LIAC at EPFL
@SchwallerGroup
Followers
2K
Following
2K
Media
50
Statuses
353
Laboratory of Artificial Chemical Intelligence at @EPFL | led by @pschwllr | AI-accelerated discovery and synthesis | part of @NCCR_Catalysis | team-run account
Lausanne, Switzerland
Joined October 2021
Word of the day: 'WOMEN'. Our mothers, grandmas, sisters, aunts, girlfriends, daughters, mentors, leaders, teachers, scientists, engineers, artists, activists who raise, inspire and lead. We stand in awe of their strength and tenderness. Here's to WOMEN, every day, everywhere! ๐
Can you decode our special message? ๐ต๏ธโโ๏ธโจ P. S. HAPPY INTERNATIONAL WOMEN'S DAY to all the inspiring women out there! ๐ฉโ๐ฌ๐ฉโ๐จ๐ฉโ๐ณ๐ฉ๐ฟโ๐ป๐ฉโ๐พ๐ฉโ๐๐ฉ๐ปโโ๏ธ๐ฉโโ๏ธ๐๐พ๐ฉโโ๏ธ๐ฉโ๐ญ๐ง๐ผ๐ฆธ๐ฝโโ๏ธ๐งโโ๏ธ๐คธโโ๏ธ๐ท๐พโโ๏ธ๐ฉ๐ผโ๐ค๐ฉโ๐๐๏ธโโ๏ธ๐ฎโโ๏ธ๐๐พโโ๏ธ๐ฉโ๐ซ๐ฉ๐ฝโ๐ผ P. P. S. Don't *see* the answer?๐ DW, our prof struggled too ๐@pschwllr
#EPFL #EPFLChem #InternationalWomensDay
0
2
22
Compared to state-of-the-art methods (e.g., MatterGen), MatInvent exhibits superior generation performance under property constraints while dramatically reducing the demand for property computation by up to 378-fold.
0
2
6
Thrilled to share our new work MatInvent, a general and efficient reinforcement learning workflow that optimizes diffusion models for goal-directed crystal generation. Thanks to @JeffGuo__ , @efssh , @pschwllr , @SchwallerGroup , @NCCR_Catalysis . https://t.co/S7ibrRNdxp
arxiv.org
Diffusion models promise to accelerate material design by directly generating novel structures with desired properties, but existing approaches typically require expensive and substantial labeled...
2
9
18
Exciting postdoc opportunity in the @SchwallerGroup at EPFL! We're hiring a postdoc to advance ML-driven synthesis planning after Zlatko Joncevโs successful exit to co-found B-12 (YC '25) ๐ Work on: - LLMs for strategic synthesis planning - Chemical reasoning at scale -
arxiv.org
While automated chemical tools excel at specific tasks, they have struggled to capture the strategic thinking that characterizes expert chemical reasoning. Here we demonstrate that large language...
0
17
38
Check out the fantastic entries, adsKRK and DynoAgent, from the @SchwallerGroup's LLMs for Materials and Chemistry Applications Hackathon team! From heterogeneous catalysis to proteins - MD simulations made simple and explainable ๐ @efssh @XuanVuNguyen18 @6ojaHa Ryo, Salomรฉ,
You donโt like molecular dynamics? We get it. Thatโs why at this yearโs LLM hackathon for Chemistry and Materials Science, we built not one, but โจtwoโจ AI agents for molecular dynamics ๐
1
3
18
You donโt like molecular dynamics? We get it. Thatโs why at this yearโs LLM hackathon for Chemistry and Materials Science, we built not one, but โจtwoโจ AI agents for molecular dynamics ๐
1
7
17
High-dimensional linear mappings, or linear layers, dominate both the parameter count and inference cost in most deep learning models. We propose a general-purpose drop-in replacement with a substantially better capacity - inference cost ratio. Check it out!๐งต
1
10
34
Psi-k @Psik_Network was intense & inspiring! Great to meet new & old faces (finally in 3D, not just 2D ๐
)! Loved sharing our work on reactive interfaces on the big stage. Hope our ideas help make sampling complex interfaces a bit easier. Soon more to come from @SchwallerGroup!
1
5
30
Out now! @pschwllr, @SchwallerGroup, @loic_roch, @VictorSabanza and colleagues provide guidelines and recommendations for when to use multi-fidelity Bayesian optimization over their single-fidelity counterparts https://t.co/vFsEPABa0Q ๐ https://t.co/mdALeF6amU
nature.com
Nature Computational Science - Multi-fidelity Bayesian optimization methods are studied on molecular and material discovery tasks, and guidelines are provided to recommend cheaper and informative...
