What does an LLM think of you?
LLMs pick up the slightest implicit cues about users. Often, they refuse to say who they think the user is.
In our new preprint, we show how to read and change these hidden beliefs. Turns out they even affect whether the model refuses
New paper🧵
Zeynep Akata
Zeynep Akata is a Liesel Beckmann Distinguished professor of computer science at the Technical University of Munich where she leads the Interpretable and Reliable Machine Learning chair. Akata is also the director of the Helmholtz Institute for Explainable Machine Learning. More on Wikipedia
-
Zeynep Akata reposted
Kirill Bykov
@kirill_bykov
-
Zeynep Akata reposted
Luis
@lusxvr
Agents are evaluated with ever more detailed metrics, but why don't we study how their own configuration shapes performance?
We study this on 4 scientific tasks, by varying 5 parts of the agents configuration across 18,000+ runs.
Here are my favorite findings: -
Zeynep Akata reposted
Lukas Thede
@lukas_thede
📣 New preprint: Replay on Demand (RoD)
Imagine you had 10 hours to refresh everything you learned in school.
Would you divide them equally across every subject?
Probably not. You’d spend your limited time on what you’ve forgotten most.
🧵 arxiv.org/abs/2609.40089 -
Zeynep Akata reposted
Lukas Thede
@lukas_thede
📣 Introducing MedKIT, our new benchmark for knowledge integration in LLMs, accepted at #neurips2026!
🧵 Why recalling updated facts ≠ integrated knowledge and why we need methods that can bridge this gap:
📄 arxiv.org/abs/2609.38543 -
Zeynep Akata reposted
Lukas Thede
@lukas_thede
🎉 CapTrack has been accepted to #neurips2026!
CapTrack extends LLM forgetting evaluation beyond general knowledge to capabilities.
We find forgetting is highly heterogeneous across capabilities.
⚠️ If you touched the weights, check what changed.
📄 arxiv.org/abs/2603.06610 -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
We will have two poster presentations and one workshop poster at #eccv2026 in Malmö🇸🇪!
Please come and say hi if you are interested in topics related to multi-modal learning. Details in the thread👇 -
Zeynep Akata reposted
Rui Xiao
@RuiXiao99
Happy to have presented our work "FINER: MLLMs Hallucinate under Fine-grained Negative Queries" at #cvpr2026 as an oral (award candidate) and at #icvss2026
[Paper]: arxiv.org/abs/2603.17662
[Code]: github.com/ExplainableML/finer -
Zeynep Akata reposted
Helmholtz Institute for Human-Centered AI
@cpilab
Deadline to apply is August 14th. If you are looking for a postdoc on foundation models, open-ended discovery, or general ml-approaches to cognitive science, we would like to work with you.
Helmholtz Institute for Human-Centered AI
@cpilab
We are looking for a postdoctoral researcher who wants to innovate at the intersection of cognitive science, neuroscience, and machine learning.
📍Munich
📅Flexible start: 10/2026–03/2027
⏳Full consideration by 14/08/2026
💼Funded for 2 years, extension possible
Please share. -
Zeynep Akata
@zeynepakata
We are looking for PhD students to start as soon as possible. Application deadline is August 10. Please apply, please share! 🙂
Explainable Machine Learning
@ExplainableML
📢 Fully funded PhD @ Helmholtz Munich + TUM
We're hiring PhD students to build trustworthy AI systems for science: explainable AI, VLMs, reliable adaptation, agentic workflows.
⏰ Priority deadline 10 Aug 2026. Application details below 👇 -
Zeynep Akata reposted
Helmholtz Institute for Human-Centered AI
@cpilab
We are looking for a postdoctoral researcher who wants to innovate at the intersection of cognitive science, neuroscience, and machine learning.
📍Munich
📅Flexible start: 10/2026–03/2027
⏳Full consideration by 14/08/2026
💼Funded for 2 years, extension possible
Please share. -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Our paper won the best paper award at the CTB workshop (sites.google.com/view/icml-ctb/home) in #icml2026📣. Please check the thread for details and huge congrats to our authors:
Sena Korkut, @MariaAleBravo, @Sanghwa30296060,
@zeynepakata
Explainable Machine Learning
@ExplainableML
4/
From Accuracy to Visual Dependence: Auditing and Filtering Modality Collapse in Traffic VideoQA
Sena Korkut, Maria Alejandra Bravo, @Sanghwa30296060, @zeynepakata
[Paper]: openreview.net/pdf?id=rKWze9irRW
[Workshop]: CTB
🗓 July 10, 11:00–12:00 & 16:00–17:00 — Hall A -
Zeynep Akata reposted
Yiran Huang
@YiranHuang9
Excited to share our ICML 2026 Spotlight paper!
MLLMs can learn new visual concepts from a few in-context examples, but how? We build a controlled microscope, uncover the mechanism, and validate its signatures in Qwen2.5-VL.
🗓️Thu, Jul 9, 2026, 10.30 a.m. to 12.15PM
📍Hall A #2803
☝️Come and check out our poster! @zeynepakata,@QBouniot, and I will be there and would love to chat!
#icml2026 #mllm #interpretability #incontextlearning -
Zeynep Akata reposted
Simon Roschmann
@SimonRoschmann
Excited to share that our paper "SOTAlign: Semi-Supervised Alignment of Unimodal Vision and Language Models via Optimal Transport" was accepted at ICML.
📄 Paper: arxiv.org/abs/2602.23353
💻 Code: github.com/ExplainableML/SOTAlign -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
We also have a great lineup of workshop presentations at #icml2026! Most of our authors will be there too—come chat with them if you're interested.
👇 Details in the thread: -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
#icml2026 kicks off in three days! We're excited to be presenting eight papers at the main conference this year. See the thread below for each paper, plus presentation times and locations 👇 Talk to our authors if you are interested!
-
Zeynep Akata
@zeynepakata
This one feels special ❤️ Our group has 8 papers at ICML main track this year, and I’ll be attending ICML in Seoul — my first ICML after 10 years, a pandemic and two children. So proud of the team and excited to be back!
Explainable Machine Learning
@ExplainableML
Happy to share that we have 8 papers accepted to #icml2026! 🎉
Most of the authors will be traveling to Seoul, Korea 🇰🇷 — feel free to reach out and chat with them if you’re interested in the work.
See the full thread below for our papers 👇 -
Zeynep Akata reposted
Karsten Roth
@confusezius
So exciting to help build encoder-free Gemma from scratch! Amazing to see all modalities intertwined in one single LLM decoder 🧬.
Bonus for my academic heart ❤️👓: it's simple to run and brings some new architectural insights to the OSS community!
🧵1/nThis media couldn't be displayed. See it on the original post.
Google Gemma
@googlegemma
Meet Gemma 4 12B!
A unified, encoder-free multimodal model designed to bring high-performance intelligence directly to your laptop, and released under an Apache 2.0 license.
Bridging the gap between edge efficiency and advanced reasoning. Here is what’s new with Gemma 4 12B: 👇 -
Zeynep Akata reposted
Communications Medicine
@CommsMedicine
🚨 NEW RESEARCH! ⚕️ 🩺
Dani et al. examine the potential and limitation of LLMs in mapping seizure descriptions to seizure onset zones in #epilepsy care. 🧠
idp.nature.com/transit?redirect_uri=https%3A%2F%2Fwww.nat…
@dani_meghal
@MuthuJeyanthi1
@steliebe
@zeynepakata
@uni_tue
@uktuebingen -
Zeynep Akata reposted
Peter Holderrieth
@peholderrieth
We release Diamond Maps💎 unlocking accurate and efficient guidance for diffusion models. Our experiments show that our methods scale incredibly well. Excited to see what people will build with this!
Accurate guidance has been a notoriously hard problem, but in this work, we’re bringing TWO (!) solutions to the table. The recipe for success:
1️⃣ Speed: Use distilled models (flow maps, mean flows, consistency models).
2️⃣ Exploration: Inject stochasticity to properly explore your search space.
Because this fundamentally improves anything using flow matching and diffusion, we see a lot of potential for applications across audio, robotics, molecules, and beyond.
Paper: arxiv.org/abs/2602.05993
Code: github.com/PeterHolderrieth/diamond_maps
Huge thanks to an amazing team: Douglas Chen, @LucaEyring, @ishin_shah, Giri Anantharaman, @electronickale, @zeynepakata, Tommi Jaakkola, @nmboffi, and @max_simchowitz. It was awesome bringing this to life together! -
Zeynep Akata reposted
Luca Eyring
@LucaEyring
Accurate reward guidance in diffusion models has been a long-standing challenge.
We propose Diamond Maps 💎, through stochastic flow maps we make reward guidance both accurate and scalable. -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Excited to share that our paper “Diamond Maps: Efficient Reward Alignment via Stochastic Flow Maps” won the Best Paper Award at the #iclr2026 ReALM-GEN Workshop 🎉✨
Check out the paper & code 👇
📕 arxiv.org/abs/2602.05993
💻 github.com/PeterHolderrieth/diamond_maps -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Our EML members are heading for Rio🇧🇷 for #iclr2026! 🌴We’re excited to share our latest research — three poster presentations. Check out the thread below and come say hi to our authors
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
🥳Our proposal "TACO-VLM: Tackling challenges in large-scale multimodal learning" was granted 1.5M GPU-h on JUPITER @fz_juelich, Europe's first exascale supercomputer with ~24k NVIDIA GH200 GPUs. With JUPITER's support, we hope to push large-scale multimodal learning further.
-
Zeynep Akata reposted
Lukas Thede
@lukas_thede
🚨 New paper! arxiv.org/abs/2603.06610
Are you sure your post-trained LLM isn’t forgetting something?
Adapting LLMs is known to cause forgetting.
We usually measure it via general knowledge benchmarks.
But if MMLU doesn’t drop…. are you really fine? -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
🥳Happy to share that we have three papers accepted to #iclr2026. Congrats to our authors and see you in Rio🌴🇧🇷. Check the thread for highlights👇
-
Zeynep Akata
@zeynepakata
🚨 PI positions with an excellent package at Helmholtz Munich. Please apply, please share! jobs.helmholtz-muenchen.de/jobposting/83fa13f55624e8e5014…
-
Zeynep Akata
@zeynepakata
We are celebrating our papers and an invited talk at #neurips2025 🎊
Explainable Machine Learning
@ExplainableML
Our EML members have arrived in beautiful San Diego for #neurips2025! 🌴
We’re excited to share our latest research — three poster presentations and one workshop presentation.
Check out the thread below 👇 and come say hi to our authors! -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Our EML members have arrived in beautiful San Diego for #neurips2025! 🌴
We’re excited to share our latest research — three poster presentations and one workshop presentation.
Check out the thread below 👇 and come say hi to our authors! -
Zeynep Akata reposted
ELIAS
@elias_project
✨ELIAS @ EurIPS 2025!
Meet us at the Start-Up Village (3–5 Dec) to explore our latest work & the upcoming ELIAS Startup Opportunities Platform.
Join our workshop “Rethinking AI — Efficiency, Frugality & Sustainability” (6 Dec) for a deep dive into sustainable, responsible AI.🌍 -
Zeynep Akata reposted
Helmholtz Institute for Human-Centered AI
@cpilab
🚀 We are hiring! 🚀
🔍 Join us as a Postdoctoral Researcher (fully-funded) at the Helmholtz Institute for Human-Centered AI in Munich.
@HelmholtzMunich @helmholtz_ai @ELLISforEurope -
Zeynep Akata
@zeynepakata
We are looking for PhD students. Please apply through the @ELLISforEurope and @MunichCenterML PhD programs!
Explainable Machine Learning
@ExplainableML
🎓PhD application season @ExplainableML is back!
This year, we’re hiring ONLY through the ELLIS Program and the MCML Program: two fantastic communities shaping the future of AI and ML.
📌Please don’t forget to denote Prof. Dr. Zeynep Akata as one of your preferred supervisors!
