xLSTM for PINNs that learn PDEs: arxiv.org/abs/2511.12512 “Across four PDEs under matched size and budget, xLSTM-PINN consistently reduces MSE, RMSE, MAE, and MaxAE with markedly narrower error bands.” “cleaner boundary transitions with attenuated high-frequency ripples”

xLSTM for Vehicle Trajectory Prediction: arxiv.org/abs/2511.00266 X-TRACK based on xLSTM achieves SOTA. “Compared to state-of-the-art baselines, X-TRACK achieves performance improvement by 79% at the 1-second prediction and 20% at the 5-second prediction in the case of highD” Again xLSTM excels.

On the flip side, when this current "data center" (more like large GPU clusters turned supercomputer) bubble bursts, they will hold the bum side of that trade and electricity might become very cheap ...

... totgesagte leben lange ...

... a less rose-tinted viewpoint of the same set of facts is that a vendor whose valuation is sky high due to solid demand is now funding its own demand pipeline. And that by funding one client over others. Red flags anyone?

Maybe they are confusing antipasto in "an Italian meal is nothing without antipasto" with being anti pasta ...

More science data loss with potential impact on hurricane season: processing and delivery of Special Sensor Microwave Imager/Sounder ends by 30 Jun. SSMIS is used by forecasters to assess tropical storm location, cloud structure, and intensity. It’ll lead to 📉 in monitoring ability & forecast skill.

Great to see xLSTMs excelling at human action segmentation. arxiv.org/abs/2506.09650

Mein Buch “Was kann Künstliche Intelligenz?“ ist erschienen. Eine leicht zugängliche Einführung in das Thema Künstliche Intelligenz. LeserInnen – auch ohne technischen Hintergrund – wird erklärt, was KI eigentlich ist, welche Potenziale sie birgt und welche Auswirkungen sie hat.

We are soooo proud. Our European-developed TiRex is leading the field—significantly ahead of U.S. competitors like Amazon, Datadog, Salesforce, and Google, as well as Chinese models from companies such as Alibaba.

Attention!! Our TiRex time series model, built on xLSTM, is topping all major international leaderboards. A European-developed model is leading the field—significantly ahead of U.S. competitors like Amazon, Datadog, Salesforce, and Google, as well as Chinese models from Alibaba.

TiRex 🦖 time series xLSTM model ranked #1 on all leaderboards. ➡️ Outperforms models by Amazon, Google, Datadog, Salesforce, Alibaba ➡️ industrial applications ➡️ limited data ➡️ embedded AI and edge devices ➡️ Europe is leading Code: lnkd.in/eHXb-XwZ Paper: lnkd.in/e8e7xnri shorturl.at/jcQeq

Ever though you could get a 35 million parameter time series state-of-the art foundation model that you can run on embedded hardware? Thanks to @hochreitersepp.bsky.social and his team at NXAI, you can. Amazing work! Paper: arxiv.org/abs/2505.23719 Code: github.com/NX-AI/tirex

If you have to say it ... it is an issue. Reminds me of football club presidents coming out and saying their manager is safe. Usually that means said manager is gone not long after. If a risk does not exist it need not be addressed.

Nice to see xLSTM slightly edging the transformer in classifying assembly tasks. arxiv.org/abs/2505.18012

Nice to see xLSTM applied and succeeding in speech enhancement. arxiv.org/abs/2501.06146

Surprising it should not be ...

Not sure "come train our proprietary AI with your data so we can sell it back to you" is such a catchy slogan ...

Some warped memory that. Is the US VP ok? Maybe he needs a break. Also, I remember the term "freedom fries" which replaced French fries which were never French to start ... anyway. youtu.be/CpuN-yM1sZU?...

Thank you @digthatdata.bsky.social . Will mull over.

Interesting observations. arxiv.org/abs/2504.07965

I hope they thought through Northern Ireland and its hybrid status of being in the Single Market but part of the UK.

Great read! arxiv.org/abs/2401.17173

Trade balance on GOODS only it seems. If the incoming reciprocal actions to these reciprocal tariffs include services into a similarly Kafkaesque formula, that would be bad news for Silicon Valley.

In summary, 1. , 4. and 5. made me into a functional analyst, 2. and 3. were key to my research in mathematics, 6. and the entire circle of ideas about memory that arose from this inspire me even today. 7. literally helped me navigate 2008/9.

Very topical again today, I feel. Helps in understanding which theories are necessarily peddled by "flat earthers" and which might have a point. I actually met Wynne, sadly long before I was at all interested in economics. A very decent gentleman, indeed.

7. The only economics books that I could actually use to a) make sense of the data I was seeing and hence b) make money from trading: Wynne Godley's "Macroeconomics" and (together with Marc Lavoie) "Monetary Economics". In AI this would be called "Sector Accounting is All You Need".

6. The original LSTM papers (1994 and 1997 version). I might be biased but still very topical and, imho, under explored.

5. Tim Gowers' paper on arithmetic progressions of length 4. Kick started an entire circle of ideas (still going) that has revolutionized what we know about primes, upper bounds of how random things can be and techniques to exploit the latter.

