LightNews LightNews
banner
eth-ai-center.bsky.social
@eth-ai-center.bsky.social
220 followers 37 following 11 posts
Welcome to ETH AI Center! We are ethz.ch/en 's central hub leading the way towards trustworthy, accessible and inclusive #artificialintelligence

ai.ethz.ch
Posts Replies Media Videos
Reposted
humerchr.bsky.social
Learn more about Generative Flow Networks in our new GFlowNet Playground - an interactive educational article to enhance intuitive understanding.

🎤 Presented at @visxai.bsky.social
👉 Explore the article here: gfn-playground.jku-vds-lab.at
🧑‍💻Flo, Alex, Andi, @alexhergar.bsky.social, Marc, and me :)
GFlowNet Playground – Building an Intuitive Understanding of GFlowNet Training
Explore the training behavior of GFlowNets, a model class that samples a diverse set of candidates in an active learning context.
gfn-playground.jku-vds-lab.at
November 10, 2025 at 2:33 PM
Reposted
alexanderhoyle.bsky.social
At #ACL2025 this week! Please reach out if you want to chat :)

We have two lovely posters:
Tues Session 2, 10:30-11:50 — Large Language Models Struggle to Describe the Haystack without Human Help

Wed Session 4 11:00-12:30 — ProxAnn: Use-Oriented Evaluations of Topic Models and Document Clustering
alexanderhoyle.bsky.social Alexander Hoyle @alexanderhoyle.bsky.social · Jul 8
Evaluating topic models (and document clustering methods) is hard. In fact, since our paper critiquing standard evaluation practices four years ago, there hasn't been a good replacement metric

That ends today (we hope)! Our new ACL paper introduces an LLM-based evaluation protocol 🧵
Screenshot of first page of paper. It is here: https://arxiv.org/pdf/2507.00828 Abstract: Topic model and document-clustering evaluations either use automated metrics that align poorly with human preferences or require expert labels that are intractable to scale. We design a scalable human evaluation protocol and a corresponding automated approximation that reflect practitioners' real-world usage of models. Annotators -- or an LLM-based proxy -- review text items assigned to a topic or cluster, infer a category for the group, then apply that category to other documents. Using this protocol, we collect extensive crowdworker annotations of outputs from a diverse set of topic models on two datasets. We then use these annotations to validate automated proxies, finding that the best LLM proxies are statistically indistinguishable from a human annotator and can therefore serve as a reasonable substitute in automated evaluations
July 27, 2025 at 9:30 AM
Reposted
alexanderhoyle.bsky.social
this looks terrific, very excited to read
emollick.bsky.social Ethan Mollick @emollick.bsky.social · Sep 15
LLMs introduce a huge range of new capabilities for research, but also make it possible for researchers to "hack" their results in new ways by how they chose to use models for annotation

This is a useful pass at quantifying some of the risk, and some mitigation strategies arxiv.org/pdf/2509.08825
September 15, 2025 at 7:53 PM
Reposted
alexanderhoyle.bsky.social
Accepted to EMNLP (and more to come 👀)! The camera ready version is now online---very happy with how this turned out

arxiv.org/abs/2507.01234
alexanderhoyle.bsky.social Alexander Hoyle @alexanderhoyle.bsky.social · Jul 17
New preprint! Have you ever tried to cluster text embeddings from different sources, but the clusters just reproduce the sources? Or attempted to retrieve similar documents across multiple languages, and even multilingual embeddings return items in the same language?

Turns out there's an easy fix🧵
Barchart of number of items in four clusters of text embeddings, with colors showing the distribution of sources in each cluster. Caption: Clustering text embeddings from disparate sources (here, U.S. congressional bill summaries and senators’ tweets) can produce clusters where one source dominates (Panel A). Using linear erasure to remove the source information produces more evenly balanced clusters that maintain semantic coherence (Panel B; sampled items relate to immigration). Four random clusters of k-means shown (k=25), trained on a combined 5,000 samples from each dataset
September 24, 2025 at 3:21 PM
Reposted
ghjuliasialelli.bsky.social
The @ai-env-summit.bsky.social is getting started 🚀 First speaker: the amazing @sarameghanbeery.bsky.social on species modeling 🦌
October 1, 2025 at 7:54 AM
eth-ai-center.bsky.social
Proud to be among the initiators & drivers of this important development for AI made in Switzerland!

& the pivotal contributions to its success by
A Ilic M Gabriel D Naeff D Dobos A Klimovic A Krause

Special congrats to the team around I Schlag, including E Ďurech & our fellows I Hakimi B Pásztor
ethz.ch ETH Zurich @ethz.ch · Sep 2
EPFL, ETH Zurich, and CSCS today released Apertus, Switzerland's first large-scale, multilingual language model (LLM). As a fully open LLM, it serves as a building block for developers and organizations to create their own applications.
ethz.ch/en/news-and-...
Apertus: a fully open, transparent, multilingual language model
EPFL, ETH Zurich and the Swiss National Supercomputing Centre (CSCS) released Apertus today, Switzerland’s first large-scale, open, multilingual language model — a milestone in generative AI for trans...
ethz.ch
September 2, 2025 at 2:18 PM
eth-ai-center.bsky.social
The Swiss AI Initiative 2nd Call for Large Grants is NOW OPEN! 🚀
Letter of Intent due: Aug 25, 2025
Full Proposal due: Sept 8, 2025
Full guidelines and submission details are available here: tinyurl.com/3utfdnsn
August 8, 2025 at 10:58 AM
Reposted
alizeepace.bsky.social
RL for real-world applications = offline learning + reward learning. How do we make this work?

Find out more at ICLR poster #377 at 10am today!

