A Metaheuristic for Amortized Search in High-Dimensional Parameter Spaces.

Area of Science:

• Computational Science
• Dynamical Systems Modeling
• Bioinformatics

Background:

• Parameter inference for dynamical models is computationally intensive due to intractable gradients and high-dimensional, non-linear spaces.
• Bayesian inference methods focus on parameter distributions rather than point estimates, presenting alternative approaches.
• Existing metaheuristics often require substantial computational resources for effective parameter inference.

Conclusions:

• DR-FFIT offers an efficient and computationally inexpensive approach for parameter inference in complex dynamical systems.
• The method effectively handles high-dimensional and non-linear parameter spaces, outperforming existing metaheuristics.
• This work provides a valuable tool for advancing parameter inference in (bio)physical modeling.