Abstract:We introduce a methodology for performing parameter inference in high-dimensional, non-linear diffusion processes. We illustrate its applicability for obtaining insights into the evolution of and relationships between species, including ancestral state reconstruction. Estimation is performed by utilising score matching to approximate diffusion bridges, which are subsequently used in an importance sampler to estimate log-likelihoods. The entire setup is differentiable, allowing gradient ascent on approximated log-likelihoods. This allows both parameter inference and diffusion mean estimation. This novel, numerically stable, score matching-based parameter inference framework is presented and demonstrated on biological two- and three-dimensional morphometry data.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the difficult problem of parameter inference in high - dimensional, non - linear diffusion processes. Specifically, it targets: 1. **Lack of closed - form likelihood functions**: For non - linear stochastic differential equation models involving hundreds of correlated dimensions, due to the lack of closed - form likelihood functions, traditional parameter estimation methods are difficult to apply. 2. **Poor numerical conditions**: The numerical calculation conditions of these high - dimensional non - linear processes are poor, causing traditional methods to face challenges when dealing with these problems. To solve the above problems, the author proposes a new method, combining score matching in deep learning and parameter estimation techniques in statistics. Specifically, the paper proposes the following innovations: - **A new, numerically stable score - matching objective function**: Allows direct simulation of diffusion bridges on hundreds of correlated dimensions, thereby achieving parameter inference. - **A fully differentiable likelihood estimator**: Uses the simulated diffusion bridges as the proposal distribution in the importance sampler, allowing for simultaneous parameter inference and diffusion mean estimation. - **A series of techniques to bypass numerical instability problems**: Avoids determinant calculations and matrix inversions through multivariate Gaussian approximation, thereby improving numerical stability. ### Application areas This method is particularly suitable for problems in evolutionary biology, such as: - **Modeling morphological feature variation**: By describing the evolution of species' shapes, establish the most likely process from an unknown common ancestor to the phenotypes of extant species (such as the outline of butterfly wings). - **Ancestral state reconstruction**: Estimate the diffusion mean to infer the most likely ancestral state of a species. - **Analysis of species relationships**: Evaluate the similarity and relationships between species by estimating the variance parameters of diffusion bridges between different species. ### Examples The paper shows the effectiveness of this method in two biological applications: 1. **Analysis of species relationships**: By comparing the parietal bone contours of the gray wolf (Canis lupus) and the red fox (Vulpes vulpes), it was found that the variance parameter of the diffusion bridge within the same species is smaller than that between different species, verifying the effectiveness of the method. 2. **Ancestral state reconstruction**: By processing the observational data of six swallowtail butterflies (Papilio genus), their ancestral morphology was successfully reconstructed, and the results showed a trend with swallowtail characteristics. In conclusion, this paper provides a novel and effective parameter inference method that can provide valuable insights in complex, high - dimensional non - linear diffusion processes, especially in the field of evolutionary biology.

Parameter Inference via Differentiable Diffusion Bridge Importance Sampling

Simulating infinite-dimensional nonlinear diffusion bridges

Modeling of 2D Diffusion Processes Based on Microscopy Data: Parameter Estimation and Practical Identifiability Analysis

Parameter Estimation with Increased Precision for Elliptic and Hypo-elliptic Diffusions

Simple simulation of diffusion bridges with application to likelihood inference for diffusions

Score matching for sub-Riemannian bridge sampling

Diffusion Bridge Implicit Models

Improved sampling via learned diffusions

Parameter Inference for Degenerate Diffusion Processes

Statistical algorithms for low-frequency diffusion data: A PDE approach

Diffusion Bridge Mixture Transports, Schrödinger Bridge Problems and Generative Modeling

Variational Markov chain Monte Carlo for Bayesian smoothing of non-linear diffusions

Statistical Spatially Inhomogeneous Diffusion Inference

Diffusion posterior sampling for simulation-based inference in tall data settings

Transport with Support: Data-Conditional Diffusion Bridges

Diffusion Schrödinger Bridge with Applications to Score-Based Generative Modeling

Multifidelity multilevel Monte Carlo to accelerate approximate Bayesian parameter inference for partially observed stochastic processes

Diffusion Schrödinger Bridge Matching

Data augmentation-based statistical inference of diffusion processes

Implicit Diffusion: Efficient Optimization through Stochastic Sampling