Corpus ID: 231846402

Provable Model-based Nonlinear Bandit and Reinforcement Learning: Shelve Optimism, Embrace Virtual Curvature

@article{Dong2021ProvableMN,
  title={Provable Model-based Nonlinear Bandit and Reinforcement Learning: Shelve Optimism, Embrace Virtual Curvature},
  author={Kefan Dong and Jia-Qi Yang and Tengyu Ma},
  journal={ArXiv},
  year={2021},
  volume={abs/2102.04168}
}

Kefan Dong, Jia-Qi Yang, Tengyu Ma

Published 2021

Computer Science

ArXiv

This paper studies model-based bandit and reinforcement learning (RL) with nonlinear function approximations. We propose to study convergence to approximate local maxima because we show that global convergence is statistically intractable even for one-layer neural net bandit with a deterministic reward. For both nonlinear bandit and RL, the paper presents a model-based algorithm, Virtual Ascent with Online Model Learner (ViOL), which provably converges to a local maximum with sample complexity… CONTINUE READING

View PDF on arXiv

Save to Library

Create Alert

Cite

Launch Research Feed