1/ A grand challenge in science: the simulation-to-experiment gap Simulators model complex systems but introduce model errors; experiments give real-world data but are costly and noisy Can we leverage both? Yes! We introduce Adversarial Distribution Alignment (ADA), at #NeurIPS2026 ADA aligns generative models—initially trained on simulations (e.g., molecular dynamics... with classical force fields)—to full real-world distributions (e.g., protein structures) using only partial, noisy observations (like cryo-EM images) ADA goes beyond just matching the expectation of observables: it can match full distributions, even in the presence of multiple correlated, high-dimensional observations Check out our website: And our paper: In collaboration with kai, Toby Kreiman, and Sergey Levineshow more

2/ To use both sources of data, we frame experimental observable alignment as a distributional constraint. If a model is well aligned, it should match the full distribution of real-world observables

3/ We satisfy the observable constraint using Wasserstein critics and apply an entropic regularizer to stay close to the simulator reference. This provides a principled framework for matching the distribution of partial observables in the absence of data from the full joint distribution Note that this is different from conditional generative modeling, as this framework can naturally handle multiple observables!

4/ ADA can handle high-dimensional, noisy observables. In this example, we start from a classical force field (inducing a Boltzmann distribution) and shift the distribution toward experimentally observed protein structures (from the PDB)—using cryo-EM image observables Despite observing only indirect, noisy images, ADA recovers meaningful structural corrections in the underlying distribution

5/ Prior works often consider only expectation alignment (EA), matching only the means of observables of interest. But matching finitely many expectations does not recover the correct distribution. ADA instead aligns to the full distribution of experimental outcomes. Even in a simple Gaussian mixture setting, EA fails to correct the energy landscape, while ADA successfully captures the true energy landscape.

6/ In the future, we envision models that use fully observed data from simulators as priors before leveraging numerous “real-world” experimental observables, spanning length and timescales, to accurately model the real world.

Closing the sim-to-experiment gap by aligning full distributions from partial noisy observations is the right problem. Matching expectations alone leaves too many modes unchecked when cryo-EM and MD disagree. Curious how ADA holds when observation channels are correlated but incomplete.

As an MSc student running short GROMACS sims, the sim-to-experiment gap is exactly what bugs me. How much experimental data did ADA need before the protein ensemble moved meaningfully away from the force-field prior? Could NMR observables like J-couplings plug in too?

how are you weighting the entropic regularizer? seems like the knob that decides whether noisy cryo-EM images pull the ensemble somewhere unphysical

simulators are always off by just enough to make me question every lab result i ever ran

Can we design some kind of autolab？