Conformal inference is (almost) free for neural networks trained with early stopping.

Ziyi Liang Yanfei Zhou Matteo Sesia

Published in: CoRR (2023)

Keyphrases

early stopping
neural network
training process
artificial neural networks
bayesian inference
support vector machine
bayesian networks
stochastic gradient descent
training set
least squares
linear combination
prediction accuracy
loss function
monte carlo
classification accuracy
cost function
active learning