Michele Ceriotti

Features reconstruction errors between the latent spaces of several universal MLIPs

No day goes by without a new universal #ML potential. But how different they really are? Sanggyu and Sofiia tried to give a quantitative answer by comparing the reconstruction errors between their latent-space features. If you are curious, check out the #preprint arxiv.org/html/2512.05...

San Diego! Must admit I did miss the perpetual summer.

Another day, another talk. Looking forward to the impromptu workshop organized by Yang Shao-Horn's lab!

Boston dawn at the #MRS25. One talk done, two to go for this 🇺🇸 pre-xmas grand tour. May the saint protector of jet lag help me.

Our yearly potato harvest

Yearly reminder that growing food is hard.

This could be because they fear retaliation, or because they knew they were dishonest in the first place, and I don't know what disturbs me more. Really a sad state for a community.

I wouldn't have guessed it would have turned into such a circus, but that was clearly wasted time from the get go, with such sloppy and/or dishonest reviews. To me the most interesting thing is that as soon as the anonymous reviewers were revealed they rushed to change their scores left and right.

the ICLR review process is rapidly turning into a theater of the absurd 🍿

PET-MAD as a lightweight universal interatomic potential for advanced materials modeling - Nature Communications PET-MAD is a fast and lightweight universal machine-learning potential, trained on a small but diverse dataset, that delivers near-quantum accuracy in atomistic simulations for both organic and inorga...

📢 PET-MAD is here! 📢 It has been for a while for those who read the #arXiv, but now you get it preciously 💸 typeset by @natcomms.nature.com Take home: unconstrained architecture + good train set choices give you fast, accurate and stable universal MLIP that just works™️ www.nature.com/articles/s41...

If you're at EPFL and want to know about the new hybrid B200+B40s cluster Lyra 😋, that will arrive mid-2026, don't miss the user meeting on Thursday, December 11th, from 2:30 to 3:30pm. Check your e-mail or ask scitas@epfl.ch to get the zoom link 😁

a meme with scilla and cariddi and ulysses ship trying to sail through, making fun of the peer review process at ICLR

made by a friend of a colleague. iykyk

Let's write one together. I also hate them - they're the counterexample to "a figure is worth a thousand words", rather "a figure that could have been one number". still, if you don't include your scatter plot that looks like a line, it's not a good model assessment. 🤷

Excited to be here!

Incidentally, received also reviews for a #JCP paper which are thoughtful, constructively critical and actionable 🤩. Thank you anonymous reviewer for doing the work as it should be done. Looks like this is a domain in which AI cannot fix human incompetence and/or bad faith.

I was very curious of the @iclr-conf.bsky.social experiment of allowing LLM use in reviews, as I've used LLMs to "self-review" @labcosmo.bsky.social manuscripts and found it quite powerful. Sadly, what I'm seeing is all slop. I thought CS grad students would be better at LLMs. Experiment failed 🤷

Electronic fibres via the thermal drawing of liquid-metal-embedded elastomers - Nature Electronics A thermal drawing method can be used to create stretchable electronic fibres from a liquid-metal-embedded elastomer that can contain conductive and insulating domains across the fibre cross-section.

💡Exciting innovation from the Laboratory of Photonic Materials and Fiber Devices! Prof. Fabien Sorin and his team have developed a innovative stretchable electronic fiber. Read more in Nature Electronics: www.nature.com/articles/s41...

Time to reach for the wallet. And maybe time for the PSF to consider relocating to a country that is not sliding into the dark ages, or at least not so quickly.

Machine learning for atomic-scale simulations: balancing speed and physical laws - ΑΙhub

‪In this blog post, Filippo Bigi, Marcel Langer (@labcosmo.bsky.social‬) and @micheleceriotti.bsky.social write about the need to balance speed and physical laws when using ML for atomic-scale simulations
aihub.org/2025/10/10/m...

Machine learning for atomic-scale simulations: balancing speed and physical laws - ΑΙhub

A primer for non conservative (& rotationally unconstrained) MLIPs, and how to use them safely. Thanks @aihub.org for the space! aihub.org/2025/10/10/m...

Heavily inspired by the @nccr-marvel.bsky.social Inspire Fellowship, that is also running its last round of applications this Fall!

🌍 EPFL’s Institute of Materials (IMX) is offering 6-month funded fellowships for external Master’s students in Materials Science.
Join one of IMX’s world-class labs and gain hands-on research experience at the cutting edge of materials science.
👉 lnkd.in/enNeKmTh

Aerial view of the EPFL campus

Are you or do you know a smart, motivated Master student in a materials-adjacent area interested in doing a research project at EPFL? The @materials-epfl.bsky.social MX Fellowship applications open today sti.epfl.ch/smx/mx-fello.... Choose a lab, send your file, and spend up to six months at EPFL!

Learning Inter-Atomic Potentials without Explicit Equivariance Accurate and scalable machine-learned inter-atomic potentials (MLIPs) are essential for molecular simulations ranging from drug discovery to new material design. Current state-of-the-art models enforc...

Looks like @ox.ac.uk forbids their researchers to do any kind of literature search, though it seems that thankfully they can still submit to the arxiv arxiv.org/abs/2510.00027 🤷

Job Opportunities at CQSL At the Computational Quantum Science Lab we typically have several openings yearly (at the PhD/ Postdoc level). Applications are reviewed twice a year, at the beginning of April and at the beginning o...

I have a new opening for several funded PHD positions in my group (the Computational Quantum Science Lab, at EPFL). If you are a talented, motivated student, please apply here www.epfl.ch/labs/cqsl/jo.... I am especially looking to hire in scientific ML applications/NQS; not in quantum computing.

Flying ITA and 🤞 I'll make it alive. Total shitshow. I think they'll end up bankrupt like all the Alitalia descendants have before them, hope after I land.

📝 We have been told (& been telling) that ML potentials are linked quite directly to the expansion of the atomic energy into pairs, triples, and so on. But is this actually true 🤔? Go read the latest from the 🧑‍🚀 team (w/QM help from Joonho's team at Harvard) to find out more arxiv.org/html/2509.14...

porcini in tha forest

porcini on tha board

porcini mit botox

'this the season! 🍄

🎉 Obv. lots of the merit goes to the hard work of the @labcosmo.bsky.social 🧑‍🚀 team, to the support of my colleagues @materials-epfl.bsky.social, and to the inexhaustible patience of my partner (coincidentally, yesterday we celebrated 15 years of being lawfully bound by article 143CC🎂) and kiddos.

error plots for the PET-MAD-DOS model on different datasets

Anticipating 🧑‍🚀 Wei Bin's talk at #psik2025 (noon@roomA), 📢 a new #preprint using PET and the MAD dataset to train a universal #ml model for the density of states, giving band gaps for solids, clusters, surfaces and molecules with MAE ~200meV. Go to the talk, or check out arxiv.org/html/2508.17...!

A cartoon explaining how mild finite-temperature conditions induce disorder and dynamical reconstruction on the surfaces of lithium thiophosphates

📢 Now out on @physrevx.bsky.social energy, journals.aps.org/prxenergy/ab... from 🧑‍🚀 @dtisi.bsky.social and Hanna Türk, our #PET -powered study of the dynamic reconstruction of LPS surfaces, and how it affects their structure, stability and reactivity.