Machine Learning in Science

The Macke lab is back at @bernsteinneuro.bsky.social in Frankfurt! 🧠🪰 This year with 2 satellite workshop talks + 5 posters on mechanistic, connectome-constrained models of the fruit fly. All posters are joint work with @srinituraga.bsky.social. Come and say hi! 👋 Details 👇 (1/8)

🪧 Poster III-55 | Wed 12:30 Isaac Omolayo: "Multi-task training of connectome-constrained models of the full fly optic lobe" (7/8)

The Macke lab is back at @bernsteinneuro.bsky.social in Frankfurt! 🧠🪰 This year with 2 satellite workshop talks + 5 posters on mechanistic, connectome-constrained models of the fruit fly. All posters are joint work with @srinituraga.bsky.social. Come and say hi! 👋 Details 👇 (1/8)

🪧Poster II-59 | Tue 18:00 @lulmer.bsky.social: "Neural activity constraints improve connectome- and task-constrained models of the fruit fly optic lobe" (5/8)

The Macke lab is back at @bernsteinneuro.bsky.social in Frankfurt! 🧠🪰 This year with 2 satellite workshop talks + 5 posters on mechanistic, connectome-constrained models of the fruit fly. All posters are joint work with @srinituraga.bsky.social. Come and say hi! 👋 Details 👇 (1/8)

🎤 Tue 10:30 | Satellite workshop "Advances in optimization of biologically constrained models and how to use them" @jakhmack.bsky.social: "Simulation-based inference: Progress, promise, open problems" (3/8)

The Macke lab is back at @bernsteinneuro.bsky.social in Frankfurt! 🧠🪰 This year with 2 satellite workshop talks + 5 posters on mechanistic, connectome-constrained models of the fruit fly. All posters are joint work with @srinituraga.bsky.social. Come and say hi! 👋 Details 👇 (1/8)

Mackelab is at @icmlconf.bsky.social in Seoul 🇰🇷 Two papers on simulation-based and model-based scientific discovery this year. Come say hi!

PRISM: Model inference at scale. We infer joint posteriors over model structure & parameters, with test-time control of complexity via a tunable prior. 📄 arxiv.org/abs/2603.15292 by Manuel Gloeckler, @josepemanzano.bsky.social, @mristam.bsky.social, @coschroeder.bsky.social, @jakhmack.bsky.social

New paper: We recast automated scientific model discovery with LLMs as Bayesian inference! LLMs write code and carry domain knowledge, great for proposing models. The key idea: discovery is inference, not just generation. What distribution of models explains the data? 🧵 arxiv.org/abs/2602.18266

Casting discovery as inference over programs lets scientists: • weigh competing mechanistic hypotheses • quantify uncertainty and non-identifiability • reuse established SMC methods for principled posterior exploration • unify proposal, refinement, and selection.

New paper: We recast automated scientific model discovery with LLMs as Bayesian inference! LLMs write code and carry domain knowledge, great for proposing models. The key idea: discovery is inference, not just generation. What distribution of models explains the data? 🧵 arxiv.org/abs/2602.18266

We treat Hodgkin-Huxley models as a structural problem: infer missing ion-channel mechanism. ModelSMC uncovers mechanistically meaningful extensions, exposing structural ambiguities and improving fit, yielding a posterior over plausible models rather than a single best model.

New paper: We recast automated scientific model discovery with LLMs as Bayesian inference! LLMs write code and carry domain knowledge, great for proposing models. The key idea: discovery is inference, not just generation. What distribution of models explains the data? 🧵 arxiv.org/abs/2602.18266

As one concrete instantiation of this framework, we introduce ModelSMC: Sequential Monte Carlo directly in model space. We keep a population of candidate programs as particles, propose refinements via the LLM, weight by likelihood, and resample.

