poster_embed.html

<!-- Built following Chenruishuo/posterly's single-file HTML poster approach. -->
<div style="font-family:Inter,Arial,sans-serif;background:#f3f8f7;color:#17322e;padding:28px;border:2px solid #28766a;border-radius:18px;max-width:1100px;margin:auto">
  <div style="background:#174f48;color:white;padding:22px 26px;border-radius:13px">
    <div style="font-size:13px;letter-spacing:.13em;text-transform:uppercase;color:#bce5dd">ICML 2026 independent reproduction · Paper 8829</div>
    <div style="font-size:34px;font-weight:750;line-height:1.08;margin-top:8px">Discretized density guidance for source-free regression</div>
    <div style="margin-top:9px;color:#dcefeb">What is source-supported, what survives a synthetic stress test, and what remains unavailable</div>
  </div>
  <div style="display:grid;grid-template-columns:1.1fr .9fr;gap:18px;margin-top:18px">
    <section style="background:white;padding:20px;border-radius:12px;border-left:7px solid #d1843c">
      <h3 style="margin:0 0 10px;color:#174f48">Outcome</h3>
      <p>The public abstract directly supports all three high-level mechanism statements. With no public implementation, supplement, or accessible full text, theorem and benchmark replication was not possible.</p>
      <p><b>Independent proxy:</b> a naive density-weighted pseudo-label fit did not improve target MSE (0.0500 vs 0.0473), warning that discretization alone is insufficient.</p>
    </section>
    <section style="background:white;padding:20px;border-radius:12px;border-left:7px solid #28766a">
      <h3 style="margin:0 0 10px;color:#174f48">Perturbation probe</h3>
      <div style="font-size:30px;font-weight:750">0.0272</div><div>mean L1 histogram-density drift</div>
      <div style="font-size:30px;font-weight:750;margin-top:12px">87.4%</div><div>samples retaining their density bin</div>
      <p style="font-size:13px">1,000 seeds × 4,000 target samples; Gaussian input perturbation σ=0.02.</p>
    </section>
  </div>
  <div style="margin-top:18px;padding:16px 20px;background:#dcefeb;border-radius:12px"><b>Reproduction boundary:</b> source audit + independent synthetic falsification/stability probe; not a reproduction of the authors' unreleased training pipeline.</div>
</div>