New Tricks for Estimating Gradients of Expectations
This work addresses a fundamental challenge in machine learning for researchers and practitioners, offering incremental improvements in gradient estimation techniques.
The paper tackles the problem of estimating gradients of expectations by introducing a family of pairwise stochastic gradient estimators, which are unbiased and provide an independent component of useful information compared to existing methods, with promising analytical and numerical examples confirming the theory.
We introduce a family of pairwise stochastic gradient estimators for gradients of expectations, which are related to the log-derivative trick, but involve pairwise interactions between samples. The simplest example of our new estimator, dubbed the fundamental trick estimator, is shown to arise from either a) introducing and approximating an integral representation based on the fundamental theorem of calculus, or b) applying the reparameterisation trick to an implicit parameterisation under infinitesimal perturbation of the parameters. From the former perspective we generalise to a reproducing kernel Hilbert space representation, giving rise to a locality parameter in the pairwise interactions mentioned above, yielding our representer trick estimator. The resulting estimators are unbiased and shown to offer an independent component of useful information in comparison with the log-derivative estimator. We provide a further novel theoretical analysis which further characterises the variance reduction afforded by the new techniques. Promising analytical and numerical examples confirm the theory and intuitions behind the new estimators.