Machine Learning Research
Scaling Bayes
Neural networks are good at making predictions, but they’re not so good at estimating how certain they are. If the training data set is small and many sets of model parameters fit the data well, for instance, the network may not realize this explicitly, leading to overly confident predictions.