👉 Link to ELLIS (ellis.eu/news/ellis-phd-program-call-for-applications-2025) and MCML (mcml.ai/opportunities/for-phd-applicants/phd2026/) -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
🎓PhD application season @ExplainableML is back!
This year, we’re hiring ONLY through the ELLIS Program and the MCML Program: two fantastic communities shaping the future of AI and ML.
📌Please don’t forget to denote Prof. Dr. Zeynep Akata as one of your preferred supervisors!
👉 Link to ELLIS (ellis.eu/news/ellis-phd-program-call-for-applications-2025) and MCML (mcml.ai/opportunities/for-phd-applicants/phd2026/) -
Zeynep Akata
@zeynepakata
Congratulations @LucaEyring for your Google PhD Award 2025 🎉
Explainable Machine Learning
@ExplainableML
We are proud of our PhD researcher Luca Eyring (@LucaEyring ) for his Google PhD Award 2025. Congratulations!🥳
-
Zeynep Akata
@zeynepakata
RT @ExplainableML: We are proud of our PhD researcher Luca Eyring (@LucaEyring ) for his Google PhD Award 2025. Congratulations!🥳
-
Zeynep Akata
@zeynepakata
What if your CBM doesn’t look at attributes when it explains the classification decision? We suggest to use this dataset and method to generate realistic images with unusual attributes to stress-test explainability methods. Jessie is @ICCVConference to talk to you about it!
Explainable Machine Learning
@ExplainableML
1/
SUB: Benchmarking CBM Generalization via Synthetic Attribute Substitutions
Jessica Bader, Leander Girrbach, Stephan Alaniz, @zeynepakata
[Paper]: arxiv.org/pdf/2507.23784
[Code]: github.com/ExplainableML/sub
📍Hall I #2142
🕑Oct 23, Poster Session 5 -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Are you at #iccv2025? 🎉 @ICCVConference
EML Munich is presenting two poster papers—come say hi to our authors!
Details in the thread 👇 -
Zeynep Akata
@zeynepakata
We got two papers of two orthogonal methods for interpretability and one paper of a GenAI method at @NeurIPSConf 2025 🔥 Please checkout the details in the thread below.
Explainable Machine Learning
@ExplainableML
🔥We celebrate 3 papers accepted to @NeurIPSConf 2025, see you in San Diego! 🥳Topics include diffusion models, sparse autoencoders (SAEs) and neural chunking. See the thread for highlights👇
-
Zeynep Akata reposted
Nathan Benaich
@nathanbenaich
🪩The one and only @stateofai 2025 is live! 🪩
It’s been a monumental 12 months for AI. Our 8th annual report is the most comprehensive it's ever been, covering what you *need* to know about research, industry, politics, safety and our new usage data.
My highlight reel:This media couldn't be displayed. See it on the original post. -
Zeynep Akata
@zeynepakata
It is an honor to receive the ZukunftsWissen Award from the German National Academy of the Sciences & Commerzbankstiftung 😊🎉🙏
Nationale Akademie der Wissenschaften Leopoldina
@Leopoldina
🎉🎬 @zeynepakata erhält den „ZukunftsWissen“-Preis von Leopoldina & Commerzbank-Stiftung. Die Informatikerin @TU_Muenchen @HelmholtzMunich spricht über ihre Forschung zur erklärbaren #ki. Zum vollständigen Video auf YouTube: www.youtube.com/watch?v=ZqIHyDOvb_k
This media couldn't be displayed. See it on the original post. -
Zeynep Akata
@zeynepakata
To integrate test-time scaling knowledge into a model during post-training in diffusion models, we replace reward guided test-time noise optimization with a Noise Hypernetwork that modulates initial input noise. Very interesting results. Checkout details 👇
Luca Eyring
@LucaEyring
Reward hacking is challenging when fine-tuning few-step Diffusion models. Direct fine-tuning on rewards can create artifacts that game metrics while degrading visual quality.
We propose Noise Hypernetworks as a theoretically grounded solution, inspired by test-time optimization. -
Zeynep Akata reposted
Luca Eyring
@LucaEyring
Reward hacking is challenging when fine-tuning few-step Diffusion models. Direct fine-tuning on rewards can create artifacts that game metrics while degrading visual quality.
We propose Noise Hypernetworks as a theoretically grounded solution, inspired by test-time optimization. -
Zeynep Akata reposted
Nataniel Ruiz
@natanielruizg
We are releasing a paper I'm very excited about. We know test-time scaling is a path to greatly improved results, and achieves reasoning in the case of LLMs. We present a new and promising way to amortize it into training using HyperNetworks for image generation models.
-
Zeynep Akata reposted
Tanishq Mathew Abraham, Ph.D.
@iScienceLuvr
Noise Hypernetworks: Amortizing Test-Time Compute in Diffusion Models
"we replace reward guided test-time noise optimization in diffusion models with a Noise Hypernetwork that modulates initial input noise."
"We show that our approach recovers a substantial portion of the quality gains from explicit test-time optimization at a fraction of the computational cost." -
Zeynep Akata
@zeynepakata
Congratulations Karsten @confusezius for a wonderful PhD journey that ended with a summa cum laude 🎉 It is rare that a PhD researcher creates a paradigm shift in a research group: I am grateful for everything that I learned from you 😉 Best of luck at @GoogleDeepMind in Zurich!
Karsten Roth
@confusezius
💫 After four PhD years on all things multimodal, pre- and post-training, I’m super excited for a new research chapter @GoogleDeepMind 🇨🇭!
Biggest thanks to @zeynepakata and @OriolVinyalsML for all the guidance, support, and incredibly eventful and defining research years ♥️! -
Zeynep Akata
@zeynepakata
Congratulations Karsten @confusezius for a wonderful PhD journey that ended with a summa cum laude 🎉 It is rare that a PhD researcher creates a paradigm shift in a research group: I am grateful for everything that I learned from you 😉 Best of luck at @GoogleDeepMind in Zurich!
Explainable Machine Learning
@ExplainableML
🎓PhD Spotlight: Karsten Roth
Celebrate @confusezius, who defended his PhD on June 24th summa cum laude!
Karsten has been an @ELLISforEurope and IMPRS-IS PhD student since May 2021, supervised by both @zeynepakata and @OriolVinyalsML. His research has been centered around robust and effective deployment of (large) neural networks in the real world, with particular focus on a method and data-centric perspective for :
🦾 (Multimodal) model pretraining
🧠 Model generalization, reuse and transferability research
🎆 Continual (multimodal) training of such models.
During his PhD, Karsten interned @AIatMeta and @GoogleDeepMind, working on generalization in representation learning and large-scale multimodal pretraining techniques.
🏁 His next stop: @GoogleDeepMind in Zurich!
Join us in celebrating Karsten's achievements and wishing him the best for his future endeavors! 🥳
👇Checkout his selected publications in top-tier conferences such as NeurIPS, ICLR, CVPR or ICCV: -
Zeynep Akata reposted
Sweta Mahajan
@SwetaMahajan1
Introducing the speakers for the eXCV workshop @ICCVConference. Get ready for stimulating and insightful talks and discussions.
Our Non-proceedings track is still open!
Paper submission deadline: July 18, 2025
More info at: excv-workshop.github.io -
Zeynep Akata reposted
Fartash Faghri
@FartashFg
Is your AI keeping Up with the world?
Announcing #neurips2025 CCFM Workshop: Continual and Compatible Foundation Model Updates
When/Where: Dec. 6-7 San Diego
Submission deadline: Aug. 22, 2025. (opening soon!)
sites.google.com/view/ccfm-neurips2025
#foundationmodels #continuallearning -
Zeynep Akata
@zeynepakata
It is a great honor to receive the ZukunftsWissen Prize 2025 from the German Academy of the Sciences @Leopoldina with generous support of the @CoBaStiftung 🎉 This achievement wouldn’t have been possible without my wonderful group @ExplainableML @TU_Muenchen @HelmholtzMunich
Nationale Akademie der Wissenschaften Leopoldina
@Leopoldina
🎉Der „ZukunftsWissen“-Preis 2025 von Leopoldina und @CoBaStiftung geht an @zeynepakata @TU_Muenchen/@HelmholtzMunich. Die Informatikerin erhält die mit 50.000 € dotierte Auszeichnung für ihre Forschung zur erklärbaren #ki.
www.leopoldina.org/presse-1/nachrichten/zukunftswissen-pr… -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
🎓PhD Spotlight: Shyamgopal Karthik
Celebrate @ShyamgopalKart1 , who will defend his PhD on 23rd June! Shyam has been a PhD student @uni_tue since October 2021, supervised by @zeynepakata.
His research has been centred around compositionality in vision and language. In particular, he has focused on:
🧠 Inference-time alignment and preference optimization for multimodal models
🏁 Composed retrieval, text-to-image generation, and uncertainty estimation
More recently, he interned at @Snap , where he worked on enhancing diffusion models through preference tuning.
Join us in celebrating Shyam’s achievements and wishing him the best for his future endeavors! 🥳
👇Checkout his selected publications in top-tier conferences such as CVPR, ICCV, ECCV, NeurIPS, ICLR: -
Zeynep Akata reposted
Sukrut Rao
@sukrutrao
Join us in taking stock of the state of the field of explainability in computer vision, at our Workshop on Explainable Computer Vision: Quo Vadis? at #iccv2025!
@ICCVConference @XAI_Research -
Zeynep Akata reposted
Karsten Roth
@confusezius
In Nashville for my last PhD conference 🥲.
Come join today 10:30-12:30 in Hall D (#391) to talk insights, tips and tricks to modify pretraining for representation reuse - scalably.
🚀Joint work w/ @zeynepakata, @dimadamen, @ibalazevic & @olivierhenaff while at @GoogleDeepMind. -
Zeynep Akata reposted
Rui Xiao
@RuiXiao99
Interested in VLM training with synthetic data? Wondering how to unlock fine-grained visual representations from captions?
I'm happy to present our work FLAIR @CVPR from 10:30 to 12:30 on 15th June, ExHall D Poster #368.
#cvpr2025 -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
📢 Landed in Nashville🎺 for #cvpr2025! The EML group is presenting 4 exciting papers — come say hi at our poster sessions! More details in the thread — see you there! 🏁🌟
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
🎓PhD Spotlight: Leonard Salewski
Celebrate @L_Salewski, who will defend his PhD on 24th June! 📸🎉
Leonard has been a PhD student in the EML and @CG_Tuebingen groups at @uni_tue since May 2020. He is part of the IMPRS-IS program (@MPI_IS), jointly supervised by @zeynepakata and Hendrik Lensch.
His research centers on explainable and multimodal AI, and can be summarized in three key contributions:
💡 Natural Language Explanations: Developing models, datasets, and benchmarks for generation and evaluation of natural language explanations.
🗣️ Multimodal Learning: Specialized in vision-language and audio-language models.
🧠 LLM Impersonation: Investigating LLM's in-context impersonation ability with its impact on performance and bias in downstream tasks.
During his PhD, Leonard broadened his expertise through an internship at @nvidia, working on yet another multimodal challenge—enabling vision-language models to understand and describe actions in popular computer games. 🎮🤖
Join us in celebrating Leonard’s achievements and wishing him the best for his defense and future endeavors! 🥳🎓
👇 Check out his amazing works in this thread🔥 -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
🎓PhD Spotlight: Thomas Hummel
A spotlight on our one and only @hummelth_ , who will defend his PhD on 23rd June! 🎉
Thomas started his PhD in 2020 at @uni_tue as part of the IMPRS-IS program under the supervision of @zeynepakata . His main research focuses on multi-modal learning and video understanding using language as guidance, in particular:
📽️👂Tackling audio-visual video classification in challenging zero- and few-shot scenarios
🤸🤔 Enhancing fine-grained understanding of how actions unfold
🎞️💡 Investigating fine-grained temporal reasoning capabilities of VLMs
Beyond this, he also contributed to semantic image synthesis using VQ-models and had the opportunity to intern at @SonyAI_global in Zurich, where his research focused on audio-language alignment for sound effects.
