Ansgar Scherp

An interesting new paper on the Distances Between Formal Concept Analysis Structures just published in @tgdkjournal.bsky.social See: drops.dagstuhl.de/entities/doc...

America Is Now a Dictatorship We Are There Now

I wonder how many conferences taking place this year in the US will be online only / in parts virtual because of what is going on there - Or to put it in MeidasTouch's Tennessee Brando's words, "America Is Now a Dictatorship", see substack.com/inbox/post/1...

It was recorded. Once I have the details, I will share them :-)

When Springer renames a journal from IR to "Discover Computing", it sucks! Now it looks like our article with @lukasgalke.bsky.social and Iacopo Vagliano appeared in there. I don't even know what "Discover Computing" is, nor does my AI-companion.

Stop doing this. The journal impact factor is now 0.

Discrete Key-Value Bottleneck Deep neural networks perform well on classification tasks where data streams are i.i.d. and labeled data is abundant. Challenges emerge with non-stationary training data streams such as continual lear...

🤖 Classic ML Still Matters: I make my case

🔍 Discrete Key-Value Bottleneck
arxiv.org/abs/2207.11240
➡️ k-means for codebook—buried in refs

🌲 CascadeXML
arxiv.org/abs/2211.00640
➡️ Transformer + hierarchical k-means label tree

🧠 Mirage
arxiv.org/abs/2310.09486
➡️ Graph distillation via pattern mining

Thank you, we still have sane people in the US!
@meidastouch.com

Now it is time for US research institutions to stand up! @columbiauniversity.bsky.social you can do better!

Make Ron happy!

In the US, classes are back online — not due to a pandemic, but because students fear being snatched off the street and deported 😟📚

Meanwhile, in Germany, some still act like it’s a game

To students & scholars: you’re welcome here.
To Europe’s academics: wake up. This is not a drill. ⏰🌍

#50501

The AI that DOGE allegedly uses is obviously a simple list of regular expressions (aka a Gazetteer approach) and aims to delete all web content, fire all people, etc., that match on trans*, gay*, dei * ...

see also the "Enola Gay" story, www.youtube.com/watch?v=CQ90...

📢 A major tech-related legal case is back in the spotlight! A well-documented scandal in the British Post Office—caused by programming errors—led to serious miscarriages of justice.
Great to see *Comm. of the ACM* covering this important issue in their latest edition.
cacm.acm.org/news/how-sof...

YouTube video by Scientists for EU American Scientists - come to Europe.

American scientists, come to Europe! 🌍✈️ We all speak English, we have excellent affordable health care 🏥💊, safe schools 🏫🔒, and free university education for your kids 🎓👨‍👩‍👧‍👦. Enjoy more money after deduction of taxes and living expenses 💰📉. We care for you! ❤️🤝

www.youtube.com/watch?v=9_l0...

Isotropy Matters: Soft-ZCA Whitening of Embeddings for Semantic Code Search Low isotropy in an embedding space impairs performance on tasks involving semantic inference. Our study investigates the impact of isotropy on semantic code search performance and explores post-proces...

🗞️ A simple trick improves embedding retrieval performance even without further training.

ZCA whitening increases isotropy of the embedding space and thereby helps retrieval

Paper by Andor Diera and with @lukasgalke.bsky.social at ESANN 2025.

Preprint: arxiv.org/abs/2411.17538

GitHub - ascherp/text-classification-survey: Companion website to a long-running survey on text classification. Companion website to a long-running survey on text classification. - ascherp/text-classification-survey

🚀 Text News Alert! 🚀

Our long-running text classification survey just got a fresh new home—and an epic boost! 🎉 /w @lukasgalke.bsky.social

* Added hierarchical classification (HTC) 📚

* Latest research papers on leveraging LLMs 🤖

👉 Explore More: github.com/ascherp/text...

When fact's don't matter anymore, you end up with Elon and Donald. -

German public news authority @tagesschau.de.web.brid.gy's front page calls the Trump government by the right words: dictatorship and coup.

As a researcher, I need unfiltered facts.

I deleted the @cnn.com app weeks ago.

Tune in to Prof @meiselasb.bsky.social and @meidastouch.com

Src:

Mark Zuckerberg Is a Surrender Monkey Capitulation is contagious.

