Lightnews — Scholar-powered news

Martin Mundt @martinmundt.bsky.social · 2d

Wonder how LLMs learn over long time horizons & how hate-checks deal with time?

Look at our new work "Chronoberg": an open-source dataset spanning 250 years of books with analysis of shifts in meaning & continual learning of LLMs:

arxiv.org/pdf/2509.22360

huggingface.co/datasets/spa...

1 4

Martin Mundt @martinmundt.bsky.social · 22d

🎉"Aligning generalization between humans and machines" (w/ 25 incredible authors) is out now in #Nature Machine Intelligence: www.nature.com/articles/s42...

In short, we identified interdisciplinary commonalities & differences with respect to notions of, methods for, & evaluation of generalization

5

Martin Mundt @martinmundt.bsky.social · Aug 19

Feeling really inspired after @collasconf.bsky.social: wonderful keynotes, exceptionally talented early-career spotlights & interactions that mattered!
And of course our own oral presentation as well :)

Very excited to be the next program chair & ensure its success for CoLLAs in Romania next year!

2

Martin Mundt @martinmundt.bsky.social · Aug 11

Just arrived in Philadelphia for @collasconf.bsky.social

Very excited about a fantastic program!

I’ll be around all week, but feel free to drop me a message or catch me directly at the debate on Monday morning, after our oral presentation on Tuesday, or in any of the poster sessions

3

Reposted by Martin Mundt

collasconf.bsky.social @collasconf.bsky.social · Aug 5

🎉 Announcing the winners of the #CoLLAs2025 Early-Career Spotlight Program!

These rising stars are pushing the boundaries of lifelong & continual learning, adaptation, and embodied AI.

👇 A thread introducing their work & talks.

#AI #MachineLearning #ContinualLearning #LifelongLearning

1 4 4

Reposted by Martin Mundt

collasconf.bsky.social @collasconf.bsky.social · Jul 31

🧠 Don't miss the tutorials at #CoLLAs2025!

We’re excited to host two deep dives into the theory and practice of lifelong learning:

📍Information-Theoretic Measures for Multi-Expert FM Adaptation

📍Mathematics of Continual Learning

Details below 👇

1 3 2

Martin Mundt @martinmundt.bsky.social · Jul 15

I'm not at #ICML2025 myself, but @florianbusch.bsky.social & Roshni are presenting our spotlight poster on continual confounding + our respective ConCon dataset on Thursday 17 Jul 4:30 pm - 7 pm PDT in East Exhibition Hall A-B: E-3309 icml.cc/virtual/2025...

Please talk to them! :)

Details below ⬇️

2 8

Reposted by Martin Mundt

Queer in AI @queerinai.com · Jul 1

🏳️‍🌈 Queer in AI is thrilled to announce another season of our affinity workshop at #NeurIPS2025! We announce a Call for Contributions to the workshop, with visa-friendly submissions due by 📅 July 31, 2025 (AOE), all other submissions due by 📅 August 14, 2025. #QueerInAI #CallForPapers

1 4 9

Reposted by Martin Mundt

ACM FAccT @facct.bsky.social · Jun 30

We're happy to officially announce the location of #FAccT2026!

Next year's conference will be held in Montreal, Canada 🇨🇦

Su Lin Blodgett and Zeerak Talat will be General Chairs, and Michael Madaio will be PC Chair 🎉

(thanks to MindView for the photo!)

Photo of the Town Hall meeting at last week's conference. The audience watches as Su Lin Blodgett announces the location of next year's conference. Su Lin Blodgett and Zeerak Talat will be General Chairs, and Michael Madaio will be PC Chair.

16 57

Reposted by Martin Mundt

Dr Abeba Birhane @abeba.bsky.social · Jun 29

dear computer vision people,

labelling someone* an "activist" is not the disparage you think it is

(*there is no claim in our paper that's not supported by empirical evidence so our work adheres to science by any standard. still...)

3 7 66

Reposted by Martin Mundt

collasconf.bsky.social @collasconf.bsky.social · Jun 28

📢 Present your published work at #CoLLAs2025!

Submit recent papers (JMLR, TMLR, NeurIPS, etc.) on lifelong/continual learning, LLMs, online learning & more.

🗓 Deadline: July 5
🔗 lifelong-ml.cc/Conferences/...

#AI #MachineLearning #LifelongLearning #ContinualLearning

CoLLAs

Collas

lifelong-ml.cc

2 3

Martin Mundt @martinmundt.bsky.social · Jun 6

We are hiring a tenure-track prof in the computer science department at University of Bremen for the field of Computer Engineering with a focus on Distributed Systems

Applications are still open until June 18th, so come join a fantastic set of colleagues!

www.uni-bremen.de/en/universit...

Job Vacancies - Universität Bremen

Offene Stellen

www.uni-bremen.de

2 4

Reposted by Martin Mundt

ACM FAccT @facct.bsky.social · Jun 1

Are you based in Greece? 🇬🇷 Do you want to attend #FAccT2025 and have your entire registration fee waived?

Volunteer for just 4 hours at the conference!

