Lightnews — Scholar-powered news

Reposted by Lorenz Linhardt

Laure Ciernik @lciernik.bsky.social · Jul 16

🎉 Presenting at #ICML2025 tomorrow!
Come and explore how representational similarities behave across datasets :)

📅 Thu Jul 17, 11 AM-1:30 PM PDT
📍 East Exhibition Hall A-B #E-2510

Huge thanks to @lorenzlinhardt.bsky.social, Marco Morik, Jonas Dippel, Simon Kornblith, and @lukasmut.bsky.social!

3 9

Reposted by Lorenz Linhardt

BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · Jun 16

Join us - we have four open positions for doctoral/postdoctoral researchers at BIFOLD, doing cutting edge research in data management and machine learning as well as their intersections. More information: www.bifold.berlin/about-us/opp...

5 5

Reposted by Lorenz Linhardt

Laure Ciernik @lciernik.bsky.social · Jun 6

I am deeply grateful to @lorenzlinhardt.bsky.social, Marco Morik, Jonas Dippel, Simon Kornblith, and @lukasmut.bsky.social for their great work and support in this project! We also thank our collaborators, @bifold.berlin and HFA 7/7
📄Paper: arxiv.org/abs/2411.05561
💻Code: github.com/lciernik/sim...

Objective drives the consistency of representational similarity across datasets

The Platonic Representation Hypothesis claims that recent foundation models are converging to a shared representation space as a function of their downstream task performance, irrespective of the obje...

arxiv.org

3 6

Lorenz Linhardt @lorenzlinhardt.bsky.social · May 3

🖼️ At the Re-Align workshop, @tomneuhaeuser.bsky.social and I presented "Cat, Rat, Meow: On the Alignment of Language Model and Human Term-Similarity Judgments", joint work with Lenka Tětková and @eberleoliver.bsky.social .
📃 arxiv.org/abs/2504.07965

Cat, Rat, Meow: On the Alignment of Language Model and Human Term-Similarity Judgments

Small and mid-sized generative language models have gained increasing attention. Their size and availability make them amenable to being analyzed at a behavioral as well as a representational level, a...

arxiv.org

1 3

Lorenz Linhardt @lorenzlinhardt.bsky.social · May 3

🖼️ At the DeLTa workshop, Jonas Loos and I presented "Latent Diffusion U-Net Representations Contain Positional Embeddings and Anomalies".
📃 arxiv.org/abs/2504.07008

Latent Diffusion U-Net Representations Contain Positional Embeddings and Anomalies

Diffusion models have demonstrated remarkable capabilities in synthesizing realistic images, spurring interest in using their representations for various downstream tasks. To better understand the rob...

arxiv.org

1 1

Lorenz Linhardt @lorenzlinhardt.bsky.social · May 3

🎉 Excited to have had the opportunity to present two posters at the ICLR2025 workshops! 🖼️🖼️

A big thanks to my coauthors and to everyone who dropped by to discuss!

Also, thanks to the Re-Align and DeLTa organizers for hosting such an inspiring workshop day. ✨

@bifold.berlin @tuberlin.bsky.social

1 4

Reposted by Lorenz Linhardt

BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · Feb 3

CALL FOR PAPERS: #XAI2025, Special Track: Actionable explainable AI. Submit your paper and check the Submission deadlines: xaiworldconference.com/2025/importa...

Actionable Explainable AI
xaiworldconference.com/2025/actiona...

1 5

Lorenz Linhardt @lorenzlinhardt.bsky.social · Jan 13

Thanks to everyone for the lively discussions on concept convexity and alignment in DL models at #NLDL (Northern Lights Deep Learning conference)! ❄️❄️

Had a great time in beautiful Tromsø connecting with fellow researchers 🤝

📜 Feel free to check out our preprint: arxiv.org/abs/2409.06362

1 7

Lorenz Linhardt @lorenzlinhardt.bsky.social · Jan 13

Happy to co-chair this year's special track on "Actionable Explainable AI" with a great team! 📄🦾
Please consider submitting! (Abstract deadline: February 10th) ☝️

BIFOLD Berlin Institute for the Foundations of Learning and Data @bifold.berlin · Jan 9

CALL FOR PAPERS: #𝗫𝗔𝗜20𝟮𝟱, 𝗦𝗽𝗲𝗰𝗶𝗮𝗹 𝗧𝗿𝗮𝗰𝗸: Actionable explainable AI. Submit your paper until February 15, 2025.

xaiworldconference.com/2025/actiona...

#XAI #LRP #counterfactuals #shapley #models #deeplearning #interpretability #decisionmaking @lorenzlinhardt.bsky.social @tuberlin.bsky.social

Following the success of Explainable AI in generating faithful and understandable explanations of complex ML models, there has been increasing attention on how the outcomes of Explainable AI can be systematically used to enable meaningful actions. These considerations are studied within the subfield of Actionable XAI. In particular, research questions relevant to this subfield include (1) what types of explanations are most helpful in enabling human experts to achieve more efficient and accurate decision-making, (2) how one can systematically improve the robustness and generalization ability of ML models or align them with human decision making and norms based on human feedback on explanations, (3) how to enable meaningful actioning of real-world systems via interpretable ML-based digital twins, and (4) how to evaluate and improve the quality of actions derived from XAI in an objective and reproducible manner. This special track will address both the technical and practical aspects of Actionable XAI. This includes the question of how to build highly informative explanations that form the basis for actionability, aiming for solutions that are interoperable with existing explanation techniques such as Shapley values, LRP or counterfactuals, and existing ML models. This special track will also cover the exploration of real-world use cases where these actions lead to improved outcomes.

1 4

Reposted by Lorenz Linhardt

Lukas Muttenthaler @lukasmut.bsky.social · Dec 4

Excited that Re-Align will have its second iteration at ICLR 2025 in Singapore! More soon! 🧠🤖

Nezihe Merve Gürel @nmervegurel.bsky.social · Dec 3

The list of accepted workshops for ICLR 2025 is available at openreview.net/group?id=ICL...
@iclr-conf.bsky.social

We received 120 wonderful proposals, with 40 selected as workshops.

ICLR 2025 Workshop Proposals

Welcome to the OpenReview homepage for ICLR 2025 Workshop Proposals

openreview.net

2 19