| <!doctype html> |
| <html lang="en"> |
| <head> |
| <meta charset="utf-8"> |
| <meta name="viewport" content="width=device-width,initial-scale=1"> |
| <style> |
| body{font-family:Inter,system-ui,sans-serif;margin:0;background:#f7f5ff;color:#251c3b} |
| main{max-width:1080px;margin:auto;padding:28px} |
| h1{font-size:2rem;margin:.2rem 0}.lead{color:#62547b} |
| .grid{display:grid;grid-template-columns:repeat(auto-fit,minmax(205px,1fr));gap:14px;margin:22px 0} |
| .card{background:white;border:1px solid #ddd5ef;border-radius:14px;padding:16px} |
| .n{font-size:1.7rem;font-weight:750;color:#6941c6} |
| .scope{border-left:5px solid #0891b2;background:#ecfeff;padding:14px 18px;border-radius:8px} |
| img{width:100%;background:white;border-radius:14px;border:1px solid #ddd5ef} |
| </style> |
| </head> |
| <body> |
| <main> |
| <p class="lead">Independent reproduction · immutable arXiv v1 + current v3 · OpenReview RwiGcN2feP</p> |
| <h1>Unifying Low Dimensional Spectra</h1> |
| <p>Exact UFM objects plus literal paper-scale linear and ReLU training.</p> |
| <section class="grid"> |
| <div class="card"><div class="n">6/6</div><div>decisive literal verdicts</div></div> |
| <div class="card"><div class="n">2 × 10⁶</div><div>paper-scale epochs per paired run</div></div> |
| <div class="card"><div class="n">32</div><div>exact theorem cells</div></div> |
| <div class="card"><div class="n">4</div><div>executed destructive controls</div></div> |
| </section> |
| <img src="outputs/low_dimensional_spectra_audit.png" alt="Hessian, decomposition, gradient, and Table 2 diagnostics"> |
| <p class="scope"><strong>No proxy counted:</strong> Claim 6 uses the literal K=3, d=65, n=40, L=5, layer-4 ReLU Deep UFM for one million epochs and an independent exact NumPy Hessian/gradient oracle. Claim 5's “about 0.2” start is reported as a literal falsification.</p> |
| </main> |
| </body> |
| </html> |
|
|
