Transfer learning for multifidelity simulation-based inference in cosmology
This addresses the problem of expensive training data for researchers in cosmology, though it is incremental as it builds on existing transfer learning methods.
The paper tackled the high computational cost of simulation-based inference in cosmology by using multifidelity transfer learning, which reduced the required number of high-fidelity simulations by a factor of 8 to 15 while maintaining performance.
Simulation-based inference (SBI) enables cosmological parameter estimation when closed-form likelihoods or models are unavailable. However, SBI relies on machine learning for neural compression and density estimation. This requires large training datasets which are prohibitively expensive for high-quality simulations. We overcome this limitation with multifidelity transfer learning, combining less expensive, lower-fidelity simulations with a limited number of high-fidelity simulations. We demonstrate our methodology on dark matter density maps from two separate simulation suites in the hydrodynamical CAMELS Multifield Dataset. Pre-training on dark-matter-only $N$-body simulations reduces the required number of high-fidelity hydrodynamical simulations by a factor between $8$ and $15$, depending on the model complexity, posterior dimensionality, and performance metrics used. By leveraging cheaper simulations, our approach enables performant and accurate inference on high-fidelity models while substantially reducing computational costs.