Throughout his PhD journey, Thomas built an impressive publication record—check out the highlights in the thread below. 👇
He was also recognized as an Outstanding Reviewer at ECCV 2024 and CVPR 2024.
We’re incredibly proud of Thomas’s accomplishments and can’t wait to see what he takes on next.
Congratulations on this major milestone! 👏📸🚀 -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
🚨Happy to announce that one paper, "Understanding the Limits of Lifelong Knowledge Editing in LLMs", is accepted at #icml2025 ! Congrats to the wonderful authors @lukas_thede , @confusezius , @MatthiasBethge ,@zeynepakata , and @tom_hartvigsen. 👇 Highlights in the thread
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
🎓PhD Spotlight: Jae Myung Kim
We’re thrilled to celebrate @jaemyung__kim , who will defend his PhD on 25th June! 🎉
Jae Myung began his PhD at @uni_tue as part of the ELLIS & IMPRS-IS programs, advised by @zeynepakata and collaborating closely with @CordeliaSchmid . His research centers on efficient, data-centric machine learning, with a focus on:
🎆 Leveraging generative models for synthetic data generation
🤔 Learning from weakly aligned or noisy data
💪 Advancing zero-shot and few-shot learning with foundation models
Beyond this, he has contributed to reliable ML, including work on explainability, bias, and uncertainty.
We’re proud of Jae’s accomplishments and can’t wait to see what he takes on next. Congratulations on this huge milestone! 👏🚀
👇Check this thread for his incredible works: -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
#cvpr2025 is heading to the 'Music City' — Nashville! 🎺 Join us from June 11–15. We're thrilled to announce that we'll be presenting four papers at @CVPR! Check out the thread below for highlights, and feel free to stop by and chat with our authors! 📷👇
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
📢 We’ve landed in Singapore for #iclr2025!
The EML group is presenting 4 exciting papers — come say hi at our poster sessions! 👇Let's chat!
More details in the thread — see you there! 🌟 -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
🌏 #iclr2025 is coming to the beautiful city of Singapore — April 24–28!
We're excited to share 4 upcoming papers being presented at the conference.
Check out the thread for highlights, and come chat with the authors if you're interested! 🧵👇 -
Zeynep Akata reposted
Theo Uscidda
@theo_uscidda
Curious about the potential of optimal transport (OT) in representation learning? Join @CuturiMarco's talk at the UniReps workshop today at 2:30 PM! Marco will notably discuss our latest paper on using OT to learn disentangled representations. Details below ⬇️
-
Zeynep Akata
@zeynepakata
It is a great honor to be named as one of the Germany’s Top 40 Under 40 by the @capitalMagazin in 2024 🎉
Helmholtz Munich | @HelmholtzMunich
@HelmholtzMunich
Prof. Zeynep Akata is one of the leading "Top 40 Under 40" - shaping the future of Germany!👏
The Capital Magazine honored the #helmholtzmunich scientist as one of the talents shaping Germany by driving positive change across diverse fields!
👉Interview: t1p.de/epcuu
💬"As a scientist, it is my wish to be useful for the society as a whole and not only influence my scientific community but the society as a whole. In that sense, it is a great honor to be recognized by the Capital Magazine as one of the Top 40 Under 40."
Prof. Zeynep Akata
🌟As the Director of the Institute of Explainable Machine Learning at #helmholtzmunich, Zeynep Akata is dedicated to foundational machine learning research: Her work focuses on developing adaptable and generalizable foundation models. This research is vital for scientific progress across disciplines.
💡Discover more about her thoughts on how to extend the limits of our curiosity with the help of technological tools in her #interview!
@zeynepakata @CompHealthMuc @TU_Muenchen
#ai #machinelearning #ml #foundationmodels #research #health
#top40under40 -
Zeynep Akata
@zeynepakata
We are looking for two postdoctoral researchers in our @ExplainableML group @TU_Muenchen @HelmholtzMunich @ELLISforEurope, application deadline is November 19th (fully funded, flexible starting dates, flexible topics, duration negotiable). More details 👇
-
Zeynep Akata
@zeynepakata
The deadline for PhD applications in our @ExplainableML group @TU_Muenchen @HelmholtzMunich is tomorrow! Checkout the @ELLISforEurope PhD program if you are interested (we won’t evaluate email applications, MSc degree is a must, flexible starting dates) ellis.eu/news/ellis-phd-program-call-for-applications-2024
-
Zeynep Akata reposted
Kempner Institute at Harvard University
@KempnerInst
Watch Eric Schulz from @cpilab explain how testing LLMs with psychology tasks reveals new insight into how they work and how they can be improved. The latest talk at the #kempnerinstitute: youtu.be/-QubTOvK9mM
-
Zeynep Akata
@zeynepakata
Checkout two postdoc positions fully funded by @TU_Muenchen and @HelmholtzMunich! The application deadline is November 19th. Please apply if you are interested in our core research topics and have a good track record in ML and CV. More details in the link👇
Explainable Machine Learning
@ExplainableML
‼️👩💻🧑💻 POTENTIAL POST-DOC APPLICANTS
We are looking for two qualified post-docs to join our lab at Helmholtz Munich and TUM: if you are interested in our work, check out our application at www.eml-munich.de/jobs. We look forward to hearing from you! -
Zeynep Akata reposted
Helmholtz Institute for Human-Centered AI
@cpilab
We are happy to share Centaur! Led by @marcellbinz and a big collaboration within and beyond our lab, Centaur is a foundation model of human behavior. It can predict and simulate human behavior for any cognitive experiment and beats the best cognitive model in every experiment.
Marcel Binz
@marcel_binz
Excited to announce Centaur -- the first foundation model of human cognition. Centaur can predict and simulate human behavior in any experiment expressible in natural language. You can readily download the model from @huggingface and test it yourself: huggingface.co/marcelbinz/Llama-3.1-Centaur-70B-adapter
-
Zeynep Akata
@zeynepakata
We are looking for two PhD students via the @ELLISforEurope PhD program fully funded by @TU_Muenchen and @HelmholtzMunich!
ELLIS
@ELLISforEurope
The portal is open: Our #ellisphd Program is now accepting applications! Apply by November 15 to work with leading #ai labs across Europe and choose your advisors among 200 top #machinelearning researchers!
#joinellisforeurope #phd #phdprogram #ml ellis.eu/news/ellis-phd-program-call-for-applications-202…
Zeynep Akata
@zeynepakata
Here is the application portal for the 2024 call (for positions starting in 2025): ellis.eu/news/ellis-phd-program-call-for-applications-2024
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
🔥 We’re looking forward to presenting two papers at NeurIPS 2024, come check us out!
ReNO: Enhancing One-step Text-to-Image Models through Reward-based Noise Optimization
www.eml-munich.de/publication/ReNO
A Practitioner’s Guide to Continual Multimodal Retraining
arxiv.org/abs/2408.14471 -
Zeynep Akata reposted
Simone Scardapane
@s_scardapane
*A Practitioner’s Guide to Continual Multimodal Pretraining*
by @confusezius @vishaal_urao @MatthiasBethge @zeynepakata et al.
An in-depth monograph on best practices for continual learning of multimodal models under compute (but not memory) budgets. 🔥
arxiv.org/abs/2408.14471 -
Zeynep Akata
@zeynepakata
If you are interested in continual pretraining in foundational multimodal models, this is a great resource! Impressively hard work lead by @confusezius and @vishaal_urao
Karsten Roth
@confusezius
Keeping multimodal foundation models relevant as time progress and new data arises is crucial.
In our new paper, we provide a comprehensive guide on how continual multimodel pretraining can be done effectively in realistic scenarios:
arxiv.org/abs/2408.14471 -
Zeynep Akata
@zeynepakata
Checkout our #eccv2024 paper on how we do successful interventions on wrongly predicted concepts that explain a fine-grained classification decision.
Nishad Singhi
@nishadsinghi
🚀 Excited to share our new paper, appearing at #eccv2024! Here’s a quick dive into our research on improving the receptiveness of neural networks to test-time concept interventions. 🧵👇
-
Zeynep Akata reposted
Computational Health Center
@CompHealthMuc
👏 We are thrilled to have this many of our brilliant researchers at the @icmlconf in Vienna! Below you will find featured work of @zeynepakata, @vincefort, @Pseudomanifold, @GKaissis, @jeremy_wayland, and other exceptional researchers from @HelmholtzMunich and @helmholtz_ai 👇
-
Zeynep Akata reposted
GreenFOMOWorkshop@ECCV2024
@Green_FOMO
🚨There is still time to submit your work to #eccv2024 workshop on Green Foundation Models. Join us in our efforts to make CV greener. ♻️
Submission deadline is 12th July. ⏳
More details can be found on our webpage green-fomo.github.io/ECCV2024/ -
Zeynep Akata
@zeynepakata
An interesting paper, congratulations @cpilab and the team!
Center for Hybrid Intelligence
@CfHybridIntel
Excited to share our latest collaborative paper, "Using Games to Understand the Mind," now published in Nature Human Behaviour!🎉 Congratulations to @jacobsherson , @jrafner , and the entire team! Dive into new insights on using games to study cognition and human behavior.
-
Zeynep Akata
@zeynepakata
The abstract submission is June 24th. No registration fees, the space is limited. We are looking forward to meeting you!
ELLIS Munich unit
@ELLISmunichunit
📢Join us on July 8th at #helmholtzmunich for the ELLIS / Helmholtz AI Workshop: "Connecting Minds and Machines: a Foundation Model approach to Learning" Keynotes by @RichardShiffrin, @ZeynepAkata, and @RahafAljundi and more!
👉All details: eml-munich.de/events/external_events
#ai #ml -
Zeynep Akata reposted
Otniel-Bogdan Mercea
@MerceaOtniel
Excited to share that I’ll be presenting today our work on “Audio-Visual Generalized Zero-Shot Learning using Pre-Trained Large Multi-Modal Models” at the L3D-IVU workshop at @CVPR. Come to poster 125 to find more about our work. (1/4)
-
Zeynep Akata
@zeynepakata
If you are interested in rehearsal free continual learning, you should check this CoLLAs2024 paper from @lukas_thede. You may be surprised about his findings.
Lukas Thede
@lukas_thede
With the rise of foundation models, Continual Learning (CL) has shifted towards adapting large-scale pretrained models, often without rehearsing past data (RFCL). But how effective are these new methods? Let’s dive in! 🧵👇
-
Zeynep Akata
@zeynepakata
The emergence of one-step SD models enables being creative: in this preprint we are optimizing the noise of the SD models using several human-preference reward metrics at inference time. The results are wonderful! Please check it out!
Luca Eyring
@LucaEyring
Can we enhance the performance of T2I models without any fine-tuning?
We show that with our ReNO, Reward-based Noise Optimization, one-step models consistently surpass the performance of all current open-source Text-to-Image models within the computational budget of 20-50 sec! -
Zeynep Akata
@zeynepakata
A simple and effective alternative to LoRa finetuning, to be presented at #icml2024
Massimo Bini
@mwbini
❔Is constraining the finetuning to low-rank matrices (s.a. in LoRA) the best approach?
In our new ICML paper (ETHER) we propose to learn hyperplane reflections of pretrained weights as an alternative
• fewer parameters (from 1 vector per layer)
• interesting properties!
1/6 -
Zeynep Akata reposted
Massimo Bini
@mwbini
❔Is constraining the finetuning to low-rank matrices (s.a. in LoRA) the best approach?
In our new ICML paper (ETHER) we propose to learn hyperplane reflections of pretrained weights as an alternative
• fewer parameters (from 1 vector per layer)
• interesting properties!
1/6 -
Zeynep Akata reposted
Kristian Kersting
@kerstingAIML
Looks like an exciting event
www.eml-unitue.de/events/external_events -
Zeynep Akata
@zeynepakata
Looking forward to it!
