Sign in

Andreas Maier

@akmaier.bsky.social
82 followers 87 following 205 posts

Father of Two, Professor in Computer Science at FAU Erlangen Nuremberg

PostsRepliesMedia
Andreas Maier @akmaier.bsky.social · 11h
Great to meet Nicolas Vandewalle from the University of Liège at the Physics Underlying Life Sciences (PULS) Workshop 2026 in Zadar, Croatia. Looking forward to bringing some ML into his robots!
Image attached to post
000
Andreas Maier @akmaier.bsky.social · 08/10/2026
Great to meet Tim Tabrizi today at infoteam's QuantumDay26 in Bubenreuth. Fantastic talks on the progress of quantum computing, and a good reminder of how fast this field is moving right on our doorstep.
Image attached to post
000
Andreas Maier @akmaier.bsky.social · 08/10/2026
Congratulations to Hendrik Schröter for successfully defending his PhD at @fau.de! Great work, Hendrik. Proud to have you as a graduate of the Pattern Recognition Lab.
Image attached to post
010
Andreas Maier @akmaier.bsky.social · 05/10/2026
Deadline extended! Due to popular demand, abstracts for the 2nd Bavarian Conference on AI in Medicine are open until 9 October. Only 300 words. With @fau.de, @tum.de and five more Bavarian partners. events.hifis.net/event/3582
Image attached to post
000
Andreas Maier @akmaier.bsky.social · 05/10/2026
MICCAI 2026: 1,167 papers from 4,402 reviewed. @fau.de brought eleven, plus first place in PENGWIN and three task wins in VLM3D. A map of the week rather than a review — I did not sit through all eleven. akmaier.substack.com/p/eleven-paper…
akmaier.substack.com
Eleven Papers and Two Challenge Wins: What FAU Took to Strasbourg
Andreas Maier (2026). Eleven Papers and Two Challenge Wins: What FAU Took to Strasbourg. Conference report, MICCAI 2026, Strasbourg, 27 September – 1 October 2026.
000
Andreas Maier @akmaier.bsky.social · 02/10/2026
@anthropic.com's Opus 5.5 page says its testing ran at max effort. It never says which effort level produced the Opus 5 column it beats. That unstated setting is the size of the advertised win. akmaier.substack.com/p/one-thousand…
akmaier.substack.com
Both Columns, One Vendor: What the Opus 5.5 Benchmarks Can and Cannot Show
Anthropic (2026). Claude Opus 5.5. Product launch page, 22 September 2026. https://www.anthropic.com/claude-opus-5-5
000
Andreas Maier @akmaier.bsky.social · 01/10/2026
Congratulations to Team Beyond Normals — three task wins at VLM3D at MICCAI 2026, plus second place in MRI Abnormality Classification! A joint effort between @fau.de, Bernhard Kainz's IDEA Lab, Siemens Healthineers and d.hip.
Image attached to post
010
Andreas Maier @akmaier.bsky.social · 01/10/2026
A fantastic evening with Prof. Jang-Hwan Choi and the team from Ewha Womans University on the last day of MICCAI 2026. Our groups have been friends since our time at Stanford many years ago. Great to see @fau.de and Ewha at one table.
Image attached to post
000
Andreas Maier @akmaier.bsky.social · 01/10/2026
1925: a cartel fixed bulb life at 1,000 hours, fining lamps that lasted longer. 2026: an @anthropic.com default decides how long a model thinks — same name, same price. No public benchmark can see it. akmaier.substack.com/p/turning-ai-i…
akmaier.substack.com
Turning AI Into Light Bulbs: Planned Obsolescence Does Not Need a Broken Part
GitHub issue anthropics/claude-code#42796, maintainer comment of 2026-04-06 (issuecomment-4194007103), with thread reactions; plus first-hand observation from this newsletter's own production run, 2026-09-22.
110
Andreas Maier @akmaier.bsky.social · 30/09/2026
The Pattern Recognition Lab at MICCAI 2026 — present and former members in one photo. Great to meet up with the pattern recognition family again. And yes, the @fau.de hoodies do make us easy to spot in a crowd.
