{ "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": "2610.23415", "title": "ConQUESTA: Topology-Aware Post-Training Structural Pruning of Quantum Neural Networks", "summary": "ConQUESTA replaces one-shot saliency ranking of QNN entanglers with a constrained mask search in which each candidate pruning mask is fully transpiled across hardware coupling topologies, combining task saliency with compiler cost. The cost is post-training search overhead per mask (transpilation and re-evaluation under an accuracy constraint), plus a fixed-mask recovery step needed to reach feasibility in 22 of 24 runs. It can fail when pruning masks are infeasible for a given topology or when accuracy drops exceed the constraint on small datasets, and it applies only to quantum circuits of 4-12 qubits on simulators or NISQ hardware.", "rationale": "The method targets quantum neural network deployment on sparse quantum hardware, which is outside the reader's classical GPU/API production infrastructure and offers no actionable benefit for conventional ML workloads.", "technique": "Topology-aware post-training pruning of quantum neural networks", "family": "esparsidade_e_poda", "recommendation": "nao_aplica", "published": "2026-10-16", "as_of": "2026-10-19", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.23415", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "ConQUESTA replaces one-shot saliency ranking of QNN entanglers with a constrained mask search in which each candidate pruning mask is fully transpiled across hardware coupling topologies, combining task saliency with compiler cost. The cost is post-training search overhead per mask (transpilation and re-evaluation under an accuracy constraint), plus a fixed-mask recovery step needed to reach feasibility in 22 of 24 runs. It can fail when pruning masks are infeasible for a given topology or when accuracy drops exceed the constraint on small datasets, and it applies only to quantum circuits of 4-12 qubits on simulators or NISQ hardware.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.23415", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "At 55% pruning, mean two-qubit gate/SWAP/depth reductions of 61.6% (Breast Cancer), 72.0% (Digits01), and 82.2% (Iris), exceeding the strongest matched baseline by 3.5, 1.7, and 8.9 percentage points; mask-space search stays tractable (305 evaluated masks vs 12.1M candidates at 12 qubits).", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.23415", "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": "" }