sisl / nns-as-gps

Compares neural network ensembles and Monte Carlo dropout in their ability to mimic the behavior of a Gaussian process.

nns-as-gps

Comparison of neural network ensembles and Monte Carlo dropout in their ability to mimic the behavior of a Gaussian process.

The ensemble methods implemented are:

1. A vanilla ensemble of neural networks, each predicting a single value, and trained using MSE loss.
2. An ensemble of neural networks that each predict a mean and variance, trained with NLL loss, as advised by this paper.
3. Monte-Carlo dropout, as described in this paper

See also:

1. Another implementation of ensembles with predictive uncertainty here

Author : Kunal Menda