Image attached to post
011
Andreas Maier @akmaier.bsky.social · 30/09/2026
CodeMidas builds RL coding tasks from source alone: strip the implementation, let an agent write the spec, run the original for the expectations. The striking result: 3k filtered tasks beat 8k unfiltered ones. akmaier.substack.com/p/the-midas-to…
akmaier.substack.com
The Midas Touch for Code: Scaling AI Coding Environments Straight from Source
Ye B., Li L., Li S., Yue Z., Zhang L., Lv H., Liu Y., Ma W., Tian H., Li R., Dong J., Zhao Y., Deng X., Zhang H., Zhao L., Liu Q., Kong L., Yang T., Luo F. CodeMidas: Scaling Agentic Coding RL Environments from Code Itself. arXiv:2609.22068 (2026).
030
Andreas Maier @akmaier.bsky.social · 29/09/2026
A preprint from @jhu.edu makes ordinary chest CT look like photon-counting CT by inverting a simulated degradation. The image gains are measured on their own synthetic data, never on real paired scans. akmaier.substack.com/p/pcct-look-ei…
akmaier.substack.com
PCCT Look, EICT Information: What SUMI Does and Does Not Distil
Liu J., Zhou X., Li W., Ye S., Sitek A., Yang X., Tang Y., Xu D., Ding K., Wang K., Yang Y., Yuille A. L., Zhou Z. Distilling Photon-Counting CT into Routine Chest CT through Clinically Validated Degradation Modeling. arXiv:2604.07329 (2026).
000
Andreas Maier @akmaier.bsky.social · 28/09/2026
The Pattern Recognition Lab on the big stage at MICCAI 2026 — congratulations to Daiqi Liu for a brilliant talk on speech-guided multimodal learning for vocal tract segmentation in real-time MRI! Proud of @fau.de today.
Image attached to post
000
Andreas Maier @akmaier.bsky.social · 28/09/2026
Laya, on @hf.co: 421M params, Apache 2.0, typed questions answered with calibrated probabilities in ~40 ms, no text generated. An open clone of a closed product launched four days earlier. akmaier.substack.com/p/laya-jev-and…
akmaier.substack.com
Laya, Jev and the Return of the Discriminative Model
Nandakishor M (Convai Innovations). Laya: Multilingual, non-autoregressive System 1 decision model. Hugging Face model card, convaiinnovations/laya, Apache 2.0, first published 2026-09-18.
021
Andreas Maier @akmaier.bsky.social · 27/09/2026
Congratulations to all the award winners at RIME 2026 at MICCAI — the 2nd Workshop on Reconstruction and Imaging Motion Estimation! A wonderful workshop, and really good times with Thomas Küstner and Lina Felsner. Glad @fau.de was among the organisers.
Image attached to post
010
Andreas Maier @akmaier.bsky.social · 27/09/2026
Congratulations to Haobo Song — first place in Task One and fourth in Task Two of the PENGWIN Challenge 2026 at MICCAI! A great collaboration between @fau.de and Siemens Healthineers. Very proud of this team.
Image attached to post
010
Andreas Maier @akmaier.bsky.social · 27/09/2026
Glad to meet Julia Schnabel at MICCAI and talk about the 2nd Bavarian Conference on AI in Medicine — 30 Nov to 1 Dec in Munich. Abstract deadline is 5 October, don't forget to submit! events.hifis.net/event/3582
Image attached to post
000
Andreas Maier @akmaier.bsky.social · 27/09/2026
A real pleasure to welcome Prof. Huan Song and the team from West China Hospital to Erlangen — time at Siemens Healthineers headquarters and at @fau.de. Thank you for making the long journey!
