<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>Integrated Gradients on Cairo Cananea</title>
    <link>https://cairocananea.com.br/tags/integrated-gradients/</link>
    <description>Recent content in Integrated Gradients on Cairo Cananea</description>
    <generator>Hugo -- 0.152.2</generator>
    <language>pt</language>
    <lastBuildDate>Tue, 19 May 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://cairocananea.com.br/tags/integrated-gradients/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Além do fine-tuning: LoRA com DistilBERT para detecção precoce de insatisfação nos reviews de ecommerce</title>
      <link>https://cairocananea.com.br/projects/amazon-reviews-classification-lora/</link>
      <pubDate>Tue, 19 May 2026 00:00:00 +0000</pubDate>
      <guid>https://cairocananea.com.br/projects/amazon-reviews-classification-lora/</guid>
      <description>Além do fine-tuning total: LoRA atinge F1 0,943 e AUC 0,996 com apenas 1% dos parâmetros. Mas bugs silenciosos (módulos errados, PEFT quebrando Captum) e o equívoco conceitual sobre LoRA como redirecionador, revelam os trade-offs reais do método</description>
    </item>
  </channel>
</rss>
