| <!doctype html> |
| <html lang="en"> |
| <head> |
| <meta charset="utf-8"> |
| <meta name="viewport" content="width=device-width,initial-scale=1"> |
| <title>CGP direct reproduction poster</title> |
| <style> |
| body{font-family:system-ui,sans-serif;background:#0b1020;color:#edf2ff;margin:0;padding:2rem} |
| main{max-width:1080px;margin:auto}.grid{display:grid;grid-template-columns:repeat(auto-fit,minmax(210px,1fr));gap:1rem} |
| .card{background:#17213b;border:1px solid #33456f;border-radius:14px;padding:1rem}h1,h2{color:#9ec5ff} |
| .metric{font-size:2rem;font-weight:700;color:#81e6d9}li{margin:.55rem 0}small{color:#bac8e8} |
| img{width:100%;border-radius:12px;background:white;margin-top:1rem} |
| </style> |
| </head> |
| <body><main> |
| <h1>Certificate-Guided Pruning: direct Algorithms 1--3 audit</h1> |
| <div class="grid"> |
| <section class="card"><div class="metric">34/34</div><p>original plus direct-execution gates pass</p></section> |
| <section class="card"><div class="metric">3,840</div><p>actual CGP observations</p></section> |
| <section class="card"><div class="metric">45,000</div><p>Gaussian lower-bound trials</p></section> |
| <section class="card"><div class="metric">60/100D</div><p>executed certified trust regions</p></section> |
| </div> |
| <h2>Direct coverage</h2> |
| <ul> |
| <li>16 full Algorithm-1 runs on cone, quadratic, Branin, and Rastrigin objectives.</li> |
| <li>Every good-event active-set certificate retains the optimum and excludes only suboptimal candidates.</li> |
| <li>48 noisy Algorithm-2 doubling runs finish in the theorem's Lipschitz interval.</li> |
| <li>12 Algorithm-3 executions never restart the optimizer-containing region and certify suboptimal restarts.</li> |
| </ul> |
| <img src="outputs/direct_cgp_upgrade.png" alt="Direct CGP reproduction plots"> |
| <p><small>Source, algorithms, outputs, byte-identical replay, and recursive bundle are SHA-256 pinned. Unavailable Rover/NAS/MuJoCo environments are not claimed as reruns.</small></p> |
| </main></body> |
| </html> |
|
|