1
8
36
Burgenstock #bc_scs25 Editorial Board meeting @rebeccambuller @albrecht_lab @OlallaLab @KatayevL @HariGroupIISc R. Mulvey @EvaHeviaGroup @LACOUR_UNIGE @LcsoLab L. De Luca @HoogendoornLab @michal_juricek @SchwallerGroup F. Gallou and @GasserGroup
0
14
80
Generate property-optimized small molecules with ๐ด๐ต๐ฆ๐ฆ๐ณ๐ข๐ฃ๐ญ๐ฆ ๐ข๐ฏ๐ฅ ๐จ๐ณ๐ข๐ฏ๐ถ๐ญ๐ข๐ณ synthesizability control - allowing complete user-flexibility to impose various reaction constraints! Pre-print: https://t.co/k3XeQ2HaQz Code: https://t.co/5DkPdPZZpE (1/4)
5
19
92
๐๏ธGOLLuM is in the Swiss Alps ๐ฑ but also at the #ICLR2025 World Models workshop! ๐ ๐Peridot 201 & 206โ๐ Apr 28th, 12:00 ๐ https://t.co/hqJuUzT88Q Fun fact: Tolkienโs mountains were Swiss-inspired. ๐งโโ๏ธโจ
1
5
29
Learn how to render multiple molecules from a single multi-frame .xyz file in Blender! ๐งช Hands-on session recording from the AIChemist School: bridging explainable AI and chemistry. @SchwallerGroup Watch here: https://t.co/RjoTcnIdrU More info: https://t.co/xDEjiKtznB
github.com
Useful tips and tricks for chemists in a hurry, trying to get started with Blender 3D. - schwallergroup/blender-aichemist
2
11
38
Quick maths ๐งฎ: In a lab of 25 people from 15+ countries, what's the probability that two share the same birthday โ๐๏ธ 25 Aprilโ and they're both at #ICLR2025 ๐ธ๐ฌ? Drop your estimate below, or just wish @6ojaHa ๐ท๐ธ and @TheoNeukomm ๐จ๐ญ a huge happy birthday! ๐๐๐ฅณ๐๐ฐ
1
2
19
๐ Honored to host Bojana from @SchwallerGroup at GreenDynamics this week for our GD AI Seminar Series! She gave an inspiring talk on reframing Large Language Models as Bayesian optimizers for chemical discovery. ๐ Huge thanks to Bo #AI #LLM #BayesianOptimization #ChemistryAI
1
2
9
๐จNew preprint! โจMy preciousโจ GOLLuM: GP-Optimized LLMs for Bayesian Optimization โ the first of its kind! LLMs as deep kernels in GPs jointly optimized via marginal likelihood โ implicit metric learning, calibrated uncertainty & superior sampling ๐ ๐
arxiv.org
Scientific discovery increasingly depends on efficient experimental optimization to navigate vast design spaces under time and resource constraints. Traditional approaches often require extensive...
4
18
79
AdsMT is finally out in @NatureComms ๐ฅณ It is designed for rapid prediction of global minimum adsorption energy (GMAE) from surface graphs and adsorbate descriptors. Thanks to @XuHuang461675 , @pschwllr and @NCCR_Catalysis! Paper & code: https://t.co/eVW0W8TDtS
@SchwallerGroup
nature.com
Nature Communications - The fast evaluation of global minimum adsorption energy (GMAE) is crucial for catalyst screening. Here, authors designed a multi-modal transformer called AdsMT to rapidly...
1
8
28
Exciting new preprint! ๐งช๐ค Our work shows how LLMs can steer search processes in chemistry! Great collaboration with: @drecmb @TheoNeukomm @d_armstr and @pschwllr in amazing @SchwallerGroup at @EPFL_en Check the demo for steerable retrosynthesis ๐ฟ๐
1
7
22
Check out the updated published version of our pre-print in @ChemicalScience! (1) *General-purpose* generative model + retrosynthesis model = design molecules with optimized properties with an *explicit* predicted synthesis pathway Paper: https://t.co/y0H2wovtlF (1/2)
Molecular generative models can *directly* optimize for synthesizability using retrosynthesis models! Check out initial results which can be an alternative to synthesizability-constrained generation Pre-print: https://t.co/3PFTuhuQRZ Code: https://t.co/dcpziGIL8U (1/2)
5
15
62
๐จ New Publication Alert! ๐จ Our article just got published in MLST๐ We introduce ChemLit-QA, a large, expert-validated dataset designed to benchmark scientific LLMs in chemistry ๐งช๐ค ๐งต๐ @pschwllr @SchwallerGroup
1
7
32
LLMs are pretty bad at writing molecules, but quite good at analyzing mols and reactions! In our new work we use LLMs+search in chemical tasks, unlocking steerable synth. planning and mechanism prediction ๐ 1/ @TheoNeukomm @d_armstr @ZJoncev @pschwllr
https://t.co/O894IlZUOy
3
17
61