@gioramponi.bsky.social and I will be presenting our latest work on offline preference-based RL (joint w/ @gxxxr.bsky.social and Bernhard Schölkopf).
April 26, 2025 at 1:15 AM
Reposted
ghjuliasialelli.bsky.social
It was a pleasure to share my research with the Nordic research community! If you missed it, the recording will soon be online on RISE’s YouTube channel. In the meantime, check out the great many talks that came before: lnkd.in/dEnZpVAe and the upcoming @climateainordics.com workshop.
May 9, 2025 at 7:33 AM
eth-ai-center.bsky.social
🎉 Today, @ddmitriev.bsky.social, our second PhD Fellow of the @eth-ai-center.bsky.social, graduated! He was supervised by ETH AI Center Faculty members Prof. Fanny Yang and Prof. Afonso Bandeira. Congratulations, Dr. Daniil Dmitriev! Next, he will join @upenn.edu as a Postdoctoral Researcher.
May 16, 2025 at 8:04 PM
Reposted
vthoma.bsky.social
[1/2]Our work “Learning to steer Markovian Agents under Model Uncertainty” explores the problem of steering (unknown) learning dynamics in Markov Games towards desirable outcomes (e.g. Pareto optimal Nash).
My co-author Jiawei Huang is presenting it at ICLR on Apr 24, 3:00-5:30pm GMT+8, Poster #401.
April 22, 2025 at 12:24 PM
eth-ai-center.bsky.social
✨What an insightful academic talk & discussion today at ETH AI Center on Scientific Inference with Diffusion Models by
@stephanmandt.bsky.social (@ucirvine.bsky.social), hosted by our Faculty member Julia Vogt (@ethzurich.bsky.social)

More talks, info, & registration: ai.ethz.ch/research/eve...
April 22, 2025 at 1:15 PM
eth-ai-center.bsky.social
🪧 28 posters, 93 attendees, ∞ interactions. That sums up last week's Zurich Pre-ICLR 2025 poster session at ETH AI Center!✨

🌍 ETH AI Center together with its @ellis.eu Unit Zurich hosted an inspiring afternoon featuring posters from our talented academic community.
April 17, 2025 at 8:36 PM
eth-ai-center.bsky.social
☯ "AI at the Service of Humankind". Last week, we had an exciting spotlight academic session at ETH AI Center, including 2 engaging talks and an insightful panel with Prof. Paolo Benanti (The Vatican), Prof. em. Mario Rasetti (PoliTo), and ETH AI Center Faculty Prof. @arkrause.bsky.social

More 👇
April 11, 2025 at 4:42 PM
Reposted
arkrause.bsky.social
We've released our lecture notes for the course Probabilistic AI at ETH Zurich, covering uncertainty in ML and its importance for sequential decision making. Thanks a lot to @jonhue.bsky.social for his amazing effort and to everyone who contributed! We hope this resource is useful to you!
jonhue.bsky.social Jonas Hübotter @jonhue.bsky.social · Feb 11
I'm very excited to share notes on Probabilistic AI that I have been writing with @arkrause.bsky.social 🥳

arxiv.org/pdf/2502.05244

These notes aim to give a graduate-level introduction to probabilistic ML + sequential decision-making.
I'm super glad to be able to share them with all of you now!
February 17, 2025 at 7:20 AM
Reposted
alicebizeul.bsky.social
✨New Preprint ✨ Ever thought that reconstructing masked pixels for image representation learning seems sub-optimal?

In our new preprint, we show how masking principal components—rather than raw pixel patches— improves Masked Image Modelling (MIM).

Find out more below 🧵
March 19, 2025 at 8:44 PM
Reposted
haldaume3.bsky.social
Thanks for having me, it was great to visit!
March 13, 2025 at 1:44 PM
eth-ai-center.bsky.social
🧬👨‍🏫 We had an interesting academic talk yesterday at ETH AI Center on "Effective Antibody Engineering with Protein Language Models" by Dr. Daniel Danciu (Cradle Bio), hosted by our ETH AI Center Faculty Prof. Dr. Gunnar Rätsch.

🔥 More talks, info, and registration: ai.ethz.ch/research/eve...
March 19, 2025 at 7:57 PM
eth-ai-center.bsky.social
🤖🧑 We had an exciting academic talk today at ETH AI Center on "AI Interacting with People (through Language)" by Prof. Dr. @haldaume3.bsky.social (UMD), hosted by our ETH AI Center Postdoctoral Fellow Dr. @alexanderhoyle.bsky.social.

More talks, registration, & info: ai.ethz.ch/research/eve...
March 11, 2025 at 8:54 PM
eth-ai-center.bsky.social
🎉 Yesterday, @alizeepace.bsky.social, our first PhD Fellow of the @eth-ai-center.bsky.social, graduated! She was supervised by ETH AI Center Faculty members Prof. Rätsch @gxxxr.bsky.social and Prof. Schölkopf. Congrats, Dr. Pace! Next, she will join Google DeepMind in Zurich as a Research Scientist.
February 27, 2025 at 10:39 AM
Reposted
levilingsch.bsky.social
Still dreaming of a general analytical solution to Navier-Stokes? 😴 Keep dreaming… but maybe not for long!

Our latest work explores AI for symbolic solutions to PDEs—bringing us a step closer to solving fundamental equations in science.

Check it out!

#AI #MachineLearning #Science #PDEs #ETHZurich
Neuro-Symbolic AI for Analytical Solutions of Differential Equations
Analytical solutions of differential equations offer exact insights into fundamental behaviors of physical processes. Their application, however, is limited as finding these solutions is difficult. To...
arxiv.org
February 14, 2025 at 12:56 PM