New paper: We recast automated scientific model discovery with LLMs as Bayesian inference! LLMs write code and carry domain knowledge, great for proposing models. The key idea: discovery is inference, not just generation. What distribution of models explains the data? 🧵 arxiv.org/abs/2602.18266

Most LLM discovery systems are agentic loops: propose, evaluate, keep the best one. This gives a single point estimate, discarding the probabilistic structure scientists rely on to know what to rule out, what is uncertain, and where more data would help. Instead, we frame discovery as inference.

@mackelab.bsky.social is at @cosynemeeting.bsky.social #cosyne2026 in Lisbon with two posters presented by PhD students from the lab. Thread below on the projects 👇

Friday, 13:15 (Poster 2-040): @lulmer.bsky.social presents “Neural activity constraints improve task-optimized connectome-constrained models” (joint work with @srinituraga.bsky.social).

The PhD graduation streak continues with Dr. Janne Lappalainen (@lappalainenjk.bsky.social) and Dr. Auguste Schulz (@auschulz.bsky.social) successfully defending in January and February. Congratulations!

Janne developed deep mechanistic networks to study when and how detailed measurements of brain wiring (connectomes) can enable accurate, neuron-level predictions of neural dynamics across the brain (www.nature.com/articles/s41...).

Happy 2026 everyone! Two freshly minted PhDs 🧑‍🎓emerged from our lab at the end of last year. We congratulate Dr Julius Vetter (@vetterj.bsky.social) and Dr Guy Moss (@gmoss13.bsky.social)! Here seen celebrating with the lab 🎳. 1/3

Julius Vetter (@vetterj.bsky.social) worked on deep generative modeling and simulation-based Bayesian inference, with applications to (physiological) time series data. E.g., openreview.net/forum?id=kN0... 2/3

Our group is at NeurIPS and EurIPS this year with four papers and one workshop poster. If you are either curious about SBI with autoML, with foundation models, or on function spaces or about differentiable simulators with Jaxley, have a look below 👇 1/11

Fifth, we bring AutoML to SBI pipelines with a practical performance metric that does not require ground-truth posteriors, improving inference quality on the SBI benchmark! By @swagatam.bsky.social, @gmoss13.bsky.social, @keggensperger.bsky.social, @jakhmack.bsky.social 10/11

Our group is at NeurIPS and EurIPS this year with four papers and one workshop poster. If you are either curious about SBI with autoML, with foundation models, or on function spaces or about differentiable simulators with Jaxley, have a look below 👇 1/11

Fourth, in collaboration with Ian C Tanoh and Scott Linderman, we used the Jaxley toolbox and extended Kalman filters to estimate the marginal log-likelihood of a biophysical neuron model. We showed that this enables identifying biophysical parameters given extracellular recordings. 8/11

Our group is at NeurIPS and EurIPS this year with four papers and one workshop poster. If you are either curious about SBI with autoML, with foundation models, or on function spaces or about differentiable simulators with Jaxley, have a look below 👇 1/11

Third, in collaboration with @kyrakadhim.bsky.social, @philipp.hertie.ai, and others, we built a task- and data-constrained biophysical network of the outer plexiform layer of the mouse retina. To optimize this model, we built it on top of our Jaxley toolbox for differentiable simulation. 6/11

Our group is at NeurIPS and EurIPS this year with four papers and one workshop poster. If you are either curious about SBI with autoML, with foundation models, or on function spaces or about differentiable simulators with Jaxley, have a look below 👇 1/11

Second, come by to check out NPE-PFN: We leverage the power of tabular foundation models for training-free and simulation-efficient SBI. SBI has never been so effortless! By @vetterj.bsky.social, Manuel Gloeckler, @danielged.bsky.social, @jakhmack.bsky.social 4/11

Our group is at NeurIPS and EurIPS this year with four papers and one workshop poster. If you are either curious about SBI with autoML, with foundation models, or on function spaces or about differentiable simulators with Jaxley, have a look below 👇 1/11

First, we introduce FNOPE, a new simulation-based inference approach for efficiently and flexibly inferring function-valued parameters. By @gmoss13.bsky.social, @leahsmuhle.bsky.social, Reinhard Drews, @jakhmack.bsky.social and @coschroeder.bsky.social 2/11