English

Adversarial training for predictive tasks: theoretical analysis and limitations in the deterministic case

Machine Learning 2020-12-01 v3 Machine Learning

Abstract

To train a deep neural network to mimic the outcomes of processing sequences, a version of Conditional Generalized Adversarial Network (CGAN) can be used. It has been observed by others that CGAN can help to improve the results even for deterministic sequences, where only one output is associated with the processing of a given input. Surprisingly, our CGAN-based tests on deterministic geophysical processing sequences did not produce a real improvement compared to the use of an LpL_p loss; we here propose a first theoretical explanation why. Our analysis goes from the non-deterministic case to the deterministic one. It led us to develop an adversarial way to train a content loss that gave better results on our data.

Keywords

Cite

@article{arxiv.2011.00835,
  title  = {Adversarial training for predictive tasks: theoretical analysis and limitations in the deterministic case},
  author = {Thibault Lesieur and Jérémie Messud and Issa Hammoud and Hanyuan Peng and Céline Lacombe and Paulien Jeunesse},
  journal= {arXiv preprint arXiv:2011.00835},
  year   = {2020}
}

Comments

NeurIPS 2020, ICBINB Workshop

R2 v1 2026-06-23T19:50:22.359Z