{
  "slug": "mutable-sketches-for-fresh-embeddings-715242",
  "title": "Mutable Sketches for Fresh Embeddings",
  "dek": "Mutable sketches store user preferences in KP-trees, fit one low-rank projection, and refresh user embeddings as ratings arrive.",
  "summary": "Mutable sketches refresh recommender user embeddings after ratings using KP-trees and one fixed low-rank projection.",
  "tags": [
    "recommendation-systems",
    "mutable-sketches",
    "KP-tree",
    "embedding-staleness",
    "matrix-factorization",
    "low-rank-projection",
    "online-updates",
    "sampling",
    "cold-start",
    "KuaiRec"
  ],
  "published_at": "2026-07-18T16:09:47.523+00:00",
  "grade": 8.6,
  "agent_utility": 8.4,
  "price_usdc": 0.288,
  "stats": {
    "claims": 15,
    "entities": 25,
    "faq": 8,
    "downloads": 0,
    "paid_downloads": 0
  },
  "preview": {
    "faq_questions": [
      "What problem do mutable sketches solve?",
      "How do mutable sketches update embeddings without retraining?",
      "What is a KP-tree in this abstract?",
      "How does the method compare with ALS on KuaiRec?",
      "What is the theoretical guarantee?",
      "Does the abstract support a cold-start claim?",
      "When is norm-proportional sampling useful?",
      "Does the abstract establish production readiness?"
    ],
    "entity_names": [
      {
        "name": "Mutable sketches",
        "type": "technique"
      },
      {
        "name": "KP-tree",
        "type": "data structure"
      },
      {
        "name": "Sparse segment tree",
        "type": "data structure"
      },
      {
        "name": "Sum aggregation",
        "type": "operation"
      },
      {
        "name": "Low-rank projection",
        "type": "technique"
      },
      {
        "name": "User embedding",
        "type": "model state"
      },
      {
        "name": "Embedding staleness",
        "type": "problem"
      },
      {
        "name": "New observation",
        "type": "event"
      },
      {
        "name": "Rating",
        "type": "event"
      },
      {
        "name": "Theorem 1",
        "type": "theoretical result"
      },
      {
        "name": "Prediction error envelope",
        "type": "theoretical quantity"
      },
      {
        "name": "FunkSVD",
        "type": "model"
      },
      {
        "name": "eALS",
        "type": "model"
      },
      {
        "name": "ALS",
        "type": "model"
      },
      {
        "name": "KuaiRec",
        "type": "dataset"
      },
      {
        "name": "RMSE",
        "type": "metric"
      },
      {
        "name": "Data read",
        "type": "metric"
      },
      {
        "name": "Per-batch updates",
        "type": "evaluation setting"
      },
      {
        "name": "New user",
        "type": "user scenario"
      },
      {
        "name": "Norm-proportional sampling",
        "type": "sampling strategy"
      },
      {
        "name": "Uniform sampling",
        "type": "sampling strategy"
      },
      {
        "name": "Item coverage",
        "type": "metric"
      },
      {
        "name": "Sparse data",
        "type": "data regime"
      },
      {
        "name": "Dense matrices",
        "type": "data regime"
      },
      {
        "name": "Two-stage recommendation",
        "type": "system setting"
      }
    ],
    "related_work_titles": [
      "FunkSVD",
      "eALS",
      "ALS",
      "Uniform sampling",
      "Retrain-cycle embedding refresh"
    ],
    "application_industries": [
      "Recommendation systems",
      "Sparse recommender matrices",
      "Dense recommender matrices"
    ],
    "glossary_terms": [
      "Two-stage recommendation",
      "Embedding staleness",
      "User embedding",
      "Retrain cycle",
      "Mutable sketch",
      "KP-tree",
      "Sparse segment tree",
      "Sum aggregation",
      "Low-rank projection",
      "On-the-fly recomputation",
      "Prediction error envelope",
      "Monotonic tightening",
      "FunkSVD",
      "eALS",
      "ALS",
      "KuaiRec",
      "RMSE",
      "Data read",
      "Per-batch updates",
      "New-user personalization",
      "Sampling strategy",
      "Norm-proportional sampling",
      "Uniform sampling",
      "Item coverage",
      "Sparse data",
      "Dense matrix"
    ]
  },
  "purchase": {
    "protocol": "x402",
    "version": 2,
    "network": "eip155:8453",
    "asset": "USDC",
    "price": 0.288,
    "full_markdown_url": "https://x402.disruptive-concepts.com/api/public/article/mutable-sketches-for-fresh-embeddings-715242/full",
    "full_json_url": "https://x402.disruptive-concepts.com/api/public/article/mutable-sketches-for-fresh-embeddings-715242/full?format=json"
  }
}