Corpus ID: 222208710

Prediction intervals for Deep Neural Networks

@article{Mancini2020PredictionIF,
  title={Prediction intervals for Deep Neural Networks},
  author={Tullio Mancini and Hector Calvo-Pardo and J. Olmo},
  journal={ArXiv},
  year={2020},
  volume={abs/2010.04044}
}
  • Tullio Mancini, Hector Calvo-Pardo, J. Olmo
  • Published 2020
  • Computer Science, Mathematics, Economics
  • ArXiv
    • The aim of this paper is to propose a suitable method for constructing prediction intervals for the output of neural network models. To do this, we adapt the extremely randomized trees method originally developed for random forests to construct ensembles of neural networks. The extra-randomness introduced in the ensemble reduces the variance of the predictions and yields gains in out-of-sample accuracy. An extensive Monte Carlo simulation exercise shows the good performance of this novel method… CONTINUE READING
    View PDF on arXiv

    Figures and Tables from this paper

    Figures and Tables

    References

    SHOWING 1-10 OF 56 REFERENCES
    Prediction intervals for neural networks via nonlinear regression
    • 160
    • Highly Influential
    • PDF
    Optimal deep neural networks by maximization of the approximation power
    • 1
    Practical Confidence and Prediction Intervals
    • 295
    • PDF
    Confidence and prediction intervals for neural network ensembles
    • 83
    • Highly Influential
    Probabilistic Forecasting Using Monte Carlo Dropout Neural Networks
    • 3
    High-Quality Prediction Intervals for Deep Learning: A Distribution-Free, Ensembled Approach
    • 55
    • Highly Influential
    • PDF
    An empirical analysis of dropout in piecewise linear networks
    • 72
    • PDF
    Prediction Intervals for Artificial Neural Networks
    • 250
    • Highly Influential
    Why M Heads are Better than One: Training a Diverse Ensemble of Deep Networks
    • 112
    • Highly Influential
    • PDF
    Bayesian Convolutional Neural Networks with Bernoulli Approximate Variational Inference
    • 363
    • Highly Influential
    • PDF