ML in PL

ML in PL ML in PL Association is a non-profit organization devoted to fostering the ML community in Poland and promoting a deep understanding of ML methods.

Even though ML in PL is based in Poland, it seeks to provide opportunities for international cooperation.

Before generative AI became everyone's favorite topic, Jakub Tomczak was quietly building the foundations — and then wri...
17/06/2026

Before generative AI became everyone's favorite topic, Jakub Tomczak was quietly building the foundations — and then writing a book about them.

"Deep Generative Modeling," published in 2022, is still the text people in the field actually recommend when someone wants to understand what's happening under the hood of modern AI. Not just what it does, but why it works — or why it sometimes doesn't.

Jakub has over 50 publications at the conferences that matter in this field: NeurIPS, ICML, ICLR, CVPR, ICCV. In 2024 he was Program Chair of NeurIPS. He has advised eBay, Qualcomm, and several startups, and currently leads AI work at the Chan Zuckerberg Initiative.

He's joining us at the 10th edition of ML in PL Conference.

Warsaw. October 8–10. Copernicus Science Centre.

Two models. Trained independently, on different data, different tasks. Average their weights, the result should be garba...
12/06/2026

Two models. Trained independently, on different data, different tasks. Average their weights, the result should be garbage. Except sometimes it isn't. Why? That's one of the questions Emanuele Rodolà has been attacking, through geometry. And model fusion is just one corner of it, his work stretches across audio, language models, and multimodal learning.

He's the next name on our 10th edition lineup.

Emanuele Rodolà is a Full Professor of Computer Science at Sapienza University of Rome, where he leads the GLADIA AI group. His work in this field has been supported by an ERC grant, a FIS grant, and a Google Research Award. In the past, he was a postdoctoral researcher at USI Lugano (2016–2017), an Alexander von Humboldt Fellow at TU Munich (2013–2016), and a JSPS Research Fellow at the University of Tokyo (2013), in addition to visiting periods at Tel Aviv University, Technion, École Polytechnique, and Stanford. He is a fellow of ELLIS and a fellow of the Young Academy of Europe. Professor Rodolà has received numerous awards for his research and plays an active role in the academic community, serving on program committees and as Area Chair for major conferences in AI and ML. His current research focuses primarily on neural model fusion, representation learning, language models, ML for audio, and multimodal learning, with around 190 publications in these areas. His work has been featured in media outlets including Fortune, Wired, Italian national broadcast and newspapers.

See you in Warsaw for the 10th edition, October 8–10 at the Copernicus Science Centre.

Every year, ML in PL is built by organizers who somehow manage to fit community work between research, studies and their...
10/06/2026

Every year, ML in PL is built by organizers who somehow manage to fit community work between research, studies and their actual jobs. Two of them are taking on team coordinator roles this year:

Weronika Piotrowska leads our Special Ops Team. She's been with ML in PL for four years, is finishing her Computer Science Master's at Warsaw University of Technology, and works on problems at the intersection of computer vision and medicine. Outside of research and community work, she runs tabletop RPG campaigns, which feels like useful preparation for a role where contingency planning is part of the job.

Arkadiusz Paterak returns for his second year coordinating the Website Team. He's finishing his Master's at AGH and works as a software engineer at Nivalit, where he's helping build AI systems for mental health. When he's away from code, you'll usually find him reading or travelling with a camera nearby.

Centrum Nauki Kopernik is, by most definitions, a place for explaining things. That's a mission we recognize.There's som...
09/06/2026

Centrum Nauki Kopernik is, by most definitions, a place for explaining things. That's a mission we recognize.
There's something fitting about bringing a community ML conference to an institution built around the idea that complex things are worth understanding — and that understanding them should be accessible to everyone. The Copernicus Centre doesn't talk down to its visitors. Neither do we. It's one of the reasons we're genuinely glad to be back.
We're honoured to have CNK's honorary patronage again this year. Thank you for continuing to make this possible.
And if you have something worth sharing and understanding, there's a place for you in the program too — our Call for Contributions is open.

