interventional flow matching

This work investigates whether flow matching in temporal generation learns a universal dynamical structure or merely reproduces historical trajectories. By analyzing the empirical flow matching objective under Gaussian conditional paths, we derive—for the first time—a closed-form expression for its optimal velocity field, revealing it to be a similarity-weighted mixture of historical instantaneous velocities. This formulation constitutes a nonparametric, memory-augmented continuous-time dynamical system. Building on this insight, we propose a training-free closed-form sampler that directly generates high-quality probabilistic forecasts from historical transitions. Evaluated on nonlinear dynamical system benchmarks, our method substantially improves sampling efficiency and numerical stability while offering an explicit, interpretable mechanism for data-dependent dynamics.

This work addresses the theoretical gap in sample complexity analysis for flow-matching generative models. Unlike prior studies relying on empirical risk minimization (ERM) assumptions, we establish the first end-to-end upper bound on sample complexity without such assumptions. Methodologically, we model the continuous flow via ordinary differential equations and parameterize the velocity field using neural networks; we then introduce a triple-error decomposition framework—comprising neural approximation error, statistical error, and optimization error—and rigorously analyze its convergence. Our theoretical analysis shows that $O(varepsilon^{-4})$ samples suffice to achieve $O(varepsilon)$ generative accuracy in the Wasserstein-2 distance. This constitutes the first rigorous, non-ERM-dependent sample complexity guarantee for flow matching, filling a critical theoretical void. Moreover, our result provides foundational insights for efficient training and generalization analysis of flow-based generative models.

Generative models such as diffusion models rely on multi-step numerical integration for inference, incurring high computational cost; while consistency models enable one-step generation, they lack a unified theoretical foundation. Method: We propose a novel paradigm that directly learns the flow map between two time points of an ordinary differential equation (ODE), unifying consistency modeling, progressive distillation, and other few-step generation approaches. Leveraging stochastic interpolation, we jointly optimize flow map prediction and velocity field distillation loss, integrating ODE theory with neural operator principles to enable tunable step counts for precision–efficiency trade-offs. Contribution/Results: On CIFAR-10 and ImageNet 32×32, our method achieves 10–50× speedup in sampling over standard diffusion models while preserving competitive sample quality—demonstrating both theoretical coherence and practical efficacy in accelerating generative inference.

This work addresses the modeling and learning of velocity fields governing data distribution evolution in flow matching, aiming to unify transport planning, Markov kernels, and stochastic process paradigms. Methodologically, it establishes the first theoretical equivalence framework for velocity fields characterizing absolutely continuous Wasserstein curves across these three constructions; introduces the conditional Wasserstein distance as a novel metric for Bayesian inverse problems; and unifies the geometric interpretations of continuous normalizing flows and score matching. Leveraging tools from Wasserstein geometry, optimal transport, and stochastic differential equations, the paper rigorously proves the intrinsic consistency of multiple velocity field learning approaches, thereby strengthening the mathematical foundations of flow matching. Experiments demonstrate that the proposed framework effectively generates high-fidelity conditional distributions in Bayesian inverse problems.

This work addresses the challenge of prospectively forecasting patient physiological trajectories under planned interventions—such as insulin administration and carbohydrate intake—by proposing Interventional Flow Matching (IFM), a continuous-time generative model. IFM models glucose dynamics within a bounded latent space, conditioning the velocity field on both historical patient states and future interventions. It introduces a novel solver-free Jacobian regularization that directly constrains the local sensitivity of the velocity field to treatment variables, thereby enforcing physiologically plausible directions and magnitudes of glycemic response (e.g., insulin-induced glucose reduction and carbohydrate-driven elevation). Without relying on explicit differential equations or rollout simulations, IFM achieves an optimal trade-off between observational noise and intervention responsiveness in the UVA/Padova Type 1 diabetes simulator cohort, consistently generating dose–response relationships that align with known physiology in both directionality and ranking consistency.

This work demonstrates that diffusion models, score-based generative models, and flow matching methods—despite their apparent formal differences—share a unified continuous-time generative mechanism. By constructing a measure-theoretic framework, the paper unifies these approaches as learning time-dependent vector fields that transport a reference distribution to the data distribution, with distributional evolution governed by the continuity equation and the Fokker–Planck equation. It establishes, for the first time under a common perspective, the equivalence and distinctions among the three paradigms, clarifies the relationship between probability flow ODEs and stochastic backward dynamics, and identifies flow matching as essentially a velocity field regression problem. The study further provides a systematic comparison of objective functions, sampling strategies, and discretization errors, links the framework to Schrödinger bridges and entropy-regularized optimal transport, and summarizes theoretical guarantees and open challenges regarding approximation capacity, stability, and scalability.

This work addresses the limitations of existing counterfactual distribution estimation methods, which often ignore the intrinsic connections between observed and counterfactual distributions, leading to substantial bias and poor generation quality. To overcome these issues, the authors propose a deconfounded flow matching framework that explicitly models the tight relationships in support sets, tail behaviors, and confounding-invariant features between the two distributions. The key innovations include a semiparametrically efficient estimator based on influence function correction and the first application of minimum energy flows to high-dimensional counterfactual modeling, which simplifies the flow objective and enhances training stability. Experimental results demonstrate that the proposed method significantly outperforms current debiasing approaches and effectively mitigates the failure modes commonly observed in high-dimensional flow-based counterfactual generators.

This work addresses the limitations of existing flow matching methods, which rely on straight-line trajectories and struggle to capture complex dynamical behaviors. By leveraging the principle of least action, the authors generalize the Lagrangian formalism to construct probability paths and associated velocity fields that satisfy both the continuity equation and endpoint constraints. This approach embeds flow matching within a classical mechanics framework, thereby unifying and extending prior methods such as optimal transport and diffusion-based paths. The resulting static variational objective eliminates the need for trajectory simulation and enables direct optimization. Empirical results demonstrate that the proposed method yields physically meaningful dynamical evolutions and achieves performance on par with state-of-the-art conditional flow matching models.

This work proposes a continuous-time generative modeling framework inspired by the Chow–Rashevskii theorem to address the poor parameter efficiency of conventional generative models in high-dimensional spaces, which often struggle to balance expressiveness and scalability. By composing a small set of fixed vector fields with learnable scalar control functions, the method constructs transport maps on manifolds, decoupling model parameters into low-dimensional scalar control channels. This design significantly enhances both parameter efficiency and interpretability. Leveraging Lie algebra–based controllable vector fields, continuous normalizing flows, and ordinary differential equation solvers, the framework enables efficient distribution transformation. Experiments demonstrate that the model accurately fits complex synthetic distributions using only a minimal number of control channels, confirming its strong representational capacity and computational efficiency.