| [ | |
| "GSCF (Generalized Spanning Converging Forest) generates spanning converging forests for signed graphs using loop-erased random walks in expected O(n) time, per Theorem 5.2 (Section 5.1, Algorithm 1, Theorem 5.2).", | |
| "FMDE estimates the diagonal of the forest matrix via sampling in O(ln) time, where l is the number of samples (Section 6.2, Algorithm 2).", | |
| "FMDE+ incorporates additional network information to achieve the same O(ln) time complexity as FMDE but with lower variance, and Theorem 6.3 proves its superior performance on balanced signed graphs (Section 6.2, Theorem 6.3).", | |
| "On the SlashdotZoo network, exact diagonal computation via direct matrix inversion takes 991.5 seconds while FMDE+ completes in under 1.1 seconds (Table 1).", | |
| "The methods scale to the FullUSA network with roughly 2.3×10^7 (over twenty million) nodes, where exact computation fails but FMDE+ completes within one hour (Table 3).", | |
| "FMDE+ achieves roughly ten times lower average relative error than FMDE across tested networks (Figure 2)." | |
| ] | |