AI Paper Atlas

Research map / Time series / Causal and generative time series

Detecting Abrupt Changes in Sequential Pairwise Comparison Data

Wanshan Li, Alessandro Rinaldo, Daren Wang

NeurIPS 2022 · Time series › Causal and generative time series · approach: Probabilistic and conformal forecasting · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use DPLR instead of WBS-GLRSequential BTL change-point localization · Simulation settings (i)-(iv)Hausdorff distance: 9.2 (9.1) vs 15.2 (7.9); 9.0 (9.9) vs 240.5 (220.3); 13.4 (14.4) vs 111.9 (195.6); 12.4 (12.1) vs 412.3 (495.5)Improved
Use DPLR instead of WBS-GLRSequential BTL change-point localization · Simulation settings (i)-(iv)running time: 49.7s (0.7) vs 31.9s (3.9); 118.5s (2.2) vs 144.2s (12.5); 167.4s (3.3) vs 215.9s (17.0); 402.4s (7.4) vs 400.0s (40.9)Mixed
Use DPLR instead of WBS-GLRSequential BTL change-point localization · Random-change simulationsHausdorff distance: 12.1 (13.3) vs 94.9 (174.8); 23.9 (27.6) vs 251.7 (219.9); 43.1 (103.4) vs 133.0 (194.9); 28.3 (26.5) vs 459.4 (512.8)Improved
Increase the proportion of randomly permuted entriesDPLR · n=20, K=3, Δ=800Hausdorff distance: 362.8 (502.2) -> 114.4 (251.3) -> 23.9 (27.6)Improved
Use local refinement after dynamic programmingDynamic-programming BTL change-point estimator · Theoretical localization error boundlocalization error bound: p_lb^{-4} K |E|nd_max/(κ^2 λ_2^2(L_G)) log(Tn) -> p_lb^{-2} |E|nd_max/(κ^2 λ_2^2(L_G)) log(Tn)Improved
Use WBS-SST instead of DPLRSequential BTL change-point localization · Simulation setting (iii), n=100, K=2, Δ=1000Hausdorff distance: 13.4 (14.4) vs ∞ (NA)Hurt
Use WBS-SST instead of DPLRSequential BTL change-point localization · Simulation setting (iv), n=100, K=3, Δ=2000Hausdorff distance: 12.4 (12.1) vs 1116.3 (694.8)Hurt
Use WBS-Mean instead of DPLRSequential BTL change-point localization · Simulation setting (ii), n=20, K=3, Δ=800Hausdorff distance: 9.0 (9.9) vs 5.8 (11.4)Improved
Use DPLR instead of WBS-MeanSequential BTL change-point localization · Simulation settings (i), (iii), and (iv)Hausdorff distance: 9.2 (9.1) vs 15.4 (8.4); 13.4 (14.4) vs 22.9 (98.4); 12.4 (12.1) vs 17.9 (6.1)Mixed
Replace likelihood-based detection with Borda-count CUSUMWBS change-point detection · Constructed BTL probability-matrix examplepopulation CUSUM statistic at the true change: 0Hurt
Use the SST two-sample statistic in WBS with insufficient sample sizeWBS-SST · Small-Δ sequential comparisonschange-point detection powerHurt
Vary the number of WBS-GLR random intervals from 50 to 250 (ablation)WBS-GLR · NBA gamestest negative log-likelihood: 1796.9, 1796.9, 1796.9, 1793.2, 1796.9No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy