At the intersection of extreme value theory and neural networks — When can we make neural nets model rare extremes well?
This is a subtle problem. Extremal values are hard to estimate in general for a bunch of obvious and less obvious reasons.
The risks you care about you might only see once. The distributions of the risk might be correlated in difficult ways, even spatially in ways that are challenging for neural nets even before we worry about the classical risk management.
But! Apparently, there are things that can be done!
TBC
1 References
Boulaguiem, Zscheischler, Vignotto, et al. 2022. “Modeling and Simulating Spatial Extremes by Combining Extreme Value Theory with Generative Adversarial Networks.” Environmental Data Science.
Lafon, Naveau, and Fablet. 2023. “A VAE Approach to Sample Multivariate Extremes.”
Maceda, Hector, Lenzi, et al. 2024. “A Variational Neural Bayes Framework for Inference on Intractable Posterior Distributions.”
Majumder, and Reich. 2023. “A Deep Learning Synthetic Likelihood Approximation of a Non-Stationary Spatial Model for Extreme Streamflow Forecasting.” Spatial Statistics.
Pasche, and Engelke. 2024. “Neural Networks for Extreme Quantile Regression with an Application to Forecasting of Flood Risk.” The Annals of Applied Statistics.
Richards, and Huser. 2024. “Regression Modelling of Spatiotemporal Extreme U.S. Wildfires via Partially-Interpretable Neural Networks.”
Richards, and Huser. n.d. “A Unifying Partially-Interpretable Framework for Neural Network-Based Extreme Quantile Regression.”
Richards, Sainsbury-Dale, Zammit-Mangion, et al. 2024. “Neural Bayes Estimators for Censored Inference with Peaks-over-Threshold Models.”
Richards, Tawn, and Brown. 2022. “Modelling Extremes of Spatial Aggregates of Precipitation Using Conditional Methods.” The Annals of Applied Statistics.
Walchessen, Lenzi, and Kuusela. 2024. “Neural Likelihood Surfaces for Spatial Processes with Computationally Intensive or Intractable Likelihoods.” Spatial Statistics.
Zhang, Ma, Wikle, et al. 2024. “Flexible and Efficient Spatial Extremes Emulation via Variational Autoencoders.”