Example

Bidirectional Diffusion Hyperparameter Settings (LectureBank-Full Main Runs)

The paper's main bidirectional-diffusion experiments fix the following hyperparameters. Base traversal depths: d(0)=6d_\uparrow^{(0)} = 6 (ancestors) and d(0)=2d_\downarrow^{(0)} = 2 (descendants). Adaptive-controller thresholds on the top seed's hierarchy level: (τlow,τhigh)=(2,5)(\tau_{\text{low}}, \tau_{\text{high}}) = (2, 5). Adaptive-controller thresholds on the dense-score contrast Δ=s1s5\Delta = s_1 - s_5: (γlow,γhigh)=(0.05,0.15)(\gamma_{\text{low}}, \gamma_{\text{high}}) = (0.05, 0.15). Per-hop penalties: δ=0.03\delta = 0.03 (upward) and δchild=0.025\delta_{\text{child}} = 0.025 (downward). The asymmetries d(0)>d(0)d_\uparrow^{(0)} > d_\downarrow^{(0)} and δ>δchild\delta > \delta_{\text{child}} bias the policy toward deeper but more strongly hop-penalized ancestor traversal than descendant traversal.

0

1

Updated 2026-05-17

Contributors are:

Who are from:

Tags

Science

Auditable Strict-Parity Evaluation of Prerequisite-Graph Retrieval for RAG under Leakage Controls