{
  "id": 7486890,
  "title": "Per-Matrix Optimality Is Not Enough: Three-Level Optimization for Low-Rank LLM Compression",
  "url": "https://urgent.news/2026/09/14/per-matrix-optimality-is-not-enough-three-level-optimization-for-low",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-14T16:36:46.000Z",
  "source": {
    "name": "arXiv cs.AI",
    "slug": "arxiv-cs-ai",
    "url": "https://arxiv.org/abs/2609.15838v1"
  },
  "original_language": "en",
  "account": null,
  "summary": "Per-matrix singular value decomposition (SVD) truncation is Eckart-Young optimal in the whitened Frobenius norm, but errors from independently compressed matrices compound through the block's nonlinear forward pass. Inspired in part by hierarchical variational optimization in quantum many-body methods, we introduce a three-level chain that widens optimization scope from individual matrices to…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}