iiinova — where your digital self keeps evolving

ELLIOT Project @elliot-eu.bsky.social · 21h

🌍 On 26 Nov, ELLIOT joins ELIAS in Milan for the Theme Development Workshop on AI for a Sustainable Society. The event will explore AI’s role in societal & environmental challenges, emerging research directions and new collaborations. 🔗 elias-ai.eu/event/tdw-on...

000

ELLIOT Project @elliot-eu.bsky.social · 02/10/2026

How can autonomous driving AI become more efficient? 🚗 🔬We highlight DrivoR by @valeoai.bsky.social, a transformer-based approach that compresses multi-camera information into compact scene representations while supporting safer, more adaptive driving decisions. Read it👉 arxiv.org/pdf/2601.05083

000

ELLIOT Project @elliot-eu.bsky.social · 01/10/2026

ELLIOT in Ljubljana! 🇸🇮 Cees Snoek delivered the keynote at the inaugural ELLIS Unit Slovenia event, taking the audience from 500M years of vision and language to ELLIOT’s multimodal AI and his latest work on multilingual image generation. 🧠 Thanks to the ELLIS Unit Slovenia for the invitation! 🚀

020

ELLIOT Project @elliot-eu.bsky.social · 01/10/2026

🧠 What should AI remember — and what should it forget? At NeurIPS 2026, ELLIOT is supporting PALM, a workshop on personalised, aligned and long-term AI memory, organised by researchers including our CISPA partners. 📅 13 Dec 📍 Paris 🇫🇷

001

ELLIOT Project @elliot-eu.bsky.social · 29/09/2026

Can a tabular foundation model help uncover what causes what? 🔎 In this ELLIOT Research Spotlight 🔬, we highlight “Does TabPFN Understand Causal Structures?”, a paper from the University of Freiburg and Prior Labs on causal information in TabPFN embeddings. ➕ Info in the comments.

100

ELLIOT Project @elliot-eu.bsky.social · 28/09/2026

AI can help write code, but it can also reproduce hidden bugs and security risks 🧑‍💻🔐 In this use case video, Christoph Landolt (CISPA) and Xuecong “Sabrina” Liu (OPENCHIP) discuss their work on safer AI-generated code and multimodal support for hardware design. Full video 👉 youtu.be/UUANvrIDGTs

youtu.be

Making AI-generated software and hardware more trustworthy, by Christoph Landolt and Xuecong Liu

YouTube video by ELLIOT Project

010

ELLIOT Project @elliot-eu.bsky.social · 25/09/2026

To celebrate European Researchers' Night, our infographic is taking centre stage on the European Corner webpage! Check out our new infographic featured in the European Corner to see how ELLIOT blends text, video & sensor data for real-world impact!🚀 Explore it 👇 cms.elliot-ai.eu/assets/promo...

000

ELLIOT Project @elliot-eu.bsky.social · 24/09/2026

Trustworthy AI isn't just a slogan for ELLIOT, it's a core workstream! 🧠🛡️ From EU AI Act compliance and bias mitigation to data privacy, explainability, and security, we're building safe European AI by design. Learn more 👇 www.elliot-ai.eu

020

ELLIOT Project @elliot-eu.bsky.social · 23/09/2026

What if video AI could move between different types of information? 🎥 In the final Poster Session video, Muhammad Khattak (EPFL) presents an any-to-any video model designed to take different video-related inputs and predict other modalities across space and time. Watch it 👀 youtu.be/mj7eJrHAInw

youtu.be

Any-to-Any Video Modeling in Multimodal Abstract Spaces by Muhammad Uzair Khattak (EPFL)

YouTube video by ELLIOT Project

020

ELLIOT Project @elliot-eu.bsky.social · 22/09/2026

We were in Barcelona for the Summer School on Autonomous Driving! 🚗 The event explored how AI can support safer, smarter mobility in complex real-world environments. It was great to see so much curiosity, technical exchange and energy in Barcelona. 🤝

042

ELLIOT Project @elliot-eu.bsky.social · 21/09/2026

🇪🇺 How can Europe strengthen frontier AI? On 8 Oct, ELLIOT will host “Training Frontier AI Models in Europe” at the German AI Centers & ELIAS Alliance event, , exploring trustworthy foundation models. 📍 Messe Stuttgart, Germany 📅 8 October 2026 🔗 More details: elliot-ai.eu/events/all-h...

