Mechanism-AI coupled dynamical systems are increasingly used to study complex biological systems. However, uncertainty associated with noisy, sparse, and partially observed data limits their reliability. This paper presents a systematic mathematical formulation for treating these uncertainties from the perspective of optimal transport and information geometry. We formulate the uncertainty quantification problem as the minimization of an energy functional on a probability density space equipped with a Riemannian metric, with the loss evaluated through a differential equation, and give a unified description of the gradient flows of the energy functional under specific metrics, such as the Wasserstein and Fisher-Rao metrics. Furthermore, we present corresponding numerical methods for solving these gradient flows using abstract adjoint sensitivity analysis. The formulation connects a range of Bayesian inference methods for dynamical systems; numerical experiments compare posterior approximation and predictive uncertainty under a common computational budget. Moreover, it also provides a geometric perspective on dynamical modeling and the study of heterogeneity evolution.
Algorithm 2.1 Particle-Based Constrained Gradient Flow with Adjoint Sensitivities
Figure 1: Conditional last-layer inference from synthetic daily counts. (a) Observations, the true generating mean incidence, and the posterior mean incidence. The shaded band is the central 95% posterior predictive interval, including negative-binomial observation noise. (b) The true effective reproduction number Rt , its posterior mean, and its pointwise 95% conditional credible band. The horizontal dotted line marks Rt=1 . Filled and open circles denote training and held-out observations, respectively; the grey background marks days 87–100, excluded from both network fitting and posterior inference. All posterior summaries use the independently initialized, longer MALA reference run and condition on the fitted hidden layers, initial state, γ , and κ .
Figure 2: Conditional last-layer inference from Ontario daily reported cases, February 25–July 25, 2020. (a) Reported daily counts, posterior mean incidence, and the central 95% posterior predictive interval, including negative-binomial observation noise. (b) The model-inferred effective reproduction number: posterior mean Rt and pointwise 95% conditional credible band. The horizontal dotted line marks Rt=1 . Filled and open circles denote training and held-out observations; the grey background marks July 12–25, excluded from both network fitting and posterior inference. No ground-truth reproduction-number curve is available. All posterior summaries use the independently initialized, longer MALA reference run and condition on the fitted hidden layers, initial state, γ , and κ .
Figure 3: Accuracy of conditional-posterior approximation at increasing computational budgets: (a) synthetic data and (b) Ontario data. The vertical axis is the sliced W2 distance to the empirical distribution from an independently initialized, longer MALA reference run, evaluated in the common nine-dimensional affine coordinates. The horizontal axis counts full likelihood-gradient evaluations at individual parameter vectors. Curves show means and coloured bands show ± one sample standard deviation across three algorithmic seeds on the same dataset. MALA uses all retained post-warmup states at each budget; SVGD and ULA use their 64 current particles. The horizontal grey band is the central 95% range of distances from 100 independently resampled 64-point reference sets to the full empirical reference. It measures finite-ensemble sampling variation, not posterior uncertainty or uncertainty in an algorithm’s mean curve. Comparison budgets include initialization and MALA warmup but exclude common preprocessing and the reference run.
Dataset
Method
Sliced W2
Mean error
Covariance error
Ontario
MALA
0.022±0.003
0.013±0.004
0.037±0.003
Ontario
SVGD
0.172±0.012
0.054±0.009
0.274±0.027
Ontario
ULA
0.215±0.035
0.138±0.040
0.338±0.024
Synthetic
MALA
0.021±0.001
0.013±0.001
0.036±0.001
Synthetic
SVGD
0.175±0.014
0.055±0.008
0.275±0.027
Synthetic
ULA
0.215±0.034
0.140±0.039
0.338±0.027
Table 1: Errors relative to the independent conditional-posterior reference at 32,768 likelihood-gradient evaluations per run. Values are means ± sample standard deviations over three algorithmic seeds. The mean and covariance errors are defined in the text. MALA uses its 28,764 retained states; SVGD and ULA each use 64 terminal particles.
Figure 4: Selection and drift–diffusion in the trait-structured SIR model. (a,b) The normalized trait density among infected individuals, ρ(t,a)=I(t,a)/Iˉ(t) , under selection only and selection plus drift–diffusion, respectively. The heatmaps share a density scale; white curves mark the mean trait aˉ(t)=∫Aaρ(t,a)da . (c) The corresponding mean traits aˉ(t) within the infected population. (d) Total infected fractions Iˉ(t)/N . In the combined case, trait transport consists of the drift −U′(a) and diffusion with coefficient σ2/2 , with zero flux at the trait boundaries; both terms are absent in the selection-only case. Transmission and recovery rates and initial conditions are identical. All curves are deterministic numerical solutions, with no statistical uncertainty bands.
The distinction between aleatoric and epistemic uncertainty has received considerable attention in machine learning research, mainly in the context of supervised learning but also in other settings such as generative modeling. In this paper, we offer a machine learning perspective on uncertainty modeling for dynamical systems, which has been studied much less so far. In particular, we ask: what uncertainties do we need for dynamical systems? We discuss sources of uncertainty, clarify their nature (aleatoric or epistemic), and consider how the objectives of representing and quantifying uncertainty vary across different tasks.
Yusuf Sale, Christopher Bülte, Felix Czaja +2
Institute of Computer Science, LMU Munich · Munich · Department of Mathematics, LMU Munich +1
A novel advective Fisher-Rao metric is introduced for optimization tasks on paths of probability measures governed by the continuity equation. This metric is shown to lead to optimal descent directions. It is then shown that this metric arises naturally from three different perspectives: As the rescaled zero-noise limit of the Fisher-Rao metric on path measures, as the expected value of the second variation of the Freidlin--Wentzell large deviation rate functional, and as the Hessian of the Benamou--Brenier action functional from dynamic optimal transport. We supplement this geometric construction with computational experiments. Here, we demonstrate empirically that the advective Fisher-Rao metric yields the desired optimal fitting of probability densities, whereas the Gauss--Newton method yields optimal fitting of velocity fields.
Benjamin Gess, Johannes Müller
Institut für Mathematik, Technische Universität Berlin & Max-Planck-Institut für Mathematik in den Naturwissenschaften · Institut für Mathematik, Technische Universität Berlin
Many scientific systems, such as cellular populations or economic cohorts, are naturally described by probability distributions that evolve over time. Predicting how such a system would have evolved under different forces or initial conditions is fundamental to causal inference, domain adaptation, and counterfactual prediction. However, the space of distributions often lacks the vector space structure on which classical methods rely. To address this, we introduce a general notion of parallel dynamics at a distributional level. We base this principle on parallel transport of tangent dynamics along optimal transport geodesics and call it ``Wasserstein Parallel Trends''. By replacing the vector subtraction of classic methods with geodesic parallel transport, we can provide counterfactual comparisons of distributional dynamics in applications such as causal inference, domain adaptation, and batch-effect correction in experimental settings. The main mathematical contribution is a novel notion of fanning scheme on the Wasserstein manifold that allows us to efficiently approximate parallel transport along geodesics while also providing the first theoretical guarantees for parallel transport in the Wasserstein space. We also show that Wasserstein Parallel Trends recovers the classic parallel trends assumption for averages as a special case and derive closed-form parallel transport for Gaussian measures. We deploy the method on synthetic data and two single-cell RNA sequencing datasets to impute gene-expression dynamics across biological systems.
Tristan Luca Saidi, Gonzalo Mena, Larry Wasserman +1
Department of Statistics and Data Science, Carnegie Mellon University · Machine Learning Department, Carnegie Mellon University · Department of Economics, Emory University