<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Sample Selection | Jorge Kamassury — Academic CV</title><link>https://kamassury.github.io/tags/sample-selection/</link><atom:link href="https://kamassury.github.io/tags/sample-selection/index.xml" rel="self" type="application/rss+xml"/><description>Sample Selection</description><generator>Hugo Blox Builder (https://hugoblox.com)</generator><language>pt-br</language><lastBuildDate>Thu, 15 May 2025 00:00:00 +0000</lastBuildDate><image><url>https://kamassury.github.io/media/icon_hu_982c5d63a71b2961.png</url><title>Sample Selection</title><link>https://kamassury.github.io/tags/sample-selection/</link></image><item><title>CCT: A Cyclic Co-Teaching Approach to Train Deep Neural Networks With Noisy Labels</title><link>https://kamassury.github.io/publications/journal-article/cyclic-co-teaching/</link><pubDate>Thu, 15 May 2025 00:00:00 +0000</pubDate><guid>https://kamassury.github.io/publications/journal-article/cyclic-co-teaching/</guid><description>&lt;p&gt;Propomos &lt;strong&gt;Cyclic Co-Teaching (CCT)&lt;/strong&gt;, um método de treinamento robusto para redes neurais profundas em bases com rótulos ruidosos. O método combina &lt;strong&gt;modulações cíclicas da taxa de aprendizado&lt;/strong&gt; e &lt;strong&gt;retenção de amostras&lt;/strong&gt;, criando uma dinâmica alternada entre fases de especialização (aprendizado intensivo) e consolidação (estabilização). Experimentos em bases sintéticas (CIFAR-10/100, Tiny-ImageNet) e reais (Animal-10N, Food-101N, Clothing1M) mostram ganhos consistentes sobre o estado da arte, especialmente em cenários de alta taxa de ruído.&lt;/p&gt;</description></item></channel></rss>