<div dir="ltr"><div class="gmail_default" style="font-family:verdana,sans-serif;font-size:small">Prezados(as)</div><div class="gmail_default" style="font-family:verdana,sans-serif;font-size:small">Convido a todos(as) que puderem participar</div><div class="gmail_default" style="font-family:verdana,sans-serif;font-size:small">para mais um seminário em Análise de Dados em Alta Dimensão <AD,AD></div><div class="gmail_default" style="font-family:verdana,sans-serif;font-size:small">Sejam bem vindos.</div><div class="gmail_default" style="font-family:verdana,sans-serif;font-size:small"><br></div><div class="gmail_default" style="font-family:verdana,sans-serif;font-size:small"><div class="gmail_default">Local: sala 221-IMECC</div><div class="gmail_default">Data: 29/05 as 13hs.</div><div class="gmail_default"><br></div><div class="gmail_default"><span style="font-family:Arial,Helvetica,sans-serif">Accelerating block coordinate descent methods with identification</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">strategies and Applications to Lasso and Logistic Regression</span><br style="font-family:Arial,Helvetica,sans-serif"><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">This work is about active set identification strategies aimed at</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">accelerating block-coordinate descent methods (BCDM) applied to</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">large-scale problems. We start by devising an identification function</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">tailored for bound-constrained composite minimization together with an</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">associated version of the BCDM, called Active BCDM, that is also</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">globally convergent. The identification function gives rise to an</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">efficient practical strategy for Lasso and ℓ1-regularized logistic</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">regression. The computational performance of Active BCDM is</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">contextualized using comparative sets of experiments that are based on</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">the solution of problems with data from deterministic instances from</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">the literature. These results have been compared with those of</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">well-established and state-of-the-art methods that are particularly</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">suited for the classes of applications under consideration. Active</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">BCDM has proved useful in achieving fast results due to its</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">identification strategy. Besides that, an extra second-order step was</span><br style="font-family:Arial,Helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif">used, with favorable cost-benefit.</span></div></div><br clear="all"><div><br></div>-- <br><div dir="ltr" class="gmail_signature" data-smartmail="gmail_signature"><div dir="ltr"><div><div dir="ltr"><div><div>Ronaldo Dias, Ph.D.</div>Professor<div>Dept. of Statistics-IMECC, UNICAMP</div><div><a href="http://www.ime.unicamp.br/~dias" target="_blank">www.ime.unicamp.br/~dias</a></div><div><br></div></div></div></div></div></div></div>