File size: 2,747 Bytes
9c5792c e4bc2e6 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 | {
"schema_version": "1.0",
"title": "Reproduction: Robust Sequential Experimental Design for A/B Testing",
"emoji": "🔬",
"space_id": "snaykey/repro-robust-seq-ab-testing",
"paper": {
"arxiv_id": null,
"openreview_id": "Q7lEZjtDKO"
},
"tags": [
"icml2026-repro",
"paper-Q7lEZjtDKO"
],
"updated_at": "2026-07-29T00:00:00+00:00",
"root": {
"slug": "index",
"title": "Reproduction: Robust Sequential Experimental Design for A/B Testing",
"file": "pages/index.md",
"children": [
{
"slug": "executive-summary",
"title": "Executive summary",
"file": "pages/executive-summary/page.md",
"children": []
},
{
"slug": "claim-1",
"title": "Theorem 1 formulates the optimal sequential treatment allocation problem in contextual bandits as a dynamic program with Markov state (Delta_i, Gamma_i) representing covariate imbalance statistics (Theorem 1, Section 3).",
"file": "pages/claim-1/page.md",
"children": []
},
{
"slug": "claim-2",
"title": "Theorem 2 and Theorem 3 extend the dynamic-programming formulation to dynamic settings with carryover effects, decomposing the problem into an across-day Bellman recursion over N days and a within-day finite-horizon MDP solvable via reinforcement learning (Theorem 2, Theorem 3, Section 4).",
"file": "pages/claim-2/page.md",
"children": []
},
{
"slug": "claim-3",
"title": "The proposed robust sequential design (RSD) bounds the worst-case mean squared error of the estimated treatment effect under model misspecification via an orthogonalization technique, remaining valid when the true outcome model contains unknown nonlinear components (Section 3).",
"file": "pages/claim-3/page.md",
"children": []
},
{
"slug": "claim-4",
"title": "In synthetic contextual bandit experiments with both additive and interactive treatment effects, and in dynamic settings with time horizons T in {6, 12} using 400-1000 simulation replications, RSD consistently outperforms random assignment, switchback designs, and Neyman-balanced procedures (Section 5).",
"file": "pages/claim-4/page.md",
"children": []
},
{
"slug": "claim-5",
"title": "RSD is evaluated on real-world data from a ride-sharing company in addition to synthetic experiments, demonstrating effectiveness beyond simulation (Section 5).",
"file": "pages/claim-5/page.md",
"children": []
},
{
"slug": "conclusion",
"title": "Conclusion",
"file": "pages/conclusion/page.md",
"children": []
}
]
}
} |