Image attached to post
010
Andreas Maier @akmaier.bsky.social · 25/09/2026
1995: SATAN, a network scanner, was called "rocket launchers for everyone". 2026: @anthropic.com's Mythos finds bugs on its own. Louis Anslow says same panic, same lesson. I think the rhetoric repeats but the tool does not. akmaier.substack.com/p/before-mytho…
akmaier.substack.com
Before Mythos There Was SATAN: What a 1995 Panic Does and Does Not Teach Us
Anslow L. (2026). Before MYTHOS, SATAN: a 1990s Software Satanic Panic. Pessimists Archive newsletter, 17 June 2026. https://newsletter.pessimistsarchive.org/p/before-mythos-satan-a-1990s-software
000
Andreas Maier @akmaier.bsky.social · 24/09/2026
WFM (Tencent Youtu, @monashuniversity.bsky.social et al.): agent memory as a learned dense wiki graph, not sparse triples; 3–5 points over an in-house baseline. But the draft calls some cells "constructed planning values". akmaier.substack.com/p/from-sparse-…
akmaier.substack.com
From Sparse Triples to Dense Wiki – Why Agents Need a Better Memory
Dong J., Luo L., Zhang S., Chen G., Guo T., Yu Y., Tao R., Guo T., Zhang Q.-W., An S., Qiao R., Sun X. (2026). WFM: Wiki Foundation Model for Complex Agentic Reasoning. arXiv:2609.18182. https://arxiv.org/abs/2609.18182
000
Andreas Maier @akmaier.bsky.social · 23/09/2026
JarvisGUI benchmarks GUI agents across Android, Windows and Ubuntu in one episode. Best open-source agent: 42 % on single steps, 8 % on composed workflows, 2 % when state must cross devices. akmaier.substack.com/p/crossing-the…
akmaier.substack.com
Crossing the Digital Divide: A New Benchmark for Multi-Device GUI Agents
Chen Z., Lu Y., Cheng Z., Liu Z., Bai J., Huang Z., Lin Z., Li Z., Guo Y., Wang Y., Wang H. (2026). JarvisGUI: Towards Cross-Device GUI Agents with Dynamic Task Composition. arXiv:2609.10451; accepted to EMNLP 2026 (Main Conference). https://arxiv.org/abs/2609.10451
000
Andreas Maier @akmaier.bsky.social · 22/09/2026
97-page survey of self-improving agents (KAUST, Jilin, @ualberta.bsky.social, IDSIA; Schmidhuber last author): weights + scaffold, an update operator on either. Clean formalism; the findings are plateaus, not an explosion. akmaier.substack.com/p/self-improvi…
akmaier.substack.com
Self-Improving Agents, Surveyed: A Clean Formalism and a Modest Reality
Ren Z., Chen Y., Guo D., Rong G., Li T., Xiong R. B., Lan Q., Wang W., Nanbo L., Yang Y., Zhuge M., Schmidhuber J. (2026). Self-Improvements in Modern Agentic Systems: A Survey. arXiv:2607.13104 (14 July 2026). https://arxiv.org/abs/2607.13104
000
Andreas Maier @akmaier.bsky.social · 21/09/2026
Felin & Holweg (Strategy Science 2024): AI predicts from old data; humans create novelty by believing ahead of the data. Well argued – but their “AI” is a 2023 chatbot. Machines that design experiments already run that loop. akmaier.substack.com/p/theory-is-al…
akmaier.substack.com
Theory Is All You Need: A Strong Case Against Prediction That Underrates the Machines
Felin T., Holweg M. (2024). Theory Is All You Need: AI, Human Cognition, and Causal Reasoning. Strategy Science 9(4):346-371. https://doi.org/10.1287/stsc.2024.0189
000
Andreas Maier @akmaier.bsky.social · 18/09/2026
The Scientific Challenge Atlas (@umich.edu, OpenAI, Argonne) maps what AI can solve by reasoning and compute alone – measurement is out of scope. My criticism: AI-supported acquisition is the next big thing, left in the dark. akmaier.substack.com/p/if-you-have-…
akmaier.substack.com
If You Have a Hammer: What the Scientific Challenge Atlas Leaves in the Dark
Duraisamy K., Stevens R., Pruet J. et al. (2026). The Scientific Challenge Atlas: Mapping the machine-accessible frontier of science. Michigan Institute for Computational Discovery and Engineering, University of Michigan, with OpenAI, Los Alamos and Argonne National Laboratories. https://micde.umich
001
Andreas Maier @akmaier.bsky.social · 17/09/2026
MIT’s ad hoc committee on AI in teaching and research: eight principles, a four-tier syllabus policy menu, AI-resilient exams. 12% survey response, headline finding anecdotal – a template to borrow, not evidence. @mit.edu akmaier.substack.com/p/when-ai-meet…
akmaier.substack.com
When AI Meets the “Mind-and-Hand” Ethos: MIT’s Roadmap for Teaching, Learning, and Research
MIT Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training (2026). Report of MIT's Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training. Massachusetts Institute of Technology, August 13, 2026. https://aiandeducation.mit.edu/report/
000
Andreas Maier @akmaier.bsky.social · 16/09/2026
@anthropic.com's September 2026 Threat Intelligence report: how Claude was used in real cyber and influence operations, Dec 2025–Aug 2026. Concrete IOCs, but a curated sample with no denominator. My read: akmaier.substack.com/p/ai-in-the-cr…
akmaier.substack.com
AI in the Crosshairs: Detecting and Countering Misuse, September 2026
Anthropic (September 2026). Detecting and countering misuse of AI: September 2026. Anthropic Threat Intelligence.
