Searcharxiv⌕ Search

arXiv subjects

Torkel E Loman

Publications and source records attributed to Torkel E Loman.

4 recordsLinked to original sources

Structural functional identifiability of unknown differential equation observation maps

In many differential equation models of real-world systems, the mechanistic structure of the dynamics is known, but the map from the dynamics to the observations is not. This is potentially problematic, as it can prevent us from e.g. learning model parameter values from data. Examples where this occurs include measuring a fluorescent marker as a proxy for the dynamics of a cellular circuit, and using wastewater measurements to follow the development of an epidemic. The nascent field of hybrid modelling, in which machine learning components (like neural networks) are embedded within mechanistic models, provides the tools to learn such maps. Here, the system's dynamics keep their mechanistic form, while a neural network represents the unknown observation map and is trained alongside any unknown core system parameters. This raises a question of identifiability: when can the true observation map be learnt, and when does one simply learn some map that transforms the dynamics into the available data? Building on previous work, we show how structural functional identifiability analysis can establish whether a model's unknown observation maps and core system parameters are identifiable, i.e. can be uniquely learnt from data. We both prove general results on when models with unknown observation maps are, and are not, identifiable, and demonstrate a general methodology (based on differential algebra) for assessing the identifiability of such models. This paves the way for applying classic differential equation workflows to the many real-world systems that contain such unknown maps.

math.DS↗

Structural functional identifiability and model discovery in differential equation models

Differential equation models are widely used to describe, interpret, and predict dynamical phenomena across science and engineering. In practice, however, the governing dynamics are rarely fully known and must be inferred from observational data. Traditionally, inverse problems in differential equation modelling have focused on estimating unknown parameter values. In this setting, structural identifiability determines whether parameter values can, in principle, be uniquely recovered from ideal observations and is, therefore, a prerequisite for meaningful inference. More recently, the integration of machine learning with mechanistic modelling has enabled the discovery of unknown equations, functions, and constitutive relationships, substantially expanding the space of admissible models. This raises a fundamental question: under what conditions can unknown functional components be uniquely recovered from data? In this paper, we generalise the classical notion of structural parameter identifiability to functional identifiability. We first identify broad classes of models for which unique functional recovery is impossible. We then show how functional identifiability can be assessed for differential equation models using differential algebra-based techniques which are well-established as a means of assessing structural identifiability for ordinary differential equation models. Our framework reveals new phenomena that arise in the transition from parametric to function inference and have no analogue in the classical setting. Finally, we characterise functional identifiability in several common model classes. Taken together, our results demonstrate that functional identifiability provides a theoretical foundation for modern inverse problems in differential equation modelling, particularly those that use machine learning representations of unknown system components.

math.ST↗

Functional and parametric identifiability for universal differential equations

Mechanistic knowledge of a studied system can be encoded as predictive mathematical models, but this raises the question of what to do when knowledge is uncertain or incomplete. Hybrid mechanistic/data-driven modelling, where unknown system dynamics are learnt from data using machine learning components, is a potential solution. This work focuses on one such approach, universal differential equations (UDEs), and asks when these can learn unknown dynamics from data. We investigate this question using identifiability. We show that UDE identifiability decomposes into parametric and functional identifiability, which concern the model's mechanistic and data-driven parts, respectively. We consider these two types of identifiability along a second, orthogonal, division: structural and practical identifiability, which distinguish ambiguity inherent to the model formulation from ambiguity arising under finite, noisy observations. Specifically, we consider the problem of how to assess practical identifiability for UDEs. As a modelling setting, we use ordinary differential equations derived from chemical reaction network models. Such models arise naturally across systems biology, chemistry, pharmacology, epidemiology, and population dynamics, making them a broadly relevant test case. Through a wide range of examples, we show that chemical reaction networks are a promising setting for UDEs, yielding learnt dynamics that are highly interpretable. We then examine how model properties---including the number of neural networks, the constraints imposed on them, and data availability---affect identifiability. A key finding is that generalising a fully mechanistic model to a UDE can have remarkably little impact on its identifiability. Our results provide strong evidence that UDEs offer an intuitive means of integrating data-driven elements into mechanistic models while preserving predictive capability.

math.DS↗

Exact identifiability analysis for a class of partially observed near-linear stochastic differential equation models

Stochasticity plays a key role in many biological systems, necessitating the calibration of stochastic mathematical models to interpret associated data. For model parameters to be estimated reliably, it is typically the case that they must be structurally identifiable. Yet, while theory underlying structural identifiability analysis for deterministic differential equation models is highly developed, there are currently no tools for the general assessment of stochastic models. In this work, we present a differential algebra-based framework for the structural identifiability analysis of linear and a class of near-linear partially observed stochastic differential equation (SDE) models. Our framework is based on a deterministic recurrence relation that describes the dynamics of the statistical moments of the system of SDEs. From this relation, we iteratively form a series of necessarily satisfied equations involving only the observed moments, from which we are able to establish structurally identifiable parameter combinations. We demonstrate our framework for a suite of linear (two- and $n$-dimensional) and non-linear (two-dimensional) models. Most importantly, we define the notion of structural identifiability for SDE models and establish the effect of the initial condition on identifiability. We conclude with a discussion on the applicability and limitations of our approach, and potential future research directions in this understudied area.

stat.ME↗