aiqbench analysis
Randomized Advantage Transformation (RAT) offers a practical solution to the computational bottleneck of natural policy gradients by enabling direct backpropagation without explicit Fisher matrix inversion. By leveraging randomized block Kaczmarz iterations, this method achieves convergence guarantees and performance parity with established techniques while remaining architecture-agnostic for continuous and visual control tasks.
Randomized Advantage Transformation (RAT): Computing Natural Policy Gradients via Direct Backpropagation
arXiv:2605.18591v2 Announce Type: replace-cross Abstract: Natural policy gradients improve optimization by accounting for the geometry of distribution space, but their practical use is limited by the cost of estimating and inverting the Fisher matrix. We present Randomized Advantage Transformation (RAT), a method for estimating Tikhonov-regularized natural policy gradients via direct backpropagation. By applying the Woodbury formula, we reformulate the regularized natural policy gradients as vanilla policy gradients with a transformed advantage. RAT computes this transformation efficiently via randomized block Kaczmarz iterations on on-policy mini-batches, avoiding explicit Fisher construction, conjugate-gradient solvers, and architecture-specific approximations. We provide convergence guarantees for RAT and demonstrate empirically that it matches or exceeds established natural-gradient methods across continuous and visual control benchmarks, while remaining simple to implement and compatible with various architectures.
Abstract:Natural policy gradients improve optimization by accounting for the geometry of distribution space, but their practical use is limited by the cost of estimating and inverting the Fisher matrix. We present Randomized Advantage Transformation (RAT), a method for estimating Tikhonov-regularized natural policy gradients via direct backpropagation. By applying the Woodbury formula, we reformulate the regularized natural policy gradients as vanilla policy gradients with a transformed advantage. RAT computes this transformation efficiently via randomized block Kaczmarz iterations on on-policy mini-batches, avoiding explicit Fisher construction, conjugate-gradient solvers, and architecture-specific approximations. We provide convergence guarantees for RAT and demonstrate empirically that it matches or exceeds established natural-gradient methods across continuous and visual control benchmarks, while remaining simple to implement and compatible with various architectures.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2605.18591 [cs.LG]
(or arXiv:2605.18591v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2605.18591 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 216 words
