https://mllog.dev/en/categories/popular-science/Recent content in Popular Science on MLLog.devhttps://mllog.dev/images/default_mllog.pnghttps://mllog.dev/images/default_mllog.pngHugo -- 0.147.9enTue, 15 Jul 2025 00:00:00 +0000https://mllog.dev/en/posts/target-polish/Tue, 15 Jul 2025 00:00:00 +0000https://mllog.dev/en/posts/target-polish/<p>Imagine you&rsquo;re analyzing sensor data. Suddenly one sensor shows -999°C. That&rsquo;s an <em>outlier</em> — a single data point that can completely ruin your analysis.</p> <h2 id="-what-is-factorization">🧩 What is factorization?</h2> <p>Matrix factorization means decomposing data $X$ into two non-negative components: $$ X \approx WH $$</p> <p>Where $W$ contains “features” and $H$ shows how much of each is needed.</p> <h2 id="-the-problem">💡 The problem</h2> <p>Classical methods like NMF are sensitive to noise and outliers. When data is messy, analysis breaks down.</p>