logbook.json

{
 "schema_version": 1,
 "title": "Feasibility Methods Reproduction",
 "emoji": "🎯",
 "space_id": "snaykey/repro-feasibility-methods",
 "paper": {
  "openreview_id": "1BchRVONfp",
  "arxiv_id": "2601.20076"
 },
 "tags": [
  "icml2026-repro",
  "paper-1BchRVONfp"
 ],
 "updated_at": "2026-07-25T15:23:50+00:00",
 "root": {
  "slug": "index",
  "title": "Feasibility Methods Reproduction",
  "file": "pages/index.md",
  "children": [
   {
    "slug": "executive-summary",
    "title": "Executive summary",
    "file": "pages/executive-summary/page.md",
    "children": []
   },
   {
    "slug": "claim-1-linear-convergence",
    "title": "Proposes Algorithm 2, a Gradient Method with Randomized Feasibility using an adaptive Polyak-type stepsize, and proves linear convergence in expectation to a prescribed tolerance ε for strongly convex, Lipschitz-smooth objectives, with iteration complexity O(log(1/ε)) (Theorem 4.4).",
    "file": "pages/claim-1-linear-convergence/page.md",
    "children": []
   },
   {
    "slug": "claim-2-convex-rate",
    "title": "Introduces Algorithm 3, DoWS (Distance over Weighted Subgradients) with Randomized Feasibility, a parameter-free adaptive-stepsize method for convex possibly-nonsmooth objectives, proving an O(1/√T) worst-case convergence rate in expectation (Theorem 5.3).",
    "file": "pages/claim-2-convex-rate/page.md",
    "children": []
   },
   {
    "slug": "claim-4-tdows-theorem-5-5",
    "title": "Introduces Algorithm 4, T-DoWS (Tamed DoWS), which removes the requirement that the constraint set Y be bounded, retaining an O(1/√T) rate up to logarithmic factors (Theorem 5.5).",
    "file": "pages/claim-4-tdows-theorem-5-5/page.md",
    "children": []
   },
   {
    "slug": "claim-3-infeasibility",
    "title": "Proves (Lemma 3.1) that infeasibility of iterates under the randomized feasibility update decreases geometrically almost surely with the number of feasibility updates, without requiring compactness of Y.",
    "file": "pages/claim-3-infeasibility/page.md",
    "children": []
   },
   {
    "slug": "claim-5-qcqp-figure-1",
    "title": "Validates the theory on quadratically constrained quadratic program (QCQP) simulations under strongly convex and convex objectives with known/unknown optimal value f*, confirming predicted function-value decay and geometric infeasibility reduction (Figure 1).",
    "file": "pages/claim-5-qcqp-figure-1/page.md",
    "children": []
   },
   {
    "slug": "claim-6-svm-figure-2",
    "title": "Evaluates Algorithms 3 and 4 against a primal-dual baseline on SVM classification with three real datasets (Banknote Authentication, Breast Cancer Wisconsin, MNIST 3-vs-5), comparing test misclassification error (Figure 2).",
    "file": "pages/claim-6-svm-figure-2/page.md",
    "children": []
   },
   {
    "slug": "conclusion",
    "title": "Conclusion",
    "file": "pages/conclusion/page.md",
    "children": []
   }
  ]
 },
 "agent_view_tokens": 438,
 "revision": "1784461053109399900"
}