000
Andreas Maier @akmaier.bsky.social · 15/09/2026
In a September 2026 op-ed, the author likens @anthropic.com's Harness permission system to a Watschenbaum—effective on fire extinguishers but blind to gas leaks. Claude is explicitly exonerated. akmaier.substack.com/p/the-wrong-wa…
akmaier.substack.com
The Wrong Watschenbaum: Anthropic's Harness Audits the Fire Extinguisher, Ignores the Gas Leak
Andreas Maier (September 2026). The Wrong Watschenbaum: Anthropic's Harness Audits the Fire Extinguisher, Ignores the Gas Leak. Personal op-ed.
000
Andreas Maier @akmaier.bsky.social · 14/09/2026
Fastcode advertises a $100K “Research as a Service” contract that delivers a conference paper, citing nine prior works as a track record. Checks show all five published papers predate the firm and have no Fastcode affiliation. akmaier.substack.com/p/pay-for-a-pa…
akmaier.substack.com
Pay-for-a-Paper? How Fastcode AI Packs a $100 K Contract into a Conference Publication
Fastcode AI (2026). Research as a Service. https://research.fastcode.ai
000
Andreas Maier @akmaier.bsky.social · 14/09/2026
Fastcode advertises a $100K “Research as a Service” contract that delivers a conference paper, citing nine prior works as a track record. Checks show all five published papers predate the firm and have no Fastcode affiliation. akmaier.substack.com/p/pay-for-a-pa…
akmaier.substack.com
Pay-for-a-Paper? How Fastcode AI Packs a $100 K Contract into a Conference Publication
Fastcode AI (2026). Research as a Service. https://research.fastcode.ai
000
Andreas Maier @akmaier.bsky.social · 12/09/2026
A great pleasure to announce that FAU has just signed an MoU with IIT Kharagpur, together with FAU President @hornegger.bsky.social. Our collaboration will focus on student exchange and joint research in MedTech, AI, and Robotics.
(announcement)
010
Andreas Maier @akmaier.bsky.social · 11/09/2026
I was touched by the wonderful welcome at NITTE University. I'm very glad we signed an MoU, with a special focus on pattern recognition and the medical and dental sciences.
(announcement)
000
Andreas Maier @akmaier.bsky.social · 11/09/2026
Thrilled to sign our second MOU on this India trip with our new friends at Manipal Academy of Higher Education. Their focus on medical AI excellence makes them an ideal collaboration partner for us.
(announcement)
000
Andreas Maier @akmaier.bsky.social · 11/09/2026
A physics-based Fisher-KPP reaction-diffusion model applied to MRI predicts infarct growth. A perfusion-derived heterogeneous diffusion field improves on reaction-only baselines, captures penumbral expansion and advances forecasting. akmaier.substack.com/p/putting-the-…
akmaier.substack.com
Putting the Brakes on Guesswork: A Physics-Based Equation for Stroke Damage
Abbas, M. H., Balcerak, M., Ahmad, A., Menze, B., & de la Rosa, E. (2026). An Imaging-Informed Reaction-Diffusion Model of Infarct Growth. MICCAI 2026 Satellite Events (LNCS), accepted. arXiv:2608.20935.
000
Andreas Maier @akmaier.bsky.social · 11/09/2026
Had a wonderful last dinner in Bangalore with Felix Winter, @hornegger.bsky.social, Sheethal Bhat, Jan Künne, and friends from Atlas University Istanbul.