Cohere Labs
@Cohere_Labs
Next Wednesday, June 5th, join @beyzaermis as she sits down with @zeynepakata for a conversation on "Interpretable and Reliable Machine Learning via Communication!"🔥🛋️
Register Now: zoom.us/webinar/register/WN_-j2VBDYWR2KisLRHkYDl7Q -
Zeynep Akata reposted
GreenFOMOWorkshop@ECCV2024
@Green_FOMO
🌿 Exciting news! Join us at ECCV 2024 (@eccvconf) for our workshop on Green Foundation Models! 🌍✨
Discover the latest in sustainable AI research and network with experts dedicated to eco-friendly advancements.
Don't miss out—let's make AI greener together! 💚 #eccv2024 -
Zeynep Akata reposted
Helmholtz Munich | @HelmholtzMunich
@HelmholtzMunich
Ever wondered about the psychological insights behind AI's behavior?
🎤Dive into our #interview with our #helmholtzmunich AI-expert Dr. Eric Schulz & discover how #machinepsycholoy & #foundationmodels revolutionize human #health.
👉t1p.de/uyqai
#ai -
Zeynep Akata reposted
Computational Health Center
@CompHealthMuc
Congratulations to our @HelmholtzMunich @CompHealthMuc PIs and students showcasing their work at the #iclr2024 @zeynepakata @Pseudomanifold @vincefort @k__niki @fabian_theis and more
-
Zeynep Akata
@zeynepakata
Our group is presenting three papers at #iclr2024 . Unfortunately I couldn’t go this year but @confusezius, @LucaEyring and @lukas_thede are there. Please go and talk to them, they’ll be happy to see you!
Explainable Machine Learning
@ExplainableML
Join us at #iclr2024 on Wednesday and Friday for
🗓️May 8th 4:30pm, S4: Training-free text-conditional image retrieval,
🗓️May 8th 4:30pm, S4: Unbalanced optimal transport for unpaired translation tasks,
🗓️May 10th 4:30pm, S8 - Spotlight: General model merging and transfer! -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Also today (May 8th, 4:30pm at Session 4)! Ask @LucaEyring, @Dominik1Klein and @theo_uscidda on how to utilize unbalanced optimal transport to handle undesired distribution shifts in unpaired translation tasks!
With @g_palla1, @k__niki, @zeynepakata, and @fabian_theis!
Luca Eyring
@LucaEyring
Dealing with undesired distribution shifts in unpaired translation tasks? Our #iclr2024 paper shows how to mitigate them leveraging Unbalanced OT!
We propose a method to incorporate unbalancedness into any neural Monge map estimator and demonstrate the benefits of unbalancedness. -
Zeynep Akata
@zeynepakata
If you are interested in image retrieval with compositional captions without training or tuning the model, please talk to @confusezius at #iclr2024 after his presentation today. Details below!
Explainable Machine Learning
@ExplainableML
Today (May 8th, 4:30pm at Session 4) @confusezius will present our work on image retrieval under arbitrary text conditions - without explicit training data and model tuning required!
Great work by @ShyamgopalKart1 @confusezius @massi_manc @zeynepakata ! -
Zeynep Akata reposted
Luca Eyring
@LucaEyring
Excited to attend my first conference in person #iclr2024
Join us at the poster session on Wednesday 16:30-18:30 at poster #238! @Dominik1Klein @theo_uscidda
Luca Eyring
@LucaEyring
Dealing with undesired distribution shifts in unpaired translation tasks? Our #iclr2024 paper shows how to mitigate them leveraging Unbalanced OT!
We propose a method to incorporate unbalancedness into any neural Monge map estimator and demonstrate the benefits of unbalancedness. -
Zeynep Akata reposted
Burcu Sayin
@SayinBurcu
📣 Call for Papers: The 2nd Workshop on Hybrid Human-Machine Learning & Decision Making (HLDM’24). 🗓️ Deadline: June 15, 2024. 🌟 Confirmed Keynotes: Prof. @NickJChater, Prof. @zeynepakata, and Prof. @nuriaoliver 🔗 Website: sml.disi.unitn.it/hldm24.html
-
Zeynep Akata
@zeynepakata
I am looking forward to speaking at the #helmholtzimaging Conference on 14-15 May!
Helmholtz Munich | @HelmholtzMunich
@HelmholtzMunich
❗Join the 4th #helmholtzimaging Conference
📅 14-15 May 2024
📢Register & submit your abstract by 𝟮𝟳 𝗠𝗮𝗿𝗰𝗵 𝟮𝟬𝟮𝟰
t1p.de/va835
#hiconference2024 Highlights:
Keynotes by #helmholtzmunich expert @zeynepakata (@TU_Muenchen) & Joost Batenburg (Leiden Uni) -
Zeynep Akata
@zeynepakata
Fresh blogpost from Luca on our ICLR24 paper in the topic of unpaired translation using optimal transport. We show results not only on i2i but also single cell translation tasks. Please have a look!
Luca Eyring
@LucaEyring
Dealing with undesired distribution shifts in unpaired translation tasks? Our #iclr2024 paper shows how to mitigate them leveraging Unbalanced OT!
We propose a method to incorporate unbalancedness into any neural Monge map estimator and demonstrate the benefits of unbalancedness. -
Zeynep Akata reposted
Yuki
@y_m_asano
Next up is @zeynepakata from @helmholtz_ai and TUM speaking about interpretable vision and language assistants (VLAs). @ELLISmunichunit a wealth of research combining these two modalities yielding increased explainability.
-
Zeynep Akata reposted
-
Zeynep Akata
@zeynepakata
The deadline is this Friday. Looking forward to your applications!
Zeynep Akata
@zeynepakata
🚀 Exciting Opportunity Alert! 🚀
🔍 Join us as a Postdoctoral Researcher (fully funded) at the Helmholtz Munich Institute for Explainable Machine Learning and TUM Chair of Interpretable and Reliable Machine Learning! @TU_Muenchen @Helmholtz_Munich @ELLISforEurope -
Zeynep Akata
@zeynepakata
Waseda Symphony Orchestra Tokyo gave a mind blowing concert in Munich just now at Isarphilarmonie. A wonderful way to advertise a university! 😊
-
Zeynep Akata reposted
Xiaoxiang ZHU
@xiaoxiang_zhu
Why, how and when #xai in #remotesensing? as a follow-up action of our #ml4earth workshop on #xai, we have complied a symmetrical overview on the topic under the lead of my brilliant PhD students @ObadicIvica and Adrian, in collaboration with @miguelangelft @zeynepakata #ai4eo
-
Zeynep Akata reposted
Helmholtz Institute for Human-Centered AI
@cpilab
📢 Exciting opportunity ahead!
The Helmholtz Institute for Human-Centered AI in Munich is looking to hire two PhD students to work on the intersection of computational cognitive science and machine learning with a starting date in 2024. @HelmholtzMunich @ELLISforEurope -
Zeynep Akata
@zeynepakata
🚀 Exciting Opportunity Alert! 🚀
🔍 Join us as a Postdoctoral Researcher (fully funded) at the Helmholtz Munich Institute for Explainable Machine Learning and TUM Chair of Interpretable and Reliable Machine Learning! @TU_Muenchen @Helmholtz_Munich @ELLISforEuropePost not available here, view on X
Zeynep Akata
@zeynepakata
🔬 Possible Research Topics:
✨ Multimodal Learning
✨ Explainable ML
✨ Large Foundational Models
✨ Scene & Video Understanding
✨ Low-shot Learning & more!
👩💻👨💻 Join us in shaping the future of Explainable Machine Learning!
Zeynep Akata
@zeynepakata
📝 How to Apply:
Send your application to zeynep.akata@tum.de with subject line [EML open topic postdoc application] by March 15th, 2024.
Don't hesitate to reach out with any questions!
💡 More info: www.eml-unitue.de/jobs/postdoc_adverisement.png
#jobopening #academictwitter #mlresearch -
Zeynep Akata
@zeynepakata
More information (paper, code, blogpost) on our #iclr2024 spotlight paper "Fantastic Gains and Where to Find Them - On the Existence and Prospect of General Knowledge Transfer": www.eml-munich.de/publication/Fantastic-Gains-and-Where-t…
Karsten Roth
@confusezius
🤔 Weight averaging is great, but can you merge arbitrary trained models?
Our #iclr2024 spotlight shows that you can - and should!
For example, we find that weak models have unique knowledge missed by bigger and better ones, and careful integration can make them even stronger! -
Zeynep Akata
@zeynepakata
🌟Exciting news alert! 🌟
I am beyond thrilled to share that in January I am joining @HelmholtzMunich as a director & @TU_Muenchen as a Liesel Beckman distinguished professor! This opportunity marks an incredible leap in my career journey, one I'm deeply grateful for.Post not available here, view on X
Zeynep Akata
@zeynepakata
At Helmholtz, leading the Institute for Explainable Machine Learning, and at TUM, chairing the Interpretable and Reliable Machine Learning group, I'm eager to move forward explainable ML and multimodal learning.
Zeynep Akata
@zeynepakata
I am grateful for this chance to advance our research and shape the future of @ExplainableML. Let's embrace this journey together! 🚀✨We're on the lookout for exceptional talent to join our research groups—stay tuned for opportunities!
-
Zeynep Akata reposted
Leonard Salewski
@L_Salewski
Are you interested in #llm's and their ability to impersonate? Visit us during the upcoming poster session 2 at #1426, where Stephan Alaniz and I will present our #neurips2023 Spotlight.
Paper: arxiv.org/abs/2305.14930
Leonard Salewski
@L_Salewski
Thrilled to share that my research on In-Context Impersonation in Large Language Models has been accepted as a spotlight paper at #neurips2023 🎉🚀!
Joint work with Stephan Alaniz, Isabel Rio-Torto, @cpilab and @zeynepakata.
Check out the paper here: arxiv.org/abs/2305.14930 -
Zeynep Akata reposted
Marcel Binz
@marcel_binz
We asked four groups of researchers: How should the advent of large language models affect the practice of science? This led to an exciting discussion covering a broad spectrum of opinions. arxiv.org/abs/2312.03759 and osf.io/preprints/osf/yr9xb
-
Zeynep Akata reposted
Leonard Salewski
@L_Salewski
Do you want to know how we earned a #neurips2023 Spotlight for our "In-Context Impersonation Reveals Large Language Models' Strengths and Biases" paper 🧐?
Buckle up and join for a ride on the rebuttal roller coaster 🎢🧵 -
Zeynep Akata reposted
Helmholtz Institute for Human-Centered AI
@cpilab
🚨Pre-print alert:🚨
Have we built machines that think like people?
In new work, led by @lucaschubu and @elifakata and together with @MatthiasBethge, we assess multi-modal #llms reasoning abilities in three core domains: intuitive physics, causality, and intuitive psychology. -
Zeynep Akata reposted
Thomas Hummel
@hummelth_
Excited to share that I'll be presenting our work on "Video-adverb retrieval with compositional adverb-action embeddings" as an Oral today in beautiful Aberdeen at #bmvc2023 @BMVCconf! 🎉(1/4)
-
Zeynep Akata reposted
Lara Urban
@LaraUrban42
It is with great sadness that I have to resign as @eLife editor and early-career advisor after @mbeisen was fired as editor-in-chief for making use of his freedom of speech to stand up – once again – for the people who were silenced. A brief statement here:
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
📅Today - Oct. 5th, 2:30pm - Sud #161
Learn more about potential pitfalls of LLM usage across modalities, particularly for CLIP-style models!
We show semantic sanity checks, and how to improve CLIP with random words and characters, and better leverage LLMs semantics.
#iccv2023
Karsten Roth
@confusezius
CLIP + LLMs: A modern success story 🌟.
LLM provides detailed class context, feeds it back into CLIP - and you have yourself an improved open-vocabulary classifier 💫.
🤔Turns out the same gain can often be achieved with random words and characters instead.
#iccv2023 -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
📅Friday Oct. 6th, 10:30am - Sud #041
We recompose 3D objects in the wild into their semantic parts without training data!