011

ELLIOT Project @elliot-eu.bsky.social · 21/09/2026

When AI explains a decision, is the explanation useful to people — or only to the model? 🔎 In this ELLIOT Research Spotlight 🔬, we highlight PASTA, a human-aligned benchmark for evaluating XAI explanations in computer vision from University of Trento and other partners. Paper in comments 👇

100

ELLIOT Project @elliot-eu.bsky.social · 18/09/2026

Thinking about taking your ELLIOT research somewhere new? 🌍 In this short video, Damar Hoogland explains who can join the ELLIOT Mobility Programme, how to register and what happens after a mobility stay. Watch and explore how research collaboration can travel abroad ✈️: youtu.be/lpRgqNMX79w

youtu.be

How to join ELLIOT’s Mobility Programme, by Damar Hoogland

YouTube video by ELLIOT Project

001

ELLIOT Project @elliot-eu.bsky.social · 17/09/2026

ELLIOT was at #PlanetaAI2026 in Prague! 🤖 Karla Štěpánová from CTU presented how AI can help robots better understand people and support more personalised robotics. Great to see our research being shared with the wider AI community in Prague! 👏

000

ELLIOT Project @elliot-eu.bsky.social · 17/09/2026

How can vision models adapt to different compute budgets? ⚙️ In this ELLIOT Research Spotlight 🔬, we highlight SnapViT, a method from the University of Amsterdam, University of Technology Nuremberg & NVIDIA that makes pretrained Vision Transformers elastic without retraining. Paper in comments 👇

111

ELLIOT Project @elliot-eu.bsky.social · 16/09/2026

What makes an AI image edit truly convincing? 🖼️ In this video, Akis Papadopoulos (CERTH) presents SAGI, a method that improves image inpainting through foundation models and uncertainty-guided evaluation, while also helping assess manipulated content. Watch now 🎬 youtu.be/W0vatEf4bjE

youtu.be

SAGI: Semantically Aligned and Uncertainty Guided AI Image Inpainting by Akis Papadopoulos (CERTH)

YouTube video by ELLIOT Project

000

ELLIOT Project @elliot-eu.bsky.social · 10/09/2026

How can image generation become faster at test time? 🧩 In our latest Poster Session video, Zhitong Gao (EPFL) presents a method using 1D ordered tokens to support +efficient search in autoregressive image generation and open new possibilities for text-to-image. Watch now 🎬 youtu.be/GdynY98X20E

youtu.be

(1D) Ordered Tokens Enable Efficient Test-Time Search by Zhitong Gao (EPFL)

YouTube video by ELLIOT Project

000

ELLIOT Project @elliot-eu.bsky.social · 09/09/2026

ELLIOT is at ECCV 2026! 🤖🌍 Our community is joining one of the leading events in computer vision and AI research. A great opportunity to exchange ideas, connect with researchers, and explore the latest advances shaping the future of multimodal AI 🤝 Stay tuned for more highlights from Malmö! 🚀

020

ELLIOT Project @elliot-eu.bsky.social · 09/09/2026

How do we know what an AI model can really do? 🔎 In this testimonial, Joaquin Vanschoren (@tue.nl) discusses ELLIOT’s work on open, transparent evaluation, from safety tests to multimodal benchmarks that combine vision, text, sound, planning and action. 📽️ Watch it: youtu.be/T_3ml_AU5Fc

youtu.be

Making AI evaluation more realistic, by Joaquin Vanschoren

YouTube video by ELLIOT Project

000

ELLIOT Project @elliot-eu.bsky.social · 08/09/2026

Meet our partner Letxbe AI! 📄✨ They specialise in document intelligence & sovereign AI agents for complex business workflows. In ELLIOT, they adapt our foundation models for real-world document workflows. 🧠 www.elliot-ai.eu/consortium/l...