(announcement)
010
Andreas Maier @akmaier.bsky.social · 11/09/2026
A physics-based Fisher-KPP reaction-diffusion model applied to MRI predicts infarct growth. A perfusion-derived heterogeneous diffusion field improves on reaction-only baselines, captures penumbral expansion and advances forecasting. akmaier.substack.com/p/putting-the-…
akmaier.substack.com
Putting the Brakes on Guesswork: A Physics-Based Equation for Stroke Damage
Abbas, M. H., Balcerak, M., Ahmad, A., Menze, B., & de la Rosa, E. (2026). An Imaging-Informed Reaction-Diffusion Model of Infarct Growth. MICCAI 2026 Satellite Events (LNCS), accepted. arXiv:2608.20935.
000
Andreas Maier @akmaier.bsky.social · 10/09/2026
Glad to announce that FAU just signed an MoU with RV University Bangalore, India, together with President @hornegger.bsky.social. Excited for many joint research projects ahead.
(announcement)
010
Andreas Maier @akmaier.bsky.social · 10/09/2026
Glad to visit my friends at Siemens Healthineers in Bangalore.
(announcement)
000
Andreas Maier @akmaier.bsky.social · 10/09/2026
The NeurIPS 2025 paper “Reading Between the Lines” presents vec2vec, a method that maps text embeddings between language models without paired data, using shared geometry. It builds on work on cross-model convergence. akmaier.substack.com/p/reading-betw…
akmaier.substack.com
Reading Between the Lines: Translating Text Embeddings Across Model Borders
Jha, R., Zhang, C., Shmatikov, V., & Morris, J. X. (2025). Harnessing the Universal Geometry of Embeddings. Advances in Neural Information Processing Systems 38, 51335-51359. DOI 10.52202/085713-1530
000
Andreas Maier @akmaier.bsky.social · 10/09/2026
The NeurIPS 2025 paper “Reading Between the Lines” presents vec2vec, a method that maps text embeddings between language models without paired data, using shared geometry. It builds on work on cross-model convergence. akmaier.substack.com/p/reading-betw…
akmaier.substack.com
Reading Between the Lines: Translating Text Embeddings Across Model Borders
Jha, R., Zhang, C., Shmatikov, V., & Morris, J. X. (2025). Harnessing the Universal Geometry of Embeddings. Advances in Neural Information Processing Systems 38, 51335-51359. DOI 10.52202/085713-1530
000
Andreas Maier @akmaier.bsky.social · 10/09/2026
Pleasure to present at CXO Leadership Square on AI & Tech in Health alongside Jan Künne, head of Grundig Akademie, and @hornegger.bsky.social in Bangalore, India. Great discussions on the future of health tech!
(announcement)
021
Andreas Maier @akmaier.bsky.social · 10/09/2026
Great to connect with the Bengaluru Siemens Healthineers team after a successful work meeting with NIMHANS. Looking forward to what’s next!
Great to connect with the Bengaluru Siemens Healthineers team after a successful work meeting with NIMHANS.
000
Andreas Maier @akmaier.bsky.social · 10/09/2026
Great selfie today at NIMHANS with Dr. Janardhan Reddy, the new interim director, alongside FAU President @hornegger.bsky.social, Sheethal Bhat and Felix Winter from Medical Valley. Excited for more joint research between FAU and NIMHANS.
Great selfie today at NIMHANS with Dr. Janardhan Reddy, the institute's new interim director, together with FAU President Joachim Hornegger, Sheethal Bhat, and Felix Winter from Medical Valley. Looking forward to more joint research between FAU and NIMHANS.
011
Andreas Maier @akmaier.bsky.social · 09/09/2026
Great visit today with @hornegger.bsky.social at Bosch Global Software Technologies (BGSW) in Bangalore, hosted by Gaur Dattatreya. Inspiring exchange on research, software and how a university and a global engineering company can collaborate. Thanks for the hospitality.
Great visit today with FAU President Joachim Hornegger to Bosch Global Software Technologies (BGSW) in Bangalore, hosted by Gaur Dattatreya. Good exchange on research, software, and where a university and a global engineering company can work together. Thanks for the hospitality.
