CNN / ResNet + Heat2D UQ¶

Notebook: CNN_ResNet_Heat2D_UQ_Tutorial.ipynb

This tutorial trains CNNRegressor2D and ResNetRegressor2D on a 2D heat source-to-solution problem. MC Dropout is the primary uncertainty path, with an optional ensemble comparison for stronger calibration.

Primary references:

• He et al. (2016), Deep Residual Learning for Image Recognition
• Lakshminarayanan et al. (2017), Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles