Reinforcement Learning-Aided Design of Efficient Polarization Kernels

TL;DR

This paper introduces PolarZero, a reinforcement learning method that efficiently designs large polarization kernels for polar codes, matching exhaustive search results and enabling scalable kernel discovery.

Contribution

The paper presents PolarZero, a reinforcement learning approach based on Gumbel AlphaZero for designing low-complexity polarization kernels, outperforming traditional search methods.

Findings

PolarZero matches exhaustive search in kernel identification.

Discovered a size-16 kernel with low complexity.

PolarZero is scalable for large-kernel design.

Abstract

Polar codes with large kernels achieve optimal error exponents but are difficult to construct when low decoding complexity is also required. We address this challenge under recursive maximum likelihood decoding (RMLD) using a rein-forcement learning approach based on the Gumbel AlphaZero algorithm. The resulting method, PolarZero, consistently matches exhaustive search in identifying low-complexity kernels, and discovers a size-16 kernel with complexity comparable to handcrafted designs. Our results suggest that PolarZero is a scalable tool for large-kernel design, where brute-force search is no longer feasible.