Fast-Forward(ing) Models with AI Emulators
summary
Emulator pipeline replaces costly forward runs of the Sorocco radiative transfer code with a neural-network surrogate that shrinks a 30+ parameter space to nine key inputs, trains on 100,000+ spectra sampled via Latin hypercube sampling, compresses 15,000-wavelength outputs with PCA, and yields ~94% reconstruction accuracy within 2σ while cutting per-evaluation time from ~30 minutes to ~1 millisecond, enabling real-time inference and ensemble uncertainty estimates via model ensembles and Otuna-powered hyperparameter search.
AI-written summary · about
chapters (9)
- 0:00Astrophysics and space imaging
- 3:58Accretion discs and stellar winds
- 6:47AI emulators and their use
- 13:11Training data for emulators
- 16:20Parameter reduction for efficiency
- 19:29Building neural network emulators
- 24:46Real-time plasma control
- 31:02Extreme weather prediction
- 35:57Future of emulators
related talks
25mConvergence of AI, Simulations and HPC
10mPruna AI
49mModeling our future: Advancing climate research with AI
40mYou can AI like an Expert
19mThe Small Model Infrastructure Nobody Built (So We Did)
06mMETASAT Demonstrator: Mixed Criticality, Accelerated AI Computing for Future Space Systems
40mUncertain Models - Learning from the past and optimizing for the future
17mPhantora: Maximizing Code Reuse in Simulation-based Machine Learning...