000

ELLIOT Project @elliot-eu.bsky.social · 07/09/2026

We’re in Naples for the 6th Workshop on Bias and Fairness in AI ⚖️ A space to ask what matters: how do we recognise bias, evaluate fairness and build AI people can trust? We’re looking forward to the discussions, insights and perspectives emerging from this conversation 🤝

000

ELLIOT Project @elliot-eu.bsky.social · 07/09/2026

Do videos always need to be modelled in full detail? 🎥 In this ELLIOT Research Spotlight 🔬, we highlight VideoFlexTok from EPFL and Apple, a flexible video tokenisation approach where early tokens capture semantics and motion, while later tokens add finer visual details. Paper in comments 👇

110

ELLIOT Project @elliot-eu.bsky.social · 04/09/2026

Interested in multimodal diffusion-based LLMs? 🧠✨ Join AImageLab at UNIMORE in Modena, Italy, for a 16-week research stay! Work with multi-GPU HPC tools to co-author cutting-edge AI research. MSc & PhD students can apply now 👇 www.elliot-ai.eu/mobility-pro...

000

ELLIOT Project @elliot-eu.bsky.social · 03/09/2026

How can multimodal AI move from research to real-world impact? 🧠 In this testimonial, Nicu Sebe (University of Trento – UNITN) shares his perspective on ELLIOT, collaboration and the future of open, trustworthy AI in Europe. 🎬Watch it now: www.youtube.com/watch?v=weZR...

youtube.com

Connecting technical, legal and societal perspectives, by Nicu Sebe

YouTube video by ELLIOT Project

000

ELLIOT Project @elliot-eu.bsky.social · 01/09/2026

👋 Hello from Atlanta! ELLIOT is at the ACM AI Leadership Summit We are joining conversations on where AI is heading and how to shape it responsibly. If you are around, come say hi!

000

ELLIOT Project @elliot-eu.bsky.social · 01/09/2026

Some visual questions need more than image recognition. 👀 In our latest Poster Session video, Alberto Compagnoni and Marco Morini (UNIMORE) present ReAG, a method combining image understanding, knowledge retrieval and reasoning to produce more accurate answers. Watch now 🎬

000

ELLIOT Project @elliot-eu.bsky.social · 31/08/2026

Ever wondered how general AI becomes a specialist tool? 🧠✨ Pre-training gives AI broad general knowledge, while fine-tuning adapts it to specific jobs, like robotics or climate modelling! Learn how ELLIOT uses both to build trustworthy European AI 👇 www.elliot-ai.eu

000

ELLIOT Project @elliot-eu.bsky.social · 28/08/2026

How can AI help us read the planet more clearly? 🌍 In this ELLIOT video, Oscar Pellicer and Amir Pasha Mozaffari discuss Earth observation, climate modelling, satellite data and the challenge of making long-term predictions more reliable. Watch the full video now 🎬

youtube.com

Using multimodal AI to better understand the Earth, by Oscar Pellicer and Amir Pasha Mozaffari

YouTube video by ELLIOT Project

000

ELLIOT Project @elliot-eu.bsky.social · 27/08/2026

How can we make prompt optimisation easier to use and reproduce? 📋 In this ELLIOT Research Spotlight 🔬, we highlight promptolution: an open-source, modular framework that brings prompt optimisation tools together for researchers and practitioners working with LLMs. Paper in the comments 🔗

100

ELLIOT Project @elliot-eu.bsky.social · 25/08/2026

How do we know if an AI image edit really followed the instruction? 🖼️ In our latest Poster Session video, Davide Bucciarelli (UNIMORE) presents research on evaluating image edits more precisely by checking what changed and whether each change matches the user’s request. 📹 Watch the video.👇

youtube.com

What Changed? Evaluating Guided Image Edits with Multimodal LLMs by Davide Bucciarelli (UNIMORE)

YouTube video by ELLIOT Project

000

ELLIOT Project @elliot-eu.bsky.social · 21/08/2026

Next stop: Barcelona! 🇪🇸✨ Join our research stay at CVC-CERCA to fine-tune visual & language foundation models for complex document understanding using top-tier HPC tools. Ready to help shape open European AI? Apply now! 👇 www.elliot-ai.eu/mobility-pro...

