BEGIN:VCALENDAR
VERSION:2.0
X-WR-CALNAME;VALUE=TEXT:Statistics Colloquium: Harrison Zhou (Yale University)
PRODID:-//Harvard events data//EN
BEGIN:VEVENT
UID:event_1369590_0
SUMMARY:Statistics Colloquium: Harrison Zhou (Yale University)
DESCRIPTION:<h3>	<u>Title:</u></h3><p>	<span>Global Convergence of EM?</span></p><h3>	<u>Abstract:</u></h3><p>	<span>In this talk I will first discuss a recent joint work with Yihong Wu: <a href="https://urldefense.proofpoint.com/v2/url?u=https-3A__arxiv.org_abs_1908.10935&amp;d=DwMGaQ&amp;c=WO-RGvefibhHBZq3fL85hQ&amp;r=cSZTSObb9KOZYAv_RgNPYOFsxuPhRQql-BOH56RsWMSZzhUjq3GA4enOiQZiC44G&amp;m=QSQA8HjsOn_KvadA8q95LUjrx6NUnRcQjSHhfbN-q8s&amp;s=vwO4TeZAQJ_Niq2TLiBmE3xwvREMRGBZdbOR4DVVkSI&amp;e=">https://arxiv.org/abs/1908.10935</a>. We show that </span><span style="">the randomly initialized EM algorithm</span><span> for parameter estimation in the symmetric two-component Gaussian mixtures converges to the MLE in at most $\sqrt(n)$ iterations with high probability. Then I will mention the limitations of that work and propose an extension to general Gaussian mixtures by overparameterization. </span></p>
LOCATION:Zoom - please contact emilie_campanelli@fas.harvard.edu for more information
STATUS:CONFIRMED
DTSTART:20201123T153000Z
DTEND:20201123T163000Z
END:VEVENT
END:VCALENDAR