←
Explainable Machine Learning401 posts
Explainable Machine Learning avatar

Explainable Machine Learning

@ExplainableML

Institute for Explainable Machine Learning @HelmholtzMunich and Interpretable and Reliable Machine Learning group @TU_Muenchen

◷ Joined August 2021▧ 157 Photos and videos
Julia Herbst avatar
⟳ Explainable Machine Learning retweeted
Julia Herbst@_juliaherbst·Nov 11

Do vision and language models, trained on entirely different data, converge toward a shared representation of the world? The Platonic Representation Hypothesis suggests they do. We take a closer look.

0:33
Kirill Bykov avatar
⟳ Explainable Machine Learning retweeted
Kirill Bykov@kirill_bykov·Oct 21

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🧵

0:40
Lukas Thede avatar
⟳ Explainable Machine Learning retweeted
Lukas Thede@lukas_thede·Oct 2

📣 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

Post attachment
Lukas Thede avatar
⟳ Explainable Machine Learning retweeted
Lukas Thede@lukas_thede·Oct 1

📣 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

Lukas Thede avatar
⟳ Explainable Machine Learning retweeted
Lukas Thede@lukas_thede·Sep 28

🎉 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

Post attachment
Luis avatar
⟳ Explainable Machine Learning retweeted
Luis@lusxvr·Sep 25

FineVision was just accepted to NeurIPS! Big thanks to all my co-authors, excited to present it at the conference

Post attachment
Luis@lusxvr4 Sep 2025

Today, we are releasing FineVision, a huge open-source dataset for training state-of-the-art Vision-Language Models: > 17.3M images > 24.3M samples > 88.9M turns > 9.5B answer tokens Here are my favourite findings:

Post attachment
Explainable Machine Learning avatar
Explainable Machine Learning@ExplainableML·Sep 7

On V*Bench and HR-Bench, VisLens matches or beats prior visual search baselines, running 8.5–9.9x faster than Thyme and up to 22.2x faster than training-free multi-pass search. Every crop traces back to decoded tokens, so you can see what the model looked at.

Post attachment
Explainable Machine Learning avatar
Explainable Machine Learning@ExplainableML·Sep 7

VisLens reads the model's own hidden states. A logit lens projects each image region through the LLM head to reveal what it already says; a light tuned-lens makes early layers readable. Match decoded tokens to the query, cluster, merge into crops, answer.

Post attachment
Explainable Machine Learning avatar
Explainable Machine Learning@ExplainableML·Sep 7

Finding a small object in a high-resolution image is still hard for MLLMs. Current fixes crop, zoom and re-query the model over and over: slow, and you cannot see why a region was picked. We wanted the answer in one pass, and we wanted to watch it happen.

Explainable Machine Learning avatar
Explainable Machine Learning@ExplainableML·Sep 7

The adapter is where segmentation quality drops: alignment costs spatial detail. LLM layers claw it back. Correct tokens act as anchors, pulling wrong neighbors to the right label. Causal masking starves early tokens of anchors; bidirectional attention fixes that.

Post attachment
Explainable Machine Learning avatar
Explainable Machine Learning@ExplainableML·Sep 7

Three probes. Linear probing at every layer to track segmentation quality. Attention knockout to test whether tokens repair each other. Bidirectional attention among image tokens to measure what causal masking costs.

Post attachment
Explainable Machine Learning avatar
Explainable Machine Learning@ExplainableML·Sep 7

MLLMs are being pushed into pixel-level work like segmentation, but nobody knows where in the stack that spatial ability lives, or where it breaks. So we opened the pipeline: vision encoder, adapter, and every single LLM layer.

Post attachment