"Iterative Superquadric Recomposition of 3D Objects from Multiple Views" is joint work by Stephan Alaniz, @massi_manc and @zeynepakata
#iccv2023 -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
📅Friday Oct. 6th, 10:30am - Sud #138
Zero-shot learning without images from seen classes! Using a pre-trained network as training data, we generate new classifiers for unseen targets.
@AndersChriii, @massi_manc, A. Sophia Koepke, @OleWinther1, @zeynepakata
#iccv2023 -
Zeynep Akata
@zeynepakata
Very excited about coming to #iccv2023 and presenting 4 papers from our
@ExplainableML group (👇is one) at the main conference in person! My first since 2019 😄
Karsten Roth
@confusezius
CLIP + LLMs: A modern success story 🌟.
LLM provides detailed class context, feeds it back into CLIP - and you have yourself an improved open-vocabulary classifier 💫.
🤔Turns out the same gain can often be achieved with random words and characters instead.
#iccv2023 -
Zeynep Akata
@zeynepakata
Congratulations @L_Salewski, Isabel, Stephan and @cpilab 🎉 this was a real turnaround! More details will follow shortly, stay tuned…
Leonard Salewski
@L_Salewski
Thrilled to share that my research on In-Context Impersonation in Large Language Models has been accepted as a spotlight paper at #neurips2023 🎉🚀!
Joint work with Stephan Alaniz, Isabel Rio-Torto, @cpilab and @zeynepakata.
Check out the paper here: arxiv.org/abs/2305.14930 -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
At #gcpr23 today? Then come to our poster on "Text-to-feature diffusion for audio-visual few-shot learning" by @MerceaOtniel, @hummelth_, A. Sophia Koepke and @zeynepakata! Find out more about our work here: t.ly/xjX34
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Also come say hi at #gcpr23 if you want to talk about our work on “Zero-shot Translation of Attention Patterns in VQA Models to Natural Language” by @L_Salewski, A. Sophia Koepke, Hendrik P.A. Lensch and @zeynepakata. Check out the paper here: t.ly/WjF4N
-
Zeynep Akata
@zeynepakata
Proud of you, Eric! Can’t find too many more words to say… I have seen from the first hand how much work went into this. Very well deserved! 🤗🥰
Helmholtz Institute for Human-Centered AI
@cpilab
Extremely happy that our #ercstg proposal #tacos got funded! Over the next 5 years, we will use tools from cognitive science to better understand large language models. This really took a village. Thanks to my lab, colleagues, and family for all of their support. @ERC_research
-
Zeynep Akata
@zeynepakata
Thank you @SabrinaPatsch for her interesting questions, guidance and translation in this interview. It was a great pleasure talking to her!
Sabrina Patsch
@SabrinaPatsch
#kuenstlicheintelligenz ist undurchsichtig: Meist ist unklar, wie ein Ergebnis zustande kam und ob es vertrauenswürdig ist. @zeynepakata von @ml4science @uni_tue will das ändern – mit erklärbarem #machinelearning.
@Tagesspiegel (T+)
www.tagesspiegel.de/wissen/blick-in-die-blackbox-konnen-w… -
Zeynep Akata
@zeynepakata
A big congratulations to our Karsten Roth @confusezius for winning a QIF Europe award 🎉
Qualcomm Research & Technologies
@QCOMResearch
QIF Europe is an excellence award through which @Qualcomm rewards and mentors the most innovative PhD students in Europe working on breakthrough #ai and #cybersecurity solutions. Congratulations @tychovdo @confusezius @SiweiZhang13 and Attri Bhattacharyya www.qualcomm.com/news/releases/2023/07/outstanding-resear…
-
Zeynep Akata
@zeynepakata
It is a great honour to receive this award. I'd like to thank Alfried Krupp von Bohlen und Halbach-Foundation for their generosity, @uni_tue for the nomination, @ml4science @MPI_IS @MPICybernetics @Cyber_Valley for the excellent research environment they create for us.
-
Zeynep Akata reposted
Helmholtz Institute for Human-Centered AI
@cpilab
For more than two years now, a part of our lab has been focusing on using the behavioral sciences to better understand Large Language Models. Time for a small recap of what we have found. As these agents will interact with us increasingly more often, how do they actually behave?
-
Zeynep Akata
@zeynepakata
Compositionality capabilities of image generation with diffusion models have been extensively discussed & several model improvements have been proposed. In an orthogonal way, we show that diffusion models are actually compositional if you dig them deep.
Shyamgopal Karthik
@ShyamgopalKart1
We are much better at evaluating text-to-image alignment these days e.g. breaking up the process into a series of question-answers, or directly measure it using a model trained on human preferences. Can we use this to improve the generated images?
👉 arxiv.org/abs/2305.13308 -
Zeynep Akata
@zeynepakata
New pre-print: In-Context Impersonation Reveals Large Language Models' Strengths and Biases. Lead by @L_Salewski and in collaboration with Stephan Alaniz, Isabel Rio-Torto, @cpilab. arxiv.org/abs/2305.14930 1/4
Post not available here, view on X
Zeynep Akata
@zeynepakata
We evaluate the usefulness of descriptions generated by in-context impersonation for vision and language tasks by using them for CLIP zero-shot classification. On CUB and Stanford Cars we see increasing performance with age and expertise and reveal biases wrt race and gender. 3/4
Zeynep Akata
@zeynepakata
Check out the full paper (arxiv.org/abs/2305.14930) for more results on in-context impersonating LLMs replicating child-like behavior on multi-arm bandits and LLM task experts outperforming domain and non-domain experts in MMLU reasoning. 4/4
-
Zeynep Akata
@zeynepakata
I am looking forward to giving a keynote at the #helmholtzaicon 2023 in Hamburg!
Helmholtz AI
@helmholtz_ai
Get ready for an unforgettable conference experience! #helmholtzaicon 2023 🎙️ We're thrilled to announce our Keynote speaker lineup, and joining us on Day 1 is @zeynepakata, Professor of Computer Science at the University of Tübingen. 📢 Save the date: 12 June in Hamburg!
-
Zeynep Akata reposted
Jimmy Wales
@jimmy_wales
What Wikipedia did: we stood strong for our principles and fought to the Supreme Court of Turkey and won. This is what it means to treat freedom of expression as a principle rather than a slogan.
-
Zeynep Akata
@zeynepakata
Congratulations @cpilab 🎉 it was a pleasure to work with you, I have learned a lot! Hopefully more will come 😊
Helmholtz Institute for Human-Centered AI
@cpilab
It's a huge honor to have our work on understanding large language models using tools from psychology and psychiatry featured in this week's @derspiegel. This was a big team effort working together with @juliancodaforno @WitteKristin @akjagadish, Marcel Binz, and @zeynepakata.
-
Zeynep Akata
@zeynepakata
After 3 extremely productive years with us, our @massi_manc has decided to move on to build his own group @UniTrento. Please consider applying to work with him, we have absolutely loved it, I am sure you would too!
-
Zeynep Akata
@zeynepakata
Proud to be a part of this work! LLMs have gotten so well at standard conversational tasks, now it is time to evaluate how easy it is to effect their psychology and make them biased 😲
Helmholtz Institute for Human-Centered AI
@cpilab
New pre-print: Using computational psychiatry to understand LLMs, we find that GPT-3.5 responds to anxiety-inducing prompts with increased exploration and biases. Joint work with @juliancodaforno @WitteKristin @akjagadish Marcel Binz and @zeynepakata.
arxiv.org/abs/2304.11111 -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Aaand it's a wrap!
We would like to thank everyone who joined, all the organizers and particularly the speakers who offered their valuable time to make this workshop a success with insights into their important research! -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
We conclude our workshop on Explainability in Machine Learning with a final talk by @mreyesag on insights, challenges and opportunities of Interpretability for Deep Learning based Medical Image Computing Technologies!
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Our second-to-last invited talk by @mormontre offered some very intriguing insights on the use of Singular Value Decomposition for Concept Discovery and Dataset Exploration!
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
👋 Great way to start the second day of the Explainability in Machine Learning Workshop with an in-depth presentation on Interpretability methods for Deep Learning in Computer Vision by Bernt Schiele!
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Fantastic panel discussion to close the first workshop day, answering important questions on the role of explainability in the current AI landscape!
With @c_f_baumgartner @trevordarrell @ruthcfong @mormontre @UlrikeLuxburg @mreyesag @WojciechSamek and Bernt Schiele -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
💬 Great talk by Lisa Dunlap @UCBerkeley on using language to improve generalization of image recognition systems to new domains!
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
What directions are interesting and relevant for future research into interpretability? Valuable insights by @ruthcfong during her talk at the Explainability in Machine Learning Workshop!
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
💭 Thought-provoking talk on the key components behind and the fundamental understanding of explainability by @UlrikeLuxburg!
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
We are starting our two-day Explainability in Machine Learning Workshop with a great talk on Concept-Level Explainable AI by @WojciechSamek!
-
Zeynep Akata
@zeynepakata
Happy international women‘s day! 🎉
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Interested in XAI? Do not miss the Explainability in ML workshop!
It will take place March 28-29 in Tübingen (Germany) and will feature talks from prominent researchers in the field.
Check the program here www.eml-unitue.de/eml-workshop, and register in advance: few spots remaining! -
Zeynep Akata reposted
Karsten Roth
@confusezius
Happy to say that our work on unsupervised disentanglement ✍️ even under correlations ✍️ has found its way to #iclr23 🥳!
With @marksibrahim @zeynepakata @pascal20100 @D_Bouchacourt)
-- Paper: arxiv.org/abs/2210.07347
Karsten Roth
@confusezius
Unsupervised disentanglement methods commonly assume statistical independence of generative factors.
But correlations are everywhere!
To better disentangle under such correlations, we suggest a novel relaxation focused on a factorized support instead!
arxiv.org/abs/2210.07347 -
Zeynep Akata reposted
Helmholtz Institute for Human-Centered AI
@cpilab
Perhaps it's a good time to announce that this got accepted at PNAS 🎉 For everyone (still) out there trying psychological prompts, we had tried some as well, but prefer programmatically generated task, e.g. bandit tasks or the two-step task. Will tweet more once it's in press.
Helmholtz Institute for Human-Centered AI
@cpilab
Many things have been said about large language models (including GPT-3). In new work with @atze_dota, we used tools from cognitive psychology to study how GPT-3 makes decisions, searches for information, deliberates, and reasons causally, and compared its behavior to humans.
-
Zeynep Akata
@zeynepakata
I am looking forward to speaking at the @CambridgeEllis seminar series on 30th November afternoon!
CambridgeEllisUnit
@CambridgeEllis
We are looking forward to our next Seminar Speaker @zeynepakata from @uni_tue who will speaking about "Explainability in Deep Learning Through Communication", join us here: www.ellis.eng.cam.ac.uk/cambridge-ellis-unit-seminar-seri… #machinelearning #ai
-
Zeynep Akata
@zeynepakata
It was a pleasure to speak at the #ml4earth #xai workshop in Munich today @xiaoxiang_zhu and her group are doing super exciting work at @TU_Muenchen and @helmholtz_ai
Xiaoxiang ZHU
@xiaoxiang_zhu
Ongoing - super exciting keynote from Prof Zeynep Akata @zeynepakata @uni_tue on #xai for #vision #nlp and #earthobservation at our #ml4earth xAI workshop organized by @TU_Muenchen @Zhu_XLab @ai4eo_de @helmholtz_ai
-
Zeynep Akata
@zeynepakata
It’s an honor to receive the ECVA Young Researcher Award 2021 🎉 Special thanks to my former PhD advisor @CordeliaSchmid for the nomination & the executive board of ECVA for the recognition
Yizhak Ben-Shabat (Itzik)
@sitzikbs
ECVA young researcher awards: congrats @zeynepakata
-
Zeynep Akata
@zeynepakata
We are presenting the four papers below in the main conference of #eccv2022!
Explainable Machine Learning
@ExplainableML
👋At #eccv2022 and looking for posters to visit?
Interested in an @ELLISforEurope or #imprsis PhD with us?
