ProCreations's picture
Publish exact native committed Q-learning reproduction
98bc8ca verified
Raw
History Blame Contribute Delete
601 Bytes
<!doctype html><meta charset="utf-8"><style>body{font-family:system-ui;background:#071426;color:#f8fafc;padding:28px}.grid{display:grid;grid-template-columns:repeat(3,1fr);gap:14px}.card{background:#11233d;border:1px solid #34d399;border-radius:12px;padding:18px}b{color:#86efac}</style><h1>Committed Q-learning</h1><div class="grid"><div class="card"><b>12,000 runs</b><p>Exact authors-code Figure 3 replay.</p></div><div class="card"><b>1.000 endpoint</b><p>Committed optimality at all six lengths.</p></div><div class="card"><b>0 exact gaps</b><p>Every quasi-Markov matrix identity.</p></div></div>