Abstract:Model reduction is the construction of simple yet predictive descriptions of the dynamics of many-body systems in terms of a few relevant variables. A prerequisite to model reduction is the identification of these relevant variables, a task for which no general method exists. Here, we develop a systematic approach based on the information bottleneck to identify the relevant variables, defined as those most predictive of the future. We elucidate analytically the relation between these relevant variables and the eigenfunctions of the transfer operator describing the dynamics. Further, we show that in the limit of high compression, the relevant variables are directly determined by the slowest-decaying eigenfunctions. Our information-based approach indicates when to optimally stop increasing the complexity of the reduced model. Furthermore, it provides a firm foundation to construct interpretable deep learning tools that perform model reduction. We illustrate how these tools work in practice by considering uncurated videos of atmospheric flows from which our algorithms automatically extract the dominant slow collective variables, as well as experimental videos of cyanobacteria colonies in which we discover an emergent synchronization order parameter.

Minimum Conditional Information Loss Approach to Discrete Stochastic Model Reduction

A Minimum Output Information Loss Method for Stochastic Model Reduction

Information theoretic methods for stochastic model reduction based on state projection

Stochastic Model Reduction by Maximizing Independence

Stochastic Model Reduction Via Minimum Information Loss Approach In Frequency-Domain

State-dimension Reduction and Filtering for Linear Systems under Communication Constraints

A Revised Minimum Information Loss Method for Discrete-time Model Reduction

Reduced-order LQG controller design based on principle of minimum information loss

Model Reduction by Minimizing Information Loss Based on Cross-Gramian Matrix

Minimum information loss method based on cross-gramian matrix for model reduction (CGMIL)

Novel algorithm for attribute reduction based on mutual-information gain ratio

H∞ Model Reduction for Linear Differential Repetitive Processes

Model Reduction of Markovian Jump Systems with Uncertain Probabilities

H∞ Model Reduction for Discrete Singular Markovian Jump Systems

Nonsynchronous Model Reduction for Uncertain 2-D Markov Jump Systems.

Information-Preserving Markov Aggregation

Minimum Cost Attribute Reduction in Incomplete Systems under Decision-Theoretic Rough Set Model

Information theory for data-driven model reduction in physics and biology

Formal Error Bounds for the State Space Reduction of Markov Chains

Exact model reduction for discrete-time conditional quantum dynamics