011
Andreas Maier @akmaier.bsky.social · 09/09/2026
Researchers including @fau.de show that a machine-learning framework can learn logical operations for arbitrary quantum error-correction codes, even non-additive ones; code and data are released publicly. akmaier.substack.com/p/when-the-mac…
akmaier.substack.com
When the Machine Learns to Do Quantum Logic
Meyer, N., Mutschler, C., Seuss, D., Maier, A., & Scherer, D. D. (2026). Learning logical operations for arbitrary quantum error correction codes. Physical Review A (accepted). DOI 10.1103/qj2x-3l6n. Preprint: arXiv:2605.28162
010
Andreas Maier @akmaier.bsky.social · 08/09/2026
@tum.de study presents ZODIAC, a zero-shot octree-based diffusion model that generates complete lumbar-spine anatomy from sparse intra-op ultrasound. It avoids supervised training on simulated occlusions by using a generative prior. akmaier.substack.com/p/when-the-spi…
akmaier.substack.com
When the Spine Goes Invisible: How “ZODIAC” Tries to Fill the Gaps in Intra-operative Ultrasound
Gafencu, M.-A., Bratulescu, V., Velikova, Y., Azampour, M. F., & Navab, N. (2026). ZODIAC: Zero-shot Octree-based Diffusion for Anatomical Completion. arXiv:2608.24422.
001
Andreas Maier @akmaier.bsky.social · 07/09/2026
Researchers at @umich.edu and @carleton.ca present a method that perturbs latent variables of an unsupervised sperm-whale coda model, revealing how click count, timing regularity and spectral features shape generated sounds. akmaier.substack.com/p/listening-to…
akmaier.substack.com
Listening to the Deep: How AI Probes Whale “Talk” by Pushing Its Latent Buttons
Begus, G., Leban, A., & Gero, S. (2026). Approaching an unknown communication system by latent space exploration and causal inference. Royal Society Open Science, 13(8), 250829. DOI 10.1098/rsos.250829
000
Andreas Maier @akmaier.bsky.social · 07/09/2026
Researchers at @umich.edu and @carleton.ca present a method that perturbs latent variables of an unsupervised sperm-whale coda model, revealing how click count, timing regularity and spectral features shape generated sounds. akmaier.substack.com/p/listening-to…
akmaier.substack.com
Listening to the Deep: How AI Probes Whale “Talk” by Pushing Its Latent Buttons
Begus, G., Leban, A., & Gero, S. (2026). Approaching an unknown communication system by latent space exploration and causal inference. Royal Society Open Science, 13(8), 250829. DOI 10.1098/rsos.250829
000
Andreas Maier @akmaier.bsky.social · 04/09/2026
img2threejs converts a reference photo into editable Three.js code using a multimodal LLM rather than depth estimation or photogrammetry. The Apache-2.0 tool, released Aug 23 2026, has ~13.8 k stars and 1.1 k forks. akmaier.substack.com/p/turning-a-si…
akmaier.substack.com
Turning a Single Photo into Editable Three.js Code – What “img2threejs” Really Does
img2threejs contributors (2026). img2threejs: Rebuild the object in a reference image as a code-only, procedural, quality-gated, animation-ready Three.js model. v1.5.1, Apache-2.0. https://github.com/img2threejs/img2threejs
000
Andreas Maier @akmaier.bsky.social · 04/09/2026
img2threejs converts a reference photo into editable Three.js code using a multimodal LLM rather than depth estimation or photogrammetry. The Apache-2.0 tool, released Aug 23 2026, has ~13.8 k stars and 1.1 k forks. akmaier.substack.com/p/turning-a-si…
akmaier.substack.com
Turning a Single Photo into Editable Three.js Code – What “img2threejs” Really Does
img2threejs contributors (2026). img2threejs: Rebuild the object in a reference image as a code-only, procedural, quality-gated, animation-ready Three.js model. v1.5.1, Apache-2.0. https://github.com/img2threejs/img2threejs
000
Andreas Maier @akmaier.bsky.social · 03/09/2026
Recording available: Prof. Sepp Hochreiter (JKU Linz, NXAI) on "The Next Phase of AI" — xLSTM, TiRex, and AI moving into the physical world. lme.tf.fau.de/news/recording-availa…
lme.tf.fau.de
Recording available: Prof. Sepp Hochreiter (JKU Linz, NXAI) — The Next Phase of Artificial Intelligence
Pattern Recognition Lab seminar recording: Sepp Hochreiter on xLSTM, TiRex, and AI's shift into the physical world.
000