Talk to us at the poster sessions, with topics from audio-visual zero-shot learning, sketch abstraction, uncertainty quantification to probabilistic contrastive learning: -
Zeynep Akata reposted
Accepted papers at TMLR
@TmlrPub
Attribute Prediction as Multiple Instance Learning
Diego Marcos, Aike Potze, Wenjia Xu, Devis Tuia, Zeynep Akata
openreview.net/forum?id=nmFczdJtc2 -
Zeynep Akata
@zeynepakata
We are also looking for PhD students through the #ellisphd program (joint PhD in two institutions from two different European countries). Deadline November 15. Application link: ellis.eu/news/ellis-phd-program-call-for-applications-2022
Explainable Machine Learning
@ExplainableML
Come work with us as part of the #ellisphd family if you are interested in research ranging from explainable and interpretable ML, multimodal and zero-shot learning to contrastive and out-of-distribution capable representation learning!
-
Zeynep Akata
@zeynepakata
We are looking for PhD candidates (deadline November 1st) and looking forward to your applications!
Explainable Machine Learning
@ExplainableML
If you are interested in joining our Explainable Machine Learning Group (www.eml-unitue.de) and to work on topics ranging from XAI, OOD Generalization and Multimodal Learning to Zero-Shot Learning, consider applying through the great IMPRS-IS graduate program!
Zeynep Akata
@zeynepakata
Application link: imprs.is.mpg.de/application
-
Zeynep Akata reposted
Gerard Pons-Moll
@GerardPonsMoll1
Congrats to Yongqin @ChibaneJulian @bharat_b7 for receiving the best paper honorable mention at 3DV'22!
Nice collaboration with @zeynepakata and Bernt Schiele!
We show how to better generalise single image 3D reconstruction to novel object classes:
virtualhumans.mpi-inf.mpg.de/gin/ -
Zeynep Akata
@zeynepakata
I am looking forward to giving a keynote at @UncertaintyInAI at 9am CEST today. I will talk about explainability via conversations and my thoughts on uncertainty.
-
Zeynep Akata reposted
Shyamgopal Karthik
@ShyamgopalKart1
Check out our latest work where we provide uncertainty estimates for "free" to your favourite image translation model! @uddupa @yanbei_c @massi_manc @zeynepakata @ExplainableML @uni_tue @ml4science #eccv2022
AK
@_akhaliq
A @Gradio Demo for BayesCap: Bayesian Identity Cap for Calibrated Uncertainty in Frozen Neural Networks on @huggingface Spaces for @europython 2022
demo: huggingface.co/spaces/EuroPython2022/BayesCap
Submit demos/models/datasets to the EuroPython 2022 organization: huggingface.co/organizations/EuroPython2022/share/JsnIYMR… -
Zeynep Akata reposted
ud
@uddupa
You've trained your new comp. vision model, but it is deterministic, and are wondering how to obtain calibrated uncertainty for it?
Check out our work at ECCV-22 BayesCap: Calibrated Uncertainty in Frozen Neural Networks
ppr:arxiv.org/pdf/2207.06873 by
@ExplainableML @zeynepakata -
Zeynep Akata
@zeynepakata
4 papers are accepted to #eccv2022 🎉 The topics range from explainable sketch abstraction, uncertainty, audio-visual zero-shot learning, deep metric learning. More details will follow @ExplainableML. Supported by @uni_tue @ml4science @MPI_IS @ELLISforEurope
-
Zeynep Akata
@zeynepakata
It was great to meet @SergeBelongie today at the IMPRS-IS evaluation event and present to a large audience after such a long break!
Serge Belongie
@SergeBelongie
Prof. @ZeynepAkata presents her group’s work on zero-shot learning w/external knowledge @uni_tue @MPI_IS
-
Zeynep Akata reposted
Karsten Roth
@confusezius
If contrastive learning and language is something that sounds interesting, drop by at this mornings oral (or poster) session! #cvpr2022
AK
@_akhaliq
Integrating Language Guidance into Vision-based Deep Metric Learning
abs: arxiv.org/abs/2203.08543 -
Zeynep Akata reposted
Karsten Roth
@confusezius
If you want to know how contrastive learning and normalizing flows go together, come say 👋 at todays morning poster session #cvpr2022!
Explainable Machine Learning
@ExplainableML
🗓️06/22 | Sess 2.1 (210a)
"Non-isotropy Regularization for Proxy-based Deep Metric Learning"
Make your contrastive learning model go :rocket: with Normalising Flows & Proxies!
@confusezius @OriolVinyalsML @zeynepakata
Blog: www.eml-munich.de/publication/Non-isotropy-Regularization… -
Zeynep Akata
@zeynepakata
Interesting read to find out how GPT-3 compares with humans wrt several well established criteria from cognitive psychology.
Helmholtz Institute for Human-Centered AI
@cpilab
Many things have been said about large language models (including GPT-3). In new work with @atze_dota, we used tools from cognitive psychology to study how GPT-3 makes decisions, searches for information, deliberates, and reasons causally, and compared its behavior to humans.
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Today is another workshop day at #cvpr2022!
Members of our group are participating in several workshops as well presenting some cool works, so check out some of them 😊: -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
🗓️ 06/20 | 8:30AM - 1:00pm
✏️ 5th Multimodal Learning and Applications Workshop (MULA)
Covers topics ranging from supervised cross- or multimodal learning to unsupervised approaches thereof!
Co-organized by @massi_manc @zeynepakata -
Zeynep Akata
@zeynepakata
We are presenting six papers in the main conference of #cvpr2022 next week. Come and join us in New Orleans!
Explainable Machine Learning
@ExplainableML
If you are at #cvpr2022, come check out one of our six publications!
We have topics ranging from training on partially labelled dataset to (multi-modal) contrastive metric learning and all manners of zero-shot learning!
A quick 🧵 (sorted by date): -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Great to see four members of our team (@massi_manc Almut Sophia Koepke @ShyamgopalKart1 @confusezius) recognized for their review work #cvpr2022!
#CVPR2027
@CVPR
List of #cvpr2022 OUTSTANDING reviewers now available. Thank you for your service!
cvpr2022.thecvf.com/outstanding-reviewers -
Zeynep Akata reposted
Serge Belongie
@SergeBelongie
U. Tübingen PhD student @confusezius presents recent work on Deep Metric Learning w/@OriolVinyalsML @zeynepakata at the Pre-CVPR event @AiCentreDK #cvpr2022
-
Zeynep Akata
@zeynepakata
I am looking forward to speaking at the #growingupinscience event next Tuesday ☺️
Mindful Science
@mindfulsci_tue
Join our next #growingupinscience 🥼🧠 on May 24th 6-7pm 📆 with @zeynepakata to learn about her story, struggles, and doubts on her journey to become a professor of #computerscience 🖥 Register with the link the bio!
-
Zeynep Akata
@zeynepakata
If you‘d like to have academic independence and collaborate with us in Tübingen immediately after your PhD, this is not to miss! We sincerely hope that strong female candidates will apply as well.
-
Zeynep Akata reposted
Karsten Roth
@confusezius
🚀Supercharge Metric Learning w/
"Non-isotropy Regularization for Proxy-based Deep Metric Learning"
#cvpr22 @OriolVinyalsML @zeynepakata @ELLISforEurope
Norm. Flows + proxies = strong generalization & super fast training.
arxiv.org/abs/2203.08547
github.com/ExplainableML/NonIsotropicProxyDML
1/4 -
Zeynep Akata reposted
Karsten Roth
@confusezius
Visual Similarity 🤝 Language in
"Integrating Language Guidance into Vision-based Deep Metric Learning"
#cvpr22 @OriolVinyalsML @zeynepakata @ELLISforEurope
Equip visual similarity models w/ lang. semantics at little cost!
arxiv.org/abs/2203.08543
github.com/ExplainableML/LanguageGuidance_for_DML
1/3 -
Zeynep Akata
@zeynepakata
Happy international women’s day! I am incredibly grateful to all the strong women who I had the pleasure of meeting and working together throughout my journey.
-
Zeynep Akata
@zeynepakata
6 papers are accepted to #cvpr2022 🎉 congratulations to my students and postdocs! More details will follow @ExplainableML. Supported by @uni_tue @ml4science @MPI_IS @ELLISforEurope
-
Zeynep Akata
@zeynepakata
10 days left to apply!
Zeynep Akata
@zeynepakata
We are looking for a Research Fellow (E14 TV-L, 100%) in @ExplainableML team @ml4science and @uni_tue in collaboration with @trevordarrell from @UCBerkeley. Deadline: 15 February. Come and join us for this exceptional opportunity!
uni-tuebingen.de/en/128980 -
Zeynep Akata reposted
Michael Black
@Michael_J_Black
I get a lot of reviews that say my work is not novel and I bet I'm not alone. It's always frustrating because I see novelty where the reviewer doesn't. Rather than rebut every critique, I've written a blog post to help reviewers think about novelty. perceiving-systems.blog/en/news/novelty-in-science
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
We continue with the work spotlight!
If you're interested in searching for sounds by description:
Check out this recent work (with demo!) on Audio Retrieval with Natural Language Queries by A.-M. Oncescu, A.S. Koepke, J.F. Henriques, @zeynepakata, @SamuelAlbanie @Oxford_VGG -
Zeynep Akata reposted
Max Planck Institute for Biological Cybernetics
@MPICybernetics
The research team around Eric Schulz @cpilab and Georg Martius @GMartius received a fund of 1.5 million euros. The team aims to translate curiosity and interest into program code that promotes self-motivated learning behavior in robotic systems. More: sohub.io/le3l
-
Zeynep Akata
@zeynepakata
Fine-Grained Zero-Shot Learning with DNA as Side Information at #neurips2021 Thursday Dec 9th at poster session 4 with @badirli_ George Mohler, @SciencePicard and @mm_dundar
Paper: openreview.net/pdf?id=RqAzAoL8BER -
Zeynep Akata
@zeynepakata
Happening now!
Explainable Machine Learning
@ExplainableML
With #neurips '21 right around the corner, make sure to check out
"Robustness via Uncertainty-aware Cycle Consistency" w/ @uddupa, @yanbei_c & @zeynepakata!
@uni_tue @MPI_IS
and come visit us next week!
1/3 -
Zeynep Akata
@zeynepakata
We are looking for a Research Fellow (E14 TV-L, 100%) in @ExplainableML team @ml4science and @uni_tue in collaboration with @trevordarrell from @UCBerkeley. Deadline: 15 February. Come and join us for this exceptional opportunity!
uni-tuebingen.de/en/128980 -
Zeynep Akata
@zeynepakata
Come and join us on Tuesday Dec 7th at 8:30AM-10AM PST #neurips2021
Explainable Machine Learning
@ExplainableML
With #neurips '21 right around the corner, make sure to check out
"Robustness via Uncertainty-aware Cycle Consistency" w/ @uddupa, @yanbei_c & @zeynepakata!
@uni_tue @MPI_IS
and come visit us next week!
1/3 -
Zeynep Akata reposted
RWTH Artificial Intelligence Colloquium
@ColloquiumRwth
Save the date for the second talk of the #aic: Prof. Zeynep Akata (@zeynepakata) from Universität Tübingen @uni_tue will join us to talk about #explainable visual recognition. Freely accessible with registration: www.ai.rwth-aachen.de/go/id/roxaj/lidx/1
-
Zeynep Akata
@zeynepakata
I am looking forward to collaborating with such a team diverse in every dimension! If you are looking for a PhD in the intersection of #ml & #cognitivescience please apply! @MPICybernetics @DeepMind @uni_tue @ml4science
Helmholtz Institute for Human-Centered AI
@cpilab
We are looking to hire a fully-funded PhD-student to work on Cognitive Models of Meta-Reinforcement Learning. The student would be co-supervised by Eric and @janexwang, as well as collaborate with @zeynepakata and Matt Botvinick:
cpilab.org/img/PhD_Job_Ad.pdf
Deadline: Dec 21st -
Zeynep Akata
@zeynepakata
If you’d like to do a PhD with us and collaborate with another research lab in Europe, apply to @ELLISforEurope PhD program. We are accepting applications until November 15th!
