky.social @fdtberlin.bsky.social
✨ Happy New Year from the Fraunhofer AI Group! ✨
Thank you for making this year unforgettable with your inspiring work and collaboration. We look forward to more achievements ahead.
Wishing you a year full of new opportunities and happiness! 🥂
#FraunhoferHHI #FraunhoferAI
Thank you for making this year unforgettable with your inspiring work and collaboration. We look forward to more achievements ahead.
Wishing you a year full of new opportunities and happiness! 🥂
#FraunhoferHHI #FraunhoferAI
January 2, 2025 at 11:35 AM
✨ Happy New Year from the Fraunhofer AI Group! ✨
Thank you for making this year unforgettable with your inspiring work and collaboration. We look forward to more achievements ahead.
Wishing you a year full of new opportunities and happiness! 🥂
#FraunhoferHHI #FraunhoferAI
Thank you for making this year unforgettable with your inspiring work and collaboration. We look forward to more achievements ahead.
Wishing you a year full of new opportunities and happiness! 🥂
#FraunhoferHHI #FraunhoferAI
🎄 Season’s Greetings from the Fraunhofer AI Group! 🎄
Thank you for an amazing year of projects and collaboration. We truly appreciate your support.
Wishing you a joyful and relaxing holiday season! 🌟
#AI #Research #Innovation #Fraunhofer #HHI #MerryChristmas
Thank you for an amazing year of projects and collaboration. We truly appreciate your support.
Wishing you a joyful and relaxing holiday season! 🌟
#AI #Research #Innovation #Fraunhofer #HHI #MerryChristmas
December 23, 2024 at 10:16 AM
🎄 Season’s Greetings from the Fraunhofer AI Group! 🎄
Thank you for an amazing year of projects and collaboration. We truly appreciate your support.
Wishing you a joyful and relaxing holiday season! 🌟
#AI #Research #Innovation #Fraunhofer #HHI #MerryChristmas
Thank you for an amazing year of projects and collaboration. We truly appreciate your support.
Wishing you a joyful and relaxing holiday season! 🌟
#AI #Research #Innovation #Fraunhofer #HHI #MerryChristmas
𝙃𝙚𝙡𝙡𝙤 𝙒𝙤𝙧𝙡𝙙! 🚀
We, the AI department of the @fraunhofer.bsky.social Heinrich-Hertz Institute are excited to launch our Bluesky channel! We’ll be sharing news, research, and events about Explainability, Efficiency, and Applications of AI. Follow us to stay updated and join the AI Community! 🚀
#AI #ML
We, the AI department of the @fraunhofer.bsky.social Heinrich-Hertz Institute are excited to launch our Bluesky channel! We’ll be sharing news, research, and events about Explainability, Efficiency, and Applications of AI. Follow us to stay updated and join the AI Community! 🚀
#AI #ML
December 10, 2024 at 3:58 PM
𝙃𝙚𝙡𝙡𝙤 𝙒𝙤𝙧𝙡𝙙! 🚀
We, the AI department of the @fraunhofer.bsky.social Heinrich-Hertz Institute are excited to launch our Bluesky channel! We’ll be sharing news, research, and events about Explainability, Efficiency, and Applications of AI. Follow us to stay updated and join the AI Community! 🚀
#AI #ML
We, the AI department of the @fraunhofer.bsky.social Heinrich-Hertz Institute are excited to launch our Bluesky channel! We’ll be sharing news, research, and events about Explainability, Efficiency, and Applications of AI. Follow us to stay updated and join the AI Community! 🚀
#AI #ML