Anytime-Valid Confidence Sequences for Consistent Uncertainty Estimation in Early-Exit Neural Networks.
Metod JazbecPatrick ForréStephan MandtDan ZhangEric T. NalisnickPublished in: CoRR (2023)
Keyphrases
- neural network
- hidden markov models
- high confidence
- measurement error
- confidence measure
- robust estimation
- estimation algorithm
- genetic algorithm
- confidence values
- multilayer perceptron
- back propagation
- recurrent neural networks
- conditional probabilities
- confidence intervals
- network architecture
- sequential patterns
- decision theory
- rule extraction
- parameter estimation
- expert systems
- pattern recognition