Explainable Machine Learning
@ExplainableML
If you are interested in doing a PhD with us, consider applying through the ELLIS PhD program. Deadline: November 15th. ellis.eu/phd-postdoc
-
Zeynep Akata
@zeynepakata
We are accepting PhD applications! Deadline November 2nd.
Explainable Machine Learning
@ExplainableML
If you are interested in doing a PhD with us, consider applying through the IMPRS-IS program. Deadline: November 1st. imprs.is.mpg.de/application
-
Zeynep Akata reposted
Helmholtz Institute for Human-Centered AI
@cpilab
Please take a look at our latest work. With @mona_garvert @TankredSaanum @nico_schuck @doellerlab
-
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Starting today (friday, 2pm CEST), we are also offering a lecture on "learning with limited labeled data" for master students @uni_tue.
If you're interested in e.g. zero- and few-shot learning or self-supervised learning, check out eml-unitue.de/teaching/seminars for more details! -
Zeynep Akata reposted
Explainable Machine Learning
@ExplainableML
Starting today (2pm CEST), we are offering a lecture on "advanced topics in explainable machine learning" for master students @uni_tue.
See www.eml-unitue.de/teaching/seminars for more details! -
Zeynep Akata
@zeynepakata
Join us in #iccv2021 Session 7 at 8am EDT (2pm CEST) today and 3pm EDT (9pm CEST) on Friday if you are interested in Explainable #ml @uni_tue @ml4science @MPI_IS and at Naver AI Lab
Seong Joon Oh
@coallaoh
Introducing a new paper with Jae Myung Kim, @junsukchoe, @zeynepakata @ ICCV 2021:
Keep CALM and improve your visual feature attribution.
Paper: arxiv.org/abs/2106.07861
Code: github.com/naver-ai/calm
Keep reading if you have used CAM before - worth your time! [1/13] -
Zeynep Akata
@zeynepakata
Come and visit us in Session 1 #iccv2021 on Tuesday 12.10 at 10am EDT (4pm CEST) and on Thursday 14.10 at 5pm EDT (11pm CEST) @ExplainableML @uni_tue @ml4science @UniofOxford @MPI_IS
Maxime Kayser
@maximek3
🆕 Paper at #iccv2021💥
😈e-ViL: A Dataset and Benchmark for Natural Language Explanations in Vision-Language Tasks (VL-NLE)
arxiv.org/abs/2105.03761
with @oanacamb, L. Salewski, @CorEmde, @zeynepakata, and T. Lukasiewicz
@UniofOxford @uni_tue
#ml #cv #xai #nlproc
(1/6) -
Zeynep Akata reposted
Samarth Sinha
@_sam_sinha_
Are you an undergrad looking to get involved in ML research?
If so, then attend out our ICCV 2021 Social about undergrads in CV! Hear from experts and a student panel in the field about research and how to get involved!
Link: iccv2021-ug-in-cv.github.io/UG_in_CV/
Please RT for visibility! -
Zeynep Akata
@zeynepakata
It is an honor to receive this award today. Thanks to the German Pattern Recognition Society for their support. It is great to be in Tübingen and be a part of @ml4science, @uni_tue, @MPI_IS
-
Zeynep Akata reposted
ud
@uddupa
Excited to attend the MICCAI conference this week!
Please stop by our poster titled "Uncertainty-Guided Progressive GANs for Medical Image Translation" on September 29th (Wednesday) from 11:30 - 13:00 (CEST).
Link to paper: arxiv.org/pdf/2106.15542.pdf
@MICCAI_Society
#miccai2021 -
Zeynep Akata reposted
Maxime Kayser
@maximek3
🆕 Paper at #iccv2021💥
😈e-ViL: A Dataset and Benchmark for Natural Language Explanations in Vision-Language Tasks (VL-NLE)
arxiv.org/abs/2105.03761
with @oanacamb, L. Salewski, @CorEmde, @zeynepakata, and T. Lukasiewicz
@UniofOxford @uni_tue
#ml #cv #xai #nlproc
(1/6) -
Zeynep Akata reposted
Visual Geometry Group (VGG)
@Oxford_VGG
Want to search for sounds by description? A.-M. Oncescu is presenting work on Audio Retrieval with Natural Language Queries right now at @INTERSPEECH21!
www.robots.ox.ac.uk/~vgg/research/audio-retrieval/
with A. S. Koepke, J. F. Henriques, @zeynepakata, @SamuelAlbanie @Oxford_VGG @ExplainableML -
Zeynep Akata reposted
Seong Joon Oh
@coallaoh
Introducing a new paper with Jae Myung Kim, @junsukchoe, @zeynepakata @ ICCV 2021:
Keep CALM and improve your visual feature attribution.
Paper: arxiv.org/abs/2106.07861
Code: github.com/naver-ai/calm
Keep reading if you have used CAM before - worth your time! [1/13] -
Zeynep Akata
@zeynepakata
Checkout 'ImageNet: past, present and future' workshop at #neurips2021 for a great lineup of speakers and consider submitting your contributions (Deadline: September 18th)! @coallaoh @dlarlus @giffmana @SanghyukChun @XiaohuaZhai
Lucas Beyer (bl16)
@giffmana
Our NeurIPS'21 workshop on "ImageNet: past, present, and future" has been accepted!
I'm excited about our speaker line-up. I'm even more excited to see what papers researchers will submit to the workshop!
Please spread the word, and consider submitting.
sites.google.com/view/imagenet-workshop/home -
Zeynep Akata reposted
Oana-Maria Camburu
@oanacamb
Happening tomorrow! If you are attending @ICML2021, come by our tutorial on Natural-XAI: Explainable AI with Natural Language Explanations, Monday 19th, 8 - 11 a.m PDT / 4 - 7 p.m. BST!
@zeynepakata #xai
Oana-Maria Camburu
@oanacamb
Very excited to announce that I will be giving a #tutorial at #icml2021 together @zeynepakata, on Natural-XAI: Explainable AI with Natual Language Explanations!
-
Zeynep Akata reposted
Robert Jenssen
@jenssen_robert
Call for papers for the northern lights deep learning conference! www.nldl.org/contributors Amazing keynote speakers e.g @SergeBelongie @zeynepakata @SAnaniadou
-
Zeynep Akata reposted
inzva
@inzvaspace
📢We are delighted to have @zeynepakata at the CVPR Meetup Turkey to share 4 different papers of hers under the topic of Explainability and Compositionality in Deep Visual Recognition.
…1/10 -
Zeynep Akata
@zeynepakata
Happening today in the tenth session of #cvpr2021 (11am EDT, 5pm CET) paper ID: 5223. Come and check it out!
Zeynep Akata
@zeynepakata
Check out our @CVPR'21 paper: Our RDTC model combines classical #machinelearning with #deeplearning to improve model explanation. Joint work with Stephan Alaniz, Diego Marcos, Bernt Schiele
Paper: arxiv.org/abs/1902.01780
Code: github.com/ExplainableML/rdtc -
Zeynep Akata
@zeynepakata
Happening today/tomorrow in the fifth session of #cvpr2021 (22pm EDT, 4am CET) paper ID: 3368. Come and check it out!
Zeynep Akata
@zeynepakata
Check out our #cvpr21 paper ‘Distilling Audio-Visual Knowledge by Compositional Contrastive Learning’ joint work with @yanbei_c @xyongqin, A.Sophia Koepke, Ying Shan
Paper: arxiv.org/abs/2104.10955
Code: github.com/ExplainableML/CCL -
Zeynep Akata
@zeynepakata
Happening today in the fourth session of #cvpr2021 (11am EDT, 5pm CET) paper ID: 7256. Come and check it out!
Zeynep Akata
@zeynepakata
Can we recognize state-object compositions within ~30k classes? Yes! Check out our @CVPR
'21 paper: Open World Compositional Zero-Shot Learning. Joint work with @massi_manc, @ferjadnaeem, @xyongqin
Paper: arxiv.org/abs/2101.12609
Code: github.com/ExplainableML/czsl -
Zeynep Akata
@zeynepakata
Happening today in the first session of #cvpr2021 (11am EDT, 5pm CET) paper ID: 1098. Come and check it out!
Zeynep Akata
@zeynepakata
Check out our #cvpr21 paper ‘Learning Graph Embeddings for Compositional Zero-shot Learning’ joint work with @ferjadnaeem @xyongqin and @fedassa
Paper: arxiv.org/pdf/2102.01987.pdf
Code: github.com/ExplainableML/czsl -
Zeynep Akata reposted
Nivii Kalavakonda
@nkalavak
We are bringing together a few amazing leaders in computer vision for a panel @WiCVworkshop during #cvpr2021.
Pls submit questions for the panel here (by June 9th): docs.google.com/forms/d/e/1FAIpQLSfBSRCqqn56MzAqYqu7J8BjX…
@zeynepakata @orussakovsky @judywawira @NataliaNeverova @ftmguney
Kindly help amplify -
Zeynep Akata reposted
WiCV
@WiCVworkshop
We are very excited to share the second fully virtual Women in Computer Vision workshop as part of CVPR 2021 is happening on Saturday June 19. Thanks to our sponsors, participants, and volunteers, we have a full schedule of invited talks, a panel discussion, and mentoring session
-
Zeynep Akata reposted
Oana-Maria Camburu
@oanacamb
Very excited to announce that I will be giving a #tutorial at #icml2021 together @zeynepakata, on Natural-XAI: Explainable AI with Natual Language Explanations!
-
Zeynep Akata
@zeynepakata
Check out our #cvpr21 paper ‘Distilling Audio-Visual Knowledge by Compositional Contrastive Learning’ joint work with @yanbei_c @xyongqin, A.Sophia Koepke, Ying Shan
Paper: arxiv.org/abs/2104.10955
Code: github.com/ExplainableML/CCL
Zeynep Akata
@zeynepakata
Rather than directly aligning the representations of different modalities, we compose representations across modalities to uncover richer multi-modal knowledge for better recognition and retrieval results on video datasets.
Zeynep Akata
@zeynepakata
Our compositional embedding closes the cross-modal semantic gap and captures the task-relevant semantics by pulling together audio, image, and video representations.
-
Zeynep Akata
@zeynepakata
Check out our #cvpr21 paper ‘Learning Graph Embeddings for Compositional Zero-shot Learning’ joint work with @ferjadnaeem @xyongqin and @fedassa
Paper: arxiv.org/pdf/2102.01987.pdf
Code: github.com/ExplainableML/czsl
Zeynep Akata
@zeynepakata
We propose a graph embedding framework to model dependency relations between objects, states and their compositions for compositional zero-shot learning. Our compositional graph can be created on the fly without any external knowledge base for a wide range of compositional tasks.
Zeynep Akata
@zeynepakata
We also propose a new benchmark for the task of compositional zero-shot learning containing 9k+ compositions! We set a new SOTA on all benchmarks achieving 2x the performance of the closest baseline.
-
Zeynep Akata
@zeynepakata
Can we recognize state-object compositions within ~30k classes? Yes! Check out our @CVPR
'21 paper: Open World Compositional Zero-Shot Learning. Joint work with @massi_manc, @ferjadnaeem, @xyongqin
Paper: arxiv.org/abs/2101.12609
Code: github.com/ExplainableML/czsl
Zeynep Akata
@zeynepakata
We explore compositional zero-shot learning (CZSL) in the open world scenario, where no constraints exist in the compositional space. We tackle this problem with CompCos which assigns a feasibility score to each composition.
Zeynep Akata
@zeynepakata
CompCos uses these scores both to exclude less feasible compositions from the output space and as margins in the loss function, achieving competitive results in closed world CZSL and the state of the art in open world CZSL.