After the ICLR list, a few people asked whether we'd do the same for CVPR. So here it is: every CVPR 2026 paper we could...
05/06/2026

After the ICLR list, a few people asked whether we'd do the same for CVPR. So here it is: every CVPR 2026 paper we could find with a Polish author.

The interesting part was watching a theme surface on its own. Computer vision is enormous, but Polish-authored work this year clusters hard around 3D: novel view synthesis, Gaussian splatting, feed-forward reconstruction. The same few subfields come up again and again, and a couple of authors account for three papers each. The single oral on the list is 3D work too.

What makes a list like this worth keeping is what shows up once the work sits together. The authors are spread across labs on several continents. Some know each other. Plenty don't. A few are clearly working the same problem from different countries. And it isn't only the diaspora: several of the papers are fully Polish teams.

As with ICLR, the list is public and incomplete by definition. Missing a paper? Drop it in the comments or message us, and we'll keep adding.

https://conference.mlinpl.org/2026/cvpr-2026

Early bird registration for ML in PL 2026 is open — starting today, June 1st. It also happens to be Children's Day in Po...
01/06/2026

Early bird registration for ML in PL 2026 is open — starting today, June 1st. It also happens to be Children's Day in Poland. Apparently a good day to start things.

Register at the link in the comments.

That's a wrap on this season's recordings. The last batch goes wide: from the metal underneath your models, to ML rewrit...
29/05/2026

That's a wrap on this season's recordings. The last batch goes wide: from the metal underneath your models, to ML rewriting how science and engineering work, to what it takes to build AI that generalises beyond data-rich settings. A good one to end on.

🎓 Maciej Draguła & Artem Yerofieiev (Tenstorrent) — All Things Metal
Modern processors spend most of their time and energy waiting for data, and most developers never see it. Tenstorrent builds programmable RISC-V processors with explicit control over data movement, paired with TT-Metal, a low-level API for full hardware control, and TT-Train, a C++ multi-device training engine for transformer workloads. For anyone curious about what sits below PyTorch, this one goes all the way down.

🎓 Johannes Brandstetter — What's the Next Wave of Disruption in Science and Engineering?
From weather and climate modeling to computational fluid dynamics and multi-physics simulation, Johannes connects the dots and argues that scientific ML is past the proof-of-concept stage. The talk focuses on what it takes to build reference models for entire industry verticals and what that means for engineering process cycles.

🎓 Herke van Hoof — Modular Learning for Improving AI Assistants
Most AI success stories require abundant data. Robotics, real-world infrastructure, and scientific domains largely don't have that. Herke makes the case for modular approaches, where complex behaviour is composed from simpler elements, and walks through three projects where modularity improved generalisation, data efficiency, and instructability.

Links in the comments 👇

How far can a model generalize? Across scales, across domains, across the boundary between capability and safety? This b...
22/05/2026

How far can a model generalize? Across scales, across domains, across the boundary between capability and safety? This batch of recordings approaches that question from three very different angles — and together they cover a lot of ground worth exploring.

🎓 Jenia Jitsev — Open Foundation Models: Scaling Laws and Generalisation
Scaling laws let you predict behavior at larger scales from experiments at smaller ones. Jenia shows they can also systematically compare learning procedures for foundation models — turning intuition-driven search into something more principled. The talk also makes the case for why open models and datasets are scientifically necessary for this research, and is honest about what remains unsolved in measuring generalisation.

🎓 Alexey Dosovitskiy — From Pixels to Nucleotides
Alexey's research traces a path from computer vision through transformers to ML-based drug design. The talk covers "off-the-grid" architectures for visual data and connects that work to the core challenges of applying ML to mRNA-based drug design. A cross-domain talk that earns its framing by actually showing the connections.

