Corpus ID: 210920296

Improper Learning for Non-Stochastic Control

@article{Simchowitz2020ImproperLF,
  title={Improper Learning for Non-Stochastic Control},
  author={Max Simchowitz and Karan Singh and Elad Hazan},
  journal={ArXiv},
  year={2020},
  volume={abs/2001.09254}
}
  • Max Simchowitz, Karan Singh, Elad Hazan
  • Published in ArXiv 2020
  • Computer Science, Mathematics
  • We consider the problem of controlling a possibly unknown linear dynamical system with adversarial perturbations, adversarially chosen convex loss functions, and partially observed states, known as non-stochastic control. We introduce a controller parametrization based on the denoised observations, and prove that applying online gradient descent to this parametrization yields a new controller which attains sublinear regret vs. a large class of closed-loop policies. In the fully-adversarial… CONTINUE READING

    Create an AI-powered research feed to stay up to date with new papers like this posted to ArXiv

    Citations

    Publications citing this paper.
    SHOWING 1-2 OF 2 CITATIONS

    References

    Publications referenced by this paper.
    SHOWING 1-10 OF 42 REFERENCES

    Regret Bounds for the Adaptive Control of Linear Quadratic Systems

    VIEW 11 EXCERPTS
    HIGHLY INFLUENTIAL

    Modern Wiener-Hopf Design of Optimal Controllers. Part I

    VIEW 4 EXCERPTS
    HIGHLY INFLUENTIAL

    Naive Exploration is Optimal for Online LQR

    VIEW 3 EXCERPTS

    System Identification I

    VIEW 3 EXCERPTS
    HIGHLY INFLUENTIAL

    2019] develop error bounds in terms of the operator norm of ∆: Lemma D.4 (Simplification

    • N . Simchowitz
    • Simchowitz et al
    • 2019

    2019b] poly log T

    • Agarwal
    • 2019

    Certainty Equivalent Control of LQR is Efficient

    VIEW 2 EXCERPTS