Best reference is probably Atiyah's book on K theory (recommended anyway) and his reference of his paper with Bott on the subject.

4. A way of rigorously developing complex analysis (incl. it's topological aspects) via Fredholm operators on Hilbert Spaces, winding numbers as indices of the Cauchy-Riemann operators etc. Not how one should teach it - but an amazing demonstration of the synthesis of 19th and 20th century geometry.

Contained key ideas of how to tackle automatic continuity questions which ultimately were extended to throw up key connections to even having to involve which axioms we are using (see 2.).

3. Graham Allan's paper on "Embedding the algebra of formal power series in a Banach algebra" and the resulting notion of an element of finite closed descent (i.e. where the CLOSURE of principal ideals generated by its powers stabilizes).

2. Dana Scott's lecture notes on Boolean Valued Forcing in set theory. These are, sadly, unpublished and, I think, Scott's interpretation of the highly cited but also unpublished Solovay-Scott paper on the same topic. For anyone interested in where maths crosses over into philosophy, a must read.

Taught me that nothing really is "elementary" (Riesz's lemma in this case) in functional analysis or any second order logic system (here complete normed spaces that are infinite dimensional) and that the search for reduction to "atomic" basic notions (here density via Riesz's lemma) is really key.

Taught me that nothing really is "elementary" (Riesz's lemma in this case) in functional analysis or any second order logic system (here complete normed spaces that are infinite dimensional) and that the search for reduction to "atomic" basic notions (here density via Riesz's lemma) is really key.

Great thread @mariokrenn.bsky.social and apologies for my late answer. Here is my own necessarily very biased list. 1. Tom Ransford's ingeniously simple proof of the Bishop-Stone-Weierstrass theorem. (Math. Proc. Camb. Phil. Soc. 96, pp. 309--311).

Sounds odd, but this is part of what makes this place special. Like the "mallard" in Trinity's dining hall ...

Seeing that X (=Twitter) was just acquired by xAI, I wonder why people want to keep their accounts and supply training data for free?One lesson learnt: we should all keep ownership and custody of our own data ...

xLSTM shines at tumor segmentation: arxiv.org/abs/2502.00314 “evaluated state-of-the-art segmentation methods, including U-Net and its enhanced variants with Transformers and Mamba. Our proposed ViLU-Net [vision xLSTM-Net] model achieved superior performance with reduced complexity.” Cool.

xLSTM for time series with Granger causality: arxiv.org/abs/2502.09981 xLSTM again shows superb performance at time series analysis. "Our experimental evaluations on three datasets demonstrate the overall efficacy of our proposed GC-xLSTM model."

Join Our Research Team in Linz! We are looking for 5 PostDocs and 10 PhDs in Machine Learning working on xLSTM, NLP, robustness, density ratio. Deadline: 04/20/25. More details: www.jku.at/en/lit-artif... #MachineLearning #DeepLearning #ResearchOpportunities #PhDPositions

xLSTM for Automated Stock Trading: arxiv.org/abs/2503.09655 xLSTM outperforms LSTM. "These findings mark the potential of xLSTM for enhancing DRL-based stock trading systems." This reinforcement learning approach uses xLSTM in both actor and critic components, which increases the performance.

Visited with legendary #AI pioneer @hochreitersepp.bsky.social at #JKULinz. We discussed the efficiency of #xLSTM against #GPT and why AI's future may be located on the edge of networks, rather than in centralised data centres via @bloomberg.com www.bloomberg.com/news/article...

Huge achievement: arxiv.org/abs/2503.14376 xLSTM kernels are now the fastest kernels both for training and inference. Faster than Flashattention or Mamba kernels. High arithmetic intensity. Optimized use of GPUs that set a new state of the art. Congratulations to the team. It is highly impressive.

Exploration imitation learning architectures: Transformer, Mamba, xLSTM: arxiv.org/abs/2502.12330 *LIBERO: “xLSTM shows great potential” *RoboCasa: “xLSTM models, we achieved success rate of 53.6%, compared to 40.0% of BC-Transformer” *Point Clouds: “xLSTM model achieves a 60.9% success rate”

xLSTM outperforms Conformers (Transformers) and Mamba on Speech Enhancement: arxiv.org/abs/2501.06146 “xLSTM-SENet2, outperforms state-of-the-art Mamba- and Conformer-based systems on the Voicebank+DEMAND.” New xLSTM-Triton-Kernels are faster than FlashAttention 3 & Mamba for training and inference.

Often LLMs hallucinate because of semantic uncertainty due to missing factual training data. We propose a method to detect such uncertainties using only one generated output sequence. Super efficient method to detect hallucination in LLMs.

Very curious about answers by @sylvaingigan.bsky.social @jenseisert.bsky.social @aspuru.bsky.social @andrew.diffuse.one @franknoe.bsky.social @kylecranmer.bsky.social @milescranmer.bsky.social and many others!

After reading it in end of 2013, i was searching for applications. In early 2014, i found the first example: Computers for designing de-novo physics experiments journals.aps.org/prl/abstract... For 10y I try to improve the scope, quality and creativity of these algorithms.