After people have left X, they will also leave Facebook after this move. Luckily, not many researchers are actively pushing research content on FB anyway. So I won't care :-) @thebulwark.bsky.social

www.thebulwark.com/p/mark-zucke...

🌟 Reflecting on the Semantic Web 🌟 With Katja Hose, Maria-Esther Vidal, Gerd Groener, and Petr Škoda, we wrote "Semantic Web: Past, Present, and Future", published in TGDK. A journey that started 13 years ago! 👉 Primer: drops.dagstuhl.de/entities/doc... ✨ Interested in contributing? Let’s chat!

Screenshot of table of contents in Volume 2, Issue 2.

Screenshot of table of contents in Volume 2, Issue 3.

Transactions on Graph Data & Knowledge (TGDK), Vol. 2, Issues 2 & 3 now available, featuring 8 resource articles, and 2 research articles! All #DiamondOA (no fees for authors/readers) w/ Dagstuhl Publishing.

Enjoy!

2(2): drops.dagstuhl.de/entities/iss...
2(3): drops.dagstuhl.de/entities/iss...

Continual Learning for Encoder-only Language Models via a Discrete Key-Value Bottleneck Continual learning remains challenging across various natural language understanding tasks. When models are updated with new training data, they risk catastrophic forgetting of prior knowledge. In the...

Preventing catastrophic forgetting in NLP! 🌟 Our discrete key-value bottleneck enables efficient continual learning in encoder-only language models—no major updates, just localized tweaks. With Andor Diera and @lukasgalke.bsky.social Learn more! 🚀 arxiv.org/abs/2412.08528

@matthiaskohn.bsky.social Thanks for the great work during your BSc thesis and Project Data Science!

Edge-Splitting MLP: Node Classification on Homophilic and... Message Passing Neural Networks (MPNNs) have demonstrated remarkable success in node classification on homophilic graphs. It has been shown that they do not solely rely on homophily but on...

Standard GNNs excel on homophilic graphs but depend on neighborhood patterns. ES-MLP, a student project by Matthias Kohn and co-supervised with Marcel Hoffmann, presented at #log2024, combines Graph-MLP with edge-splitting for faster, robust, edge-free inference. openreview.net/forum?id=BQE...

Hierarchical Text Classification (HTC) vs. eXtreme Multilabel Classification (XML): Two Sides of the Same Medal Assigning a subset of labels from a fixed pool of labels to a given input text is a text classification problem with many real-world applications, such as in recommender systems. Two separate research...

Hierarchical Text Classification (HTC) vs. eXtreme Multilabel Classification (XML) are two research areas that have been disconnected so far. A team of students and with @lukasgalke.bsky.social we show that it is crucial to cross the two worlds. Read more arxiv.org/abs/2411.13687

LossVal: Efficient Data Valuation for Neural Networks Assessing the importance of individual training samples is a key challenge in machine learning. Traditional approaches retrain models with and without specific samples, which is computationally expens...

LossVal, a new efficient method for data valuation for neural networks. It allows assessing the importance of individual training samples for different tasks and constraints. Find out more arxiv.org/abs/2412.04158

I grew up with Twitter, me and my academic life. We used it at conferences, saw it struggle financially & wondered if it’d survive. I never imagined a billionaire would buy it for $44B, overrun it, & turn it into a hate platform. Now, Twitter-based research is part of scientific history?

Hierarchical text classification does not need a graph encoder or label semantics. Work at the European Conference on Artificial Intelligence 2024 (ECAI) by Yousef Younes and with @lukasgalke.bsky.social - learn more: ebooks.iospress.nl/doi/10.3233/...

... but it would be no fun if I just left X without saying anything further. From now on, I will advertise all my messages on Mastodon, BlueSky, and LinkedIn with a link from X and encourage people to move on and turn the page. Meanwhile, my X account is labeled as (legacy account) 😛 Elon

I decided weeks ago to leave Twitter, prepared it, and now it is done. So many others have done so by now, too.

YouTube video by PBS NewsHour How Trump's rhetoric compares to historic fascist language

Understand how fascists are creating an "upside-down world" thanks to @pbsnews.bsky.social and history professor @ruthbenghiat.bsky.social in "How Trump's rhetoric compares to historic fascist language", see: www.youtube.com/watch?v=6U4N...

Hi!