Tree-Coupled A/B Testing: Exact Multi-Policy Comparison via Feedback Sharing

In an arXiv paper (identifier 2608.12831) titled 'Fast A/B/n Testing: Exact Multi-Policy Comparison via Tree-Coupled Feedback Sharing', a novel approach known as Tree-Coupled A/B Testing (TCAB) is presented. This technique facilitates precise feedback sharing among various policies without altering their respective trajectory laws. Unlike traditional A/B/n frameworks that depend on JT reward outcomes, TCAB employs a structured tree to link policy histories, enhancing the coupling and allowing for a single reward per corresponding tree edge. This approach maintains each policy's independent finite-horizon trajectory law while decreasing the required outcomes, making it particularly valuable for online platforms where reward interactions are expensive. TCAB serves as a more effective option compared to conventional A/B/n testing.