010

ELLIOT Project @elliot-eu.bsky.social · 20/08/2026

📣 ELLIOT research is heading to the ACM AI Leadership Summit in Atlanta! At the end of August, ELLIOT researcher Aditya Gulati will present the paper “Earned, Not Manufactured: The Regulatory Gap in AI ‘Trust’ that Behavioral Science Can Close”. 📰 ➕ Info in the comments.

100

ELLIOT Project @elliot-eu.bsky.social · 19/08/2026

How can AI image edits become more realistic and easier to evaluate? 🖼️ In this ELLIOT Research Spotlight 🔬, we highlight SAGI, a pipeline for semantically aligned, uncertainty-guided image inpainting, plus SAGI-D, a large dataset for AI-generated inpainting detection. Paper in the comments 🔗

100

ELLIOT Project @elliot-eu.bsky.social · 18/08/2026

How do real-world needs become technical AI requirements? 🧩 now.In this ELLIOT WP leader interview, Dimosthenis Karatzas explains how end-user needs are shaping the project’s use cases & connecting research outcomes with practical applications. Dive in now 🎬

youtube.com

Connecting user needs with AI innovation, by Dimosthenis Karatzas

YouTube video by ELLIOT Project

000

ELLIOT Project @elliot-eu.bsky.social · 17/08/2026

How can AI find the right information in documents that mix text and images? 📄🖼️ In our latest Poster Session video, Davide Caffagni (UNIMORE) presents ReT, a method for multimodal document retrieval that helps AI connect visual and textual context more effectively. Watch the video 👇

youtube.com

Recurrence Meets Transformers for Universal Multimodal Retrieval by Davide Caffagni (UNIMORE)

YouTube video by ELLIOT Project

020

ELLIOT Project @elliot-eu.bsky.social · 14/08/2026

We might be in summer mode, but our updates certainly aren't! ☀️🤖 Behind the scenes, our consortium is busy putting together the very first official ELLIOT Newsletter, arriving this September. Subscribe now so you don’t miss it: www.elliot-ai.eu/newsletters

010

ELLIOT Project @elliot-eu.bsky.social · 12/08/2026

On International Youth Day, we spotlight Artemis Llabrés, one of the young researchers contributing to ELLIOT. 🌱 In this short video, she shares what daily AI research looks like, from experiments to collaboration, & why understanding model blind spots matters. 🔗 Watch the full video 👇

youtu.be

Explaining the daily work in ELLIOT, by Artemis Llabrés

YouTube video by ELLIOT Project

000

ELLIOT Project @elliot-eu.bsky.social · 11/08/2026

How can language AI stay efficient as sequences get longer? ⚙️ In our latest Poster Session video, Lukas Hauzenberger (NXAI) presents an approach that transfers transformer-based models into hybrid xLSTM models, reducing memory needs while keeping strong performance. 🔗 Watch the video 👇

youtube.com

Effective Distillation to Hybrid xLSTM Architectures by Lukas Hauzenberger (NXAI)

YouTube video by ELLIOT Project

000

ELLIOT Project @elliot-eu.bsky.social · 07/08/2026

One year into ELLIOT, what has already started to take shape? 👀 In this short interview, Project Coordinator Dr Yiannis Kompatsiaris (CERTH-ITI) reflects on first-year progress, from multimodal models and evaluation frameworks to 12 real-world use cases. Watch now 🎬

youtu.be

Progress towards open Multimodal AI, by Yiannis Kompatsiaris

YouTube video by ELLIOT Project

020

ELLIOT Project @elliot-eu.bsky.social · 06/08/2026

How can AI help design next-gen microchips? ⚙️🤖 ELLIOT is proud to support the AI for Chip Design Workshop at NeurIPS 2026 in Paris (12 Dec), initiated by the Barcelona Supercomputing Center! 📢 Call for Papers open until 30 Aug! Know more 👇 elliot-ai.eu/events/ai-fo...

