snaykey's picture
Update logbook: On the Accuracy of Newton Step and Influence Function Data Attributions
5a01788 verified
Raw
History Blame Contribute Delete
3.15 kB
{
"schema_version": 1,
"title": "On the Accuracy of Newton Step and Influence Function Data Attributions",
"emoji": "📊",
"space_id": "snaykey/repro-newton-step-influence-function-data-attributions",
"paper": {
"arxiv_id": "2512.12572",
"openreview_id": "mDo8XNqopd"
},
"tags": [
"icml2026-repro",
"paper-mDo8XNqopd"
],
"updated_at": "2026-07-24T19:41:56+00:00",
"root": {
"slug": "index",
"title": "On the Accuracy of Newton Step and Influence Function Data Attributions",
"file": "pages/index.md",
"children": [
{
"slug": "claim-1-theorem-1-5-local-strong-convexity",
"title": "Theorem 1.5 bounds the Newton-step approximation error to the true leave-one-out retrained parameter using only local strong convexity in a neighborhood of the Newton step, rather than the global strong convexity assumed in prior analyses (Theorem 1.5, Section 3).",
"file": "pages/claim-1-theorem-1-5-local-strong-convexity/page.md",
"children": []
},
{
"slug": "claim-2-theorem-1-2-error-scaling-laws",
"title": "For logistic regression with n samples, d-dimensional Gaussian features, and k removed points, the average-case Newton-step attribution error scales as Õ(kd/n²), while the average-case influence-function error scales as Õ((k^{3/2}d^{1/2}+k^{1/2}d^{3/2})/n²) (Theorem 1.2).",
"file": "pages/claim-2-theorem-1-2-error-scaling-laws/page.md",
"children": []
},
{
"slug": "claim-3-theorem-1-2-matching-bounds",
"title": "The derived upper bounds match corresponding lower bounds up to polylogarithmic factors, showing the Newton step is provably more accurate than influence functions whenever d ≫ k (Theorem 1.2).",
"file": "pages/claim-3-theorem-1-2-matching-bounds/page.md",
"children": []
},
{
"slug": "claim-4-theorem-1-6-1-7-rif-drif",
"title": "Rescaled and doubly-rescaled influence functions (RIF and DRIF) are shown to match the Newton step's Õ(kd/n²) average-case error rate while preserving the additivity property that plain influence functions have but the Newton step lacks (Theorem 1.6, Theorem 1.7).",
"file": "pages/claim-4-theorem-1-6-1-7-rif-drif/page.md",
"children": []
},
{
"slug": "claim-5-remove-lambda-dependence",
"title": "The paper's revised bounds remove the problematic 1/λ³ dependence on the regularization coefficient λ present in prior influence-function error analyses (Section 1, Discussion of prior bounds).",
"file": "pages/claim-5-remove-lambda-dependence/page.md",
"children": []
},
{
"slug": "executive-summary",
"title": "Executive summary",
"file": "pages/executive-summary/page.md",
"children": []
},
{
"slug": "conclusion",
"title": "Conclusion",
"file": "pages/conclusion/page.md",
"children": []
}
]
},
"agent_view_tokens": 5480,
"revision": "1784922116190000500"
}