Apply by June 3, 2025, at 11:59 pm anywhere on Earth: facctconference.org/2025/cfv2

1 3 6

Martin Mundt @martinmundt.bsky.social · May 28

We (+ @pranav-nlp.bsky.social) had tremendous fun presenting our cake-AI analogy to a really engaging audience. Thanks for the greatly thoughtful discussions 🙏

We are facing numerous socio-technical challenges in AI, but with talented fresh minds & increased awareness, there may be a bright future

Data Train @datatrain.bsky.social · May 27

@martinmundt.bsky.social SERVING 🍰 cake 🍰 in his Data Story @dfki.bsky.social today 🙌

6

Reposted by Martin Mundt

Pranav A @pranav-nlp.bsky.social · May 27

I am at the data story workshop hosted by @martinmundt.bsky.social where we talk about AI as cake! We redefine Yann LeCun's famous analogy beyond just learning types. From ingredients to selling - we break down how AI really works and who gets left out.

Martin presenting the intro slide of the talk

1 1 3

Martin Mundt @martinmundt.bsky.social · May 23

"Reasonable AI" got selected as a cluster of excellence www.tu-darmstadt.de/universitaet...

Overwhelmingly happy to be part of RAI & continue working with the smart minds at TU Darmstadt & hessian.AI, while also seeing my new home at Uni Bremen achieve a historic success in the excellence strategy!

1 3 8

Martin Mundt @martinmundt.bsky.social · May 14

You may know batch-norm, but did you know it is a lightweight catalyst for open world learning?

In "BOWL: A Deceptively Simple Open World Learner" we leverage BN stats to enable OoD detection & rapid active + continual learning!

arxiv.org/abs/2402.04814

🚀 now accepted @collasconf.bsky.social

5

Reposted by Martin Mundt

Hanna Wallach @hannawallach.bsky.social · May 3

Exciting news!!! This just got into @icmlconf.bsky.social as a position paper!!! 🎉 More updates to come as we work on the camera-ready version!!!

Hanna Wallach @hannawallach.bsky.social · Feb 4

Remember this @neuripsconf.bsky.social workshop paper? We spent the past month writing a newer, better, longer version!!! You can find it online here: arxiv.org/abs/2502.00561

11 51

Martin Mundt @martinmundt.bsky.social · May 8

Horizontal, vertical, hybrid data partitioning: heterogeneity is tough to handle in federated learning!

🔥In “Scaling Probabilistic Circuits via Data Partitioning" - accepted at #UAI25 - we unify the different settings through aggregation of learned client distributions: arxiv.org/abs/2503.08141

4 7

Martin Mundt @martinmundt.bsky.social · May 4

We have 2 variants of the dataset: 1. Confounders only ever appear disjointly
2. Confounders can reappear later in other “tasks”

We did observe that in disjoint scenarios, solutions like “blue + large + metallic” (concatenation of confounders) happen.

Did you have something like this in mind?

1 1

Reposted by Martin Mundt

ContinualAI @continualai.bsky.social · May 2

Join us this Thursday 8 May for:

U n l e a r n i n g
~ in continual scenarios ~

together with:
Zhenyi Wang, Sijia Liu, and Kairan Zhao

6 PM CET / 9 AM PT / 12 PM ET

Hope to see you there!!!

Registration here: stanford.zoom.us/meeting/regi...

2 4

Martin Mundt @martinmundt.bsky.social · May 2

We believe ConCon is crucial for future continual learning studies. Why?

In the presence of confounders, we show that models succeed when training on all data, but sequential training on accumulated tasks fails without further action!

This questions the common CL "upper-bound" with infinite memory

1

Martin Mundt @martinmundt.bsky.social · May 2

🔥Our work “Where is the Truth? The Risk of Getting Confounded in a Continual World" was accepted with a spotlight poster at ICML!
arxiv.org/abs/2402.06434

-> we introduce continual confounding + the ConCon dataset, where confounders over time render continual knowledge accumulation insufficient ⬇️

2 3 10

Reposted by Martin Mundt

Data Train @datatrain.bsky.social · Apr 30

Join us for a delicious Data Story by @martinmundt.bsky.social on the AI-cake analogy, hosted @dfki.bsky.social in Bremen!

👉 Info and registration: www.bremen-research.de/data-train/c...

👓 Read more about the research in Martin's thread below

🍰 Of course, there will be cake 😋

2 2

Martin Mundt @martinmundt.bsky.social · Apr 30

We are hosting a 3-hour interactive session on our Cake-AI analogy work (arxiv.org/abs/2502.03038) on May 27th @dfki.bsky.social Bremen.

If you are around, please consider signing up to learn & discuss how the process of baking helps us understand design & socio-technical tensions of AI systems!⬇️

Data Train @datatrain.bsky.social · Apr 30

Join us for a delicious Data Story by @martinmundt.bsky.social on the AI-cake analogy, hosted @dfki.bsky.social in Bremen!

👉 Info and registration: www.bremen-research.de/data-train/c...

👓 Read more about the research in Martin's thread below

🍰 Of course, there will be cake 😋

2 4