Score Function Gradient Estimation

Score Function Gradient Estimation is a gradient estimation method in variational inference that leverages the score function—the gradient of the log-density of the variational distribution—to compute unbiased but high-variance gradients of the evidence lower bound with respect to distributional parameters by differentiating through the sampling process via the likelihood ratio method.

31

Publications

2K

Citations

89

Authors

37

Institutions

Publications per year

2017–2026

19

Variants

Likelihood Ratio Gradient Estimation

Authors

89

  1. RY
    Reuven Y. Rubinstein

    Technion – Israel Institute of Technology, Israel

    2 publications · 99 citations

    2

    h-index

  2. RE
    Richard E. Turner

    University of Cambridge, United Kingdom

    2 publications · 56 citations

    2

    h-index

  3. BK
    Beomsu Kim

    Seoul National University, South Korea

    2 publications · 77 citations

    2

    h-index

  4. JS
    Junghoon Seo

    Korea Information System (South Korea), South Korea

    2 publications · 77 citations

    2

    h-index

  5. JK
    Jamyoung Koo

    Xi'an Jiaotong University, China

    2 publications · 77 citations

    2

    h-index

1–5 of 89

Venues

  1. arXiv (Cornell University)

    Repository · 11 publications · 1.1K citations

    10

    h-index

  2. Applied Mathematical Modelling

    1 publications · 24 citations

    1

    h-index

  3. Operations Research

    1 publications · 88 citations

    1

    h-index

  4. Information Sciences

    1 publications · 13 citations

    1

    h-index

  5. International Conference on Learning Representations

    Conference · 1 publications · 21 citations

    1

    h-index

1–5