020

ELLIOT Project @elliot-eu.bsky.social · 05/08/2026

Can text-to-image models count what they create? 🖼️🔢 In this ELLIOT Research Spotlight 🔬, we highlight QUOTA, an optimisation framework that helps AI generate the right number of objects across unseen domains, without retraining for each one. Paper + code in the comments 🔗

100

ELLIOT Project @elliot-eu.bsky.social · 03/08/2026

How can AI vision work better in low light or changing environments? 📷 In this video, Niccolò Biondi (UNITN) presents PEPR, a method that combines RGB and event-based camera data to improve robustness in real-world conditions. 👀 🔗 Watch the video: www.youtube.com/watch?v=AxVP...

youtube.com

PEPR: Event-Based Predictive Regularization for Domain Generalization by Niccolò Biondi (UNITN)

YouTube video by ELLIOT Project

030

ELLIOT Project @elliot-eu.bsky.social · 31/07/2026

Food recognition is more than naming a dish. 🍽️ In this interview, Eric Verdaguer, Co-Founder and CEO of LogMeal, explains how AI can identify ingredients, estimate quantities and support professionals in healthcare, sports, wellness and food services. Watch now. 🎬

000

ELLIOT Project @elliot-eu.bsky.social · 30/07/2026

Our homepage has a fresh new look! 🌐 Explore our new “Latest Updates” section, featuring project news, events, and scientific outputs (publications, datasets, software & models). Tracking open science has never been easier! Explore our website here: www.elliot-ai.eu

000

ELLIOT Project @elliot-eu.bsky.social · 29/07/2026

How can AI combine specialised models without losing what each one does best? 🧠 In this ELLIOT workshop interview, Joost van de Weijer (CVC-CERCA) explains model merging and how it can support collaboration when data cannot be shared in one place. Watch now. 🎬

000

ELLIOT Project @elliot-eu.bsky.social · 27/07/2026

ELLIOT Research Spotlight 🔬 Huge congratulations to the EPFL for their forward-looking work on Test-Space Training! This brings brilliant localised intelligence to AI via self-supervision. Pushing boundaries for European open science! 🚀🤝 Explore the results and overview video: tst-vision.epfl.ch

011

ELLIOT Project @elliot-eu.bsky.social · 24/07/2026

How can AI learn everyday actions from a person’s point of view? 👀 In this ELLIOT interview, @dimadamen.bsky.social (University of Bristol) explains why first-person vision matters and why translating human actions into language remains a major challenge for multimodal AI. Link in comments 👇

141

ELLIOT Project @elliot-eu.bsky.social · 23/07/2026

How can AI better understand 3D driving scenes? 🚗 In this ELLIOT Research Spotlight 🔬, we highlight LOSC from @valeoai.bsky.social: a method for open-vocabulary LiDAR segmentation that helps assign meaning to 3D point clouds, beyond fixed object categories. Paper + code in comments 👇

122

ELLIOT Project @elliot-eu.bsky.social · 22/07/2026

Finding the right words to talk to AI can be tough! 🧠 This is why Tom Zehle from @ellisinsttue.bsky.social created Promptolution, a modular framework that automates prompt optimisation. Watch the video below to discover how it easily boosts LLM performance! 🚀🎬

000

ELLIOT Project @elliot-eu.bsky.social · 21/07/2026

AI systems that understand human behaviour can support new digital assistants, but they also raise risks around manipulation and impersonation 🧠. In this ELLIOT workshop interview, Ioannis Patras (Queen Mary University of London & Tavus) explores safe and responsible use of AI. Watch now 👀

000