{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-02-018",
    "source_story_id": "tmp-sidebar-softserve-quasi-newton",
    "edition_id": "mp-2026-10-02-morning-0085",
    "edition_url": "https://themachinepress.com/edition/2026-10-02",
    "position": 20,
    "story_type": "ticker",
    "section": "research",
    "editorial_classification": "editorial",
    "headline": "Quasi-Newton Training Scaled to 136 Million Parameters",
    "slug": "quasi-newton-training-scaled-to-136-million-parameters",
    "dek": "Positive-definite curvature estimates avoided line searches and GPU-unfriendly decompositions.",
    "summary": "Positive-definite curvature estimates avoided line searches and GPU-unfriendly decompositions.",
    "body_text": "SoftServe builds diagonal and Kronecker-factored curvature estimates that remain positive definite under nonconvexity, using coupled Newton-Schulz matrix operations. It often reached lower losses than Adam, Muon and SOAP on ill-conditioned tasks including a 136-million-parameter physics-informed diffusion model.",
    "why_it_matters": "Positive-definite curvature estimates avoided line searches and GPU-unfriendly decompositions.",
    "limitations": [
      "SoftServe builds diagonal and Kronecker-factored curvature estimates that remain positive definite under nonconvexity, using coupled Newton-Schulz matrix operations."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-02-018/quasi-newton-training-scaled-to-136-million-parameters",
    "json_url": "https://themachinepress.com/story/mp-2026-10-02-018.json",
    "first_published_at": "2026-10-02T09:00:00.000-04:00",
    "modified_at": "2026-10-02T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-02-018-001",
        "text": "Positive-definite curvature estimates avoided line searches and GPU-unfriendly decompositions.",
        "source_ids": [
          "source-2026-10-02-020"
        ],
        "qualification": "SoftServe builds diagonal and Kronecker-factored curvature estimates that remain positive definite under nonconvexity, using coupled Newton-Schulz matrix operations."
      }
    ],
    "source_ids": [
      "source-2026-10-02-020"
    ],
    "tags": [
      "optimization",
      "deep learning",
      "quasi-Newton"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-02-020",
      "title": "arXiv preprint 2610.02182",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2610.02182",
      "canonical_url": "https://arxiv.org/abs/2610.02182",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-01T13:58:24.000-04:00",
      "accessed_at": "2026-10-02T08:44:06.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-02-018-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Quasi-Newton Training Scaled to 136 Million Parameters",
    "publisher": "The Machine Press",
    "published_at": "2026-10-02T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-02-018/quasi-newton-training-scaled-to-136-million-parameters"
  }
}