-
Zeynep Akata
@zeynepakata
Check out our @CVPR'21 paper: Our RDTC model combines classical #machinelearning with #deeplearning to improve model explanation. Joint work with Stephan Alaniz, Diego Marcos, Bernt Schiele
Paper: arxiv.org/abs/1902.01780
Code: github.com/ExplainableML/rdtc
Zeynep Akata
@zeynepakata
We solve the classification task in a two-agent framework. The recurrent decision-tree agent asks binary questions and the attribute-based agent provides the yes/no answer w.r.t. the image.
Zeynep Akata
@zeynepakata
RDTC can also provide fine-grained explanations by using human-understandable attributes as binary splits while at the same time being more scalable and flexible than other neural decision trees.
-
Zeynep Akata
@zeynepakata
We have an open PhD position (fully funded, E13 TV-L, 80%) in compositional data synthesis with @anna_khoreva in @Cyber_Valley @uni_tue (Application deadline: 31 May) For more information: cyber-valley.de/de/jobs/phd-in-compositional-data-synthesis
-
Zeynep Akata
@zeynepakata
Full Professor (W3) opening in #computerscience at the University of Tübingen @uni_tue, in the same campus as @ml4science, @MPI_IS, @MPICybernetics, @MPI_for_DB, @Cyber_Valley. Deadline for applications: March 12th!
cyber-valley.de/de/jobs/full-professor-w3-of-distributed-… -
Zeynep Akata
@zeynepakata
Looking forward to giving a talk at the TUM AI lecture series tomorrow.
Matthias Niessner
@MattNiessner
Tomorrow in our TUM AI - Lecture, we have @zeynepakata talking about Explainability and Compositionality for Visual Recognition!
Live Stream here: youtu.be/wQOkyxqXNhc
Starts tomorrow at 6pm CEST / 9am PST! -
Zeynep Akata reposted
Helmholtz Institute for Human-Centered AI
@cpilab
Our lab (PI: Eric Schulz) at @MPICybernetics in Tübingen is looking to hire 2 fully-funded postdocs
P1 (joint with @GMartius) will work on computational theories of curiosity
P2 will work on any topic related to our group.
Deadline: Feb 28th 2021
Start: April 2021 (negotiable) -
Zeynep Akata
@zeynepakata
This 👇
Helmholtz Institute for Human-Centered AI
@cpilab
In a new paper just out in @iScience_CP and together with Peter Dayan, we propose to turn the lens of computational psychiatry onto computers to illuminate their systematic failures. www.sciencedirect.com/science/article/pii/S258900422030969X
-
Zeynep Akata
@zeynepakata
We are looking for a PhD student (fully funded) within the IMPRS program. The deadline for applications is November 2nd: imprs.is.mpg.de/application
-
Zeynep Akata
@zeynepakata
Twitter has become a better place then!
Ulrike Luxburg
@UlrikeLuxburg
Finally started a twitter account ...
@CellTypist @PhilippHennig5 @jakhmack @zeynepakata @MatthiasBethge -
Zeynep Akata
@zeynepakata
Checkout our #neurips2020 paper 'Attribute Prototype Network for Zero-Shot Learning': a network that can visually explain its behavior by learning attribute prototypes that are then used to predict and localise class-discriminative attributes for an image.
Zeynep Akata
@zeynepakata
Our attribute prototypes can transfer knowledge from seen to unseen classes boosting the performance of zero-shot learning via locality enhanced image representations that can visualise class discriminative attributes accurately. arxiv.org/abs/2008.08290
Zeynep Akata
@zeynepakata
Joint work with: @wenjia43411575 @xyongqin @WangJiuniu and Bernt Schiele.
-
Zeynep Akata
@zeynepakata
ELLIS offers an interdisciplinary PhD program where every PhD student is supervised by one ELLIS fellow/scholar and one ELLIS member from a different country and conducts a 1 year exchange at the other location. Application deadline: 1.12.2020 ellis.eu/de/news/ellis-phd-program-call-for-applications
-
Zeynep Akata
@zeynepakata
Looking forward to the #wicv workshop at #eccv2020 !
WiCV
@WiCVworkshop
We are pleased to welcome @dimadamen, @FidlerSanja and @zeynepakata as our guest speakers at #wicv. They will be joined by Rahul Sukthankar for a panel discussing diversity in computer vision. Leave your anonymous questions for the panel here: forms.gle/hsxSo5NN3p2S3ujcA
-
Zeynep Akata
@zeynepakata
#daad offers a postdoctoral networking tour in #ml/#ai Come and visit us in @uni_tue @Cyber_Valley (Deadline: 16 August)
www.daad.de/en/the-daad/postdocnet/details-and-application/ -
Zeynep Akata
@zeynepakata
Looking forward to the #riiaa2020 summer school and conference on August 25-28!
RIIAA
@riiaa_org
#riiaa2020 is close and we are very happy to continue announcing more amazing speakers participating this year. Now is the turn of @zeynepakata from the University of Tübingen, panelist of Computer Vision #ai
-
Zeynep Akata
@zeynepakata
The talks from last weeks #ml4science2020 conference are on youtube now! 👇
-
Zeynep Akata reposted
DeepAI
@DeepAI
Evaluation for Weakly Supervised Object Localization: Protocol, Metrics, and Datasets
deepai.org/publication/evaluation-for-weakly-supervised-o…
by Junsuk Choe et al. including @coallaoh, @SanghyukChun, @zeynepakata
#hyperparameter #computervision -
Zeynep Akata reposted
SanghyukChun
@SanghyukChun
Happy to announce our CVPR 2020 paper on a new evaluation for weakly-supervised object localization (WSOL)! WSOL methods have many issues. E.g. often not truly "weakly-supervised". We fix the issues. Authors: @junsukchoe @coallaoh Seungho Lee @zeynepakata Hyunjung Shim (1/13)
-
Zeynep Akata reposted
Albert Ali Salah
@SzassTam
Bilim Akademisi'nin Yapay Öğrenme Yaz Okulu bu sene tamamen çevrimiçi ve kayıt ücreti yok. yazokulu.bilimakademisi.org/yapayogrenme/2020/ @BilimAkademisi @BilkentUniv @zeynepakata @TaylanCemgilML @ftmguney
@cmericli @oflazer @gcinbis #yapayogrenme #machinelearning @sarkac_org -
Zeynep Akata
@zeynepakata
We are looking for a postdoc to join our new Explainable Machine Learning lab within the Cluster of Excellence @uni_tue starting as soon as possible
uni-tuebingen.de/en/128980
#machinelearning #nlproc #computervision -
Zeynep Akata reposted
Aykut Erdem
@aykuterdemml
Our recent ACM Transaction on Graphics (TOG) paper is featured on Two Minute Papers (@karoly_zsolnai)☺️Levent (@lvntkrcn) will also present this work at #siggraph2020! @erkuterdem @zeynepakata @HUCVL
-
Zeynep Akata reposted
Marco Federici
@mfederici_
Looking forward to presenting our work "Learning Robust representations via Multi-View Information Bottleneck" at #iclr2020. Join the poster session this Wednesday at 8 am or 8 pm GMT. @duttanjan84 @NateKushman @zeynepakata
poster link: iclr.cc/virtual/poster_B1xwcyHFDr.html -
Zeynep Akata reposted
DeepAI
@DeepAI
e-SNLI-VE-2.0: Corrected Visual-Textual Entailment with Natural Language Explanations
deepai.org/publication/e-snli-ve-2-0-corrected-visual-tex…
by Virginie Do et al. including @zeynepakata
#computation #language -
Zeynep Akata
@zeynepakata
We are looking for independent research group leaders for Machine Learning combined with #nlp, #linguistics and other disciplines in Cluster of Excellence #machinelearning at the University of Tübingen @uni_tue (deadline for application is April 30).
uni-tuebingen.de/en/128980 -
Zeynep Akata
@zeynepakata
The German Conference on Pattern Recognition will take place from September 28 to October 1 in Tübingen @uni_tue (will be a virtual conference if necessary). Paper submission deadline is July 10, 2020.
www.gcpr-vmv-vcbm-2020.uni-tuebingen.de?page_id=102 -
Zeynep Akata reposted
Cyber Valley
@Cyber_Valley
The @EU_Commission offers significant support to Europe’s #ai excellence. It has decided to fund ELISE, a project by @ELLISforEurope that specifically focuses on fundamental research in #machinelearning, with 12 million Euros: fcai.fi/news/the-european-commission-offers-significant-s… #supportingellis
-
Zeynep Akata reposted
Sandro Pezzelle
@sandropezzelle
Working on projects on #nlp, #computervision, real-world #knowledge and beyond? 🚀 Consider submitting your long/short paper at #lantern2020 at @coling2020!
Deadline🏁: MAY 20
Info: shorturl.at/ackz5
See the picture for some accepted topics!
#nlproc #cv #ai #deeplearning -
Zeynep Akata reposted
Cyber Valley
@Cyber_Valley
Scientists from #cybervalley shape the future of European #ai: visiting us to gain insights into our #research is Margrethe @Vestager, Executive Vice President of the @EU_Commission for a #europe Fit for the Digital Age, Winfried Kretschmann @RegierungBW @TheresiaBauer @uni_tue
This media couldn't be displayed. See it on the original post. -
Zeynep Akata reposted
Bernhard Schölkopf
@bschoelkopf
„Data science and machine learning conferences (NeurIPS, ICML, AISTATS, ICLR, UAI, ...): Allow remote paper & poster presentations at conferences” - www.change.org/p/organizers-of-data-science-and-machine-l…
-
Zeynep Akata reposted
Cyber Valley
@Cyber_Valley
Die Stiftung Werner-von-Siemens-Ring hat 9 junge Forscher*innen in das Netzwerk hervorragender Jungwissenschaftler*innen gewählt. Darunter sind gleich zwei aus dem #cybervalley: @zeynepakata von der @uni_tue und Andreas Dörr vom @Bosch_AI: siemens-ring.de/wp-content/uploads/2019/12/Pressemitteilu… #ki #maxplanck
-
Zeynep Akata reposted
Rodolfo (Rudy) Corona
@_rodolfocorona_
Excited to present our spotlight+poster today at #neurips2019 on “Modeling Coneptual Understanding in Image Reference Games”!
Spotlight: 10:20AM, Track 4 Session 5, West Ballroom C.
Poster: 79, 10:45 East Hall B+C.
With Stephan Alaniz and @zeynepakata.
arxiv.org/abs/1910.04872 -
Zeynep Akata
@zeynepakata
Excited to share our work on “Modeling Conceptual Understanding in Image Reference Games”, appearing as a spotlight in #neurips 2019
Paper: arxiv.org/abs/1910.04872
Code: github.com/rcorona/conceptual_img_ref
with @_rodolfocorona_ and Stephan AlanizPost not available here, view on X
Zeynep Akata
@zeynepakata
We present an image reference game between artificial agents with variations in how well they understand visual attributes. To succeed, the speaker must learn to adapt to the listener.
Zeynep Akata
@zeynepakata
While leveraging gameplay history and past experience to succeed in the task, our speaker agent also provides an interpretable representation of listener agents’ understanding of attributes.
-
Zeynep Akata
@zeynepakata
Congratulations to Anna Sepliarskaia for successfully defending her PhD thesis @UvA_Amsterdam. It is great to wear one of these robes (and the neck tie that traditionally goes with it)😄with Sonja Smets and @hanedhinda.
-
Zeynep Akata
@zeynepakata
We are looking for PhD students for projects related to Explainable Machine Learning, the application deadline is November 6th!
imprs.is.mpg.de/application -
Zeynep Akata
@zeynepakata
Deeply honoured to be awarded an ERC-2019 Starting Grant!
The grant (DEXIM) comes with a funding of 1.5M EUR to pursue cutting edge research on ML, language and vision, aiming to explain the decisions of machine learning models to the user.
erc.europa.eu/news/StG-recipients-2019









































































































