🎓 Gerhard Wunder, Maura Pintor & Jakub Kałużny — Discussion Panel: AI in Security
As AI gets deployed in security-critical contexts, the questions get harder: what are the inherent vulnerabilities of AI models, what do adversarial attack vectors look like, and are traditional security measures adequate? The panel covers adversarial attacks, data privacy, and the ethical implications of AI in security — the kind of conversation that tends to be more useful than a single paper.

Links in the comments 👇

Why do deep networks train the way they do, and why do they forget what they learn? Those questions have run through fif...
19/05/2026

Why do deep networks train the way they do, and why do they forget what they learn? Those questions have run through fifteen years of Razvan Pascanu's work, from a PhD with Yoshua Bengio to Google DeepMind. He is the first name on our 10th edition lineup.

Razvan Pascanu has been a research scientist at Google DeepMind since 2014. Before this, he completed his PhD at Universite de Montréal with prof. Yoshua Bengio, where he worked on understanding deep networks, specifically recurrent neural architectures. During his career he has made significant contributions to theory of deep networks, optimization, recurrent architectures as well as deep reinforcement learning, continual learning, meta-learning and graph neural networks. For details on his work please see razp.info. He has been Program Chair for the Neural Information Processing Systems (NeurIPS) conference and currently acts as General Chair, as well as a Program Chair for the Conference on Life-long Learning Agents (CoLLAs) and the Learning on Graphs Conference (LoG). He has organized various workshops on topics such as continual learning at top-tier conferences. He is also one of the main organizers of the Eastern European Machine Learning Summer School (EEML) and EEML workshop series, as well as an organizer of the Romanian AI Days.

See you in Warsaw for the 10th edition, October 8–10 at the Copernicus Science Centre.

Medicine is one of the places where causal ML stops being a research exercise and starts having consequences. Which trea...
15/05/2026

Medicine is one of the places where causal ML stops being a research exercise and starts having consequences. Which treatment works for which patient? Which variables actually drive that difference? And can we trust the methods we're using to find out? These three talks take those questions seriously — from fundamental limits of causal discovery all the way to personalized treatment decisions in clinical data. Well worth your time.

🎓 Mateusz Gajewski & Mateusz Olko — Limits in Causal Discovery and the Path Forward
Most neural causal discovery methods rely on the faithfulness assumption: that conditional independencies in data reflect true causal structure. Their ICML 2025 paper shows that in nonlinear settings, reliable causal discovery requires exponentially many data points as graph size and density increase. The talk covers what comes next: amortized approaches, partial graph discovery, and grounding evaluations in real-world systems where ground-truth graphs are rarely available.

🎓 Paweł Morzywołek — Inference on Local Variable Importance Measures for Heterogeneous Treatment Effects
Treatment effects vary across individuals — the question is which variables drive that variation, tested rigorously enough for high-stakes domains like medicine. Paweł's framework provides local importance measures that can differ across individuals and global inference that tests whether a variable matters for anyone at all. Built on semiparametric theory, valid even when ML algorithms are used to estimate heterogeneity.

🎓 Michael Vollenweider — Learning Personalized Treatment Decisions in Precision Medicine
Not all treatment assignment bias hurts equally. Michael models different bias types using mutual information and shows that some — particularly those unrelated to outcome mechanisms — have minimal effect on counterfactual prediction accuracy. Benchmarked on semi-synthetic TCGA data and real-world drug and CRISPR screen outcomes.

Links in the comments 👇

Adres

Warsaw
02-097

Strona Internetowa

Ostrzeżenia

Bądź na bieżąco i daj nam wysłać e-mail, gdy ML in PL umieści wiadomości i promocje. Twój adres e-mail nie zostanie wykorzystany do żadnego innego celu i możesz zrezygnować z subskrypcji w dowolnym momencie.

Skontaktuj Się Z Firmę

Wyślij wiadomość do ML in PL:

Skróty

Udostępnij

Kategoria