{
  "schema_version": 1,
  "builder_plan": {
    "scenario": "A self-hosted model uses more compute or memory than your workload can justify.",
    "change": "Check for a usable pruned checkpoint and runtime support before attempting model surgery.",
    "baseline": "The original unpruned model at the same precision, context, and load.",
    "steps": [
      "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.",
      "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.",
      "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.",
      "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test."
    ],
    "measure": "Task quality, actual GPU memory, serving throughput, and preparation cost.",
    "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation."
  },
  "builder_review": null,
  "arxiv_id": "2610.11853",
  "title": "DADP: Dynamic Activity-Dependent Pruning, A Reverse Hebbian-Inspired Structural Pruning Method",
  "summary": "DADP replaces post-hoc magnitude thresholds, static initialization heuristics such as SNIP, and manual per-layer sparsity budgets with a single global importance threshold computed during training from pre-synaptic activations times post-synaptic error gradients. Its cost is the added per-step importance accumulation and mask bookkeeping during training, plus a residual accuracy gap (about 2.4 points at 99% sparsity on ResNet-18), and it does not eliminate the need to retrain from scratch. It can fail through threshold sensitivity that prunes critical neurons early, workload-dependent sparsity allocation that does not transfer across architectures, and unvalidated behavior at production LLM scale, since the largest model tested is MiniBERT.",
  "rationale": "The method is implementable on a single 24 GB GPU and could reduce training memory without per-layer tuning, but the evidence covers only small architectures, so the reader should validate it on their own workload before production adoption.",
  "technique": "Dynamic activity-dependent structural pruning during training",
  "family": "esparsidade_e_poda",
  "recommendation": "testar",
  "published": "2026-10-08",
  "as_of": "2026-10-18",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2610.11853",
  "independent_implementations": 0,
  "total_implementations": 0,
  "stars": 0,
  "citations": null,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "DADP replaces post-hoc magnitude thresholds, static initialization heuristics such as SNIP, and manual per-layer sparsity budgets with a single global importance threshold computed during training from pre-synaptic activations times post-synaptic error gradients. Its cost is the added per-step importance accumulation and mask bookkeeping during training, plus a residual accuracy gap (about 2.4 points at 99% sparsity on ResNet-18), and it does not eliminate the need to retrain from scratch. It can fail through threshold sensitivity that prunes critical neurons early, workload-dependent sparsity allocation that does not transfer across architectures, and unvalidated behavior at production LLM scale, since the largest model tested is MiniBERT.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.11853",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "claim_id": "claim-02",
      "statement": "Retains 73.67% accuracy versus a 76.06% dense baseline at 99% sparsity on ResNet-18, and reports matching or outperforming Magnitude, SNIP, and RigL across MLP, VGG-16, ResNet-18, BiLSTM-CRF, and MiniBERT.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.11853",
      "source_page": null,
      "source_excerpt": ""
    }
  ],
  "exposure_map": [
    {
      "dimension": "quality",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "compute",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "latency",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "operations",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "compatibility",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "security",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "data_and_training",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "dimension": "reproducibility",
      "basis": "not_evaluated",
      "finding": "",
      "interpretation": "",
      "limitation": "",
      "next_check": "",
      "source_url": "",
      "source_page": null,
      "source_excerpt": ""
    }
  ],
  "risks": [],
  "minimum_test": [],
  "open_questions": [],
  "independent_tests": [],
  "lab_reproductions": [],
  "report_available": false,
  "validation_tier": "unknown",
  "evidence_tier": "unknown",
  "infrastructure_basis": "unknown",
  "equations": [],
  "equations_status": "not_fetched",
  "core_kind": "",
  "equations_fetched_at": "",
  "equations_source_url": "",
  "equations_source_sha256": "",
  "exposures_reviewed_at": "",
  "exposures_source_sha256": ""
}