<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>The Glacius Blog</title><link>https://glacius.ai/blog</link><description>Visual explanations of mathematics and machine learning.</description><language>en</language><atom:link href="https://glacius.ai/feed.xml" rel="self" type="application/rss+xml"/><item><title>How vector length changes dot product and cosine similarity</title><link>https://glacius.ai/blog/dot-product-vs-cosine-similarity</link><guid isPermaLink="true">https://glacius.ai/blog/dot-product-vs-cosine-similarity</guid><pubDate>Thu, 10 Sep 2026 00:00:00 GMT</pubDate><description>Dot product vs cosine similarity, explained visually. See how vector length changes a similarity score, with a worked example and the normalization formula.</description></item><item><title>Why gradient descent can oscillate</title><link>https://glacius.ai/blog/why-gradient-descent-oscillates</link><guid isPermaLink="true">https://glacius.ai/blog/why-gradient-descent-oscillates</guid><pubDate>Thu, 10 Sep 2026 00:00:00 GMT</pubDate><description>Why gradient descent oscillates: a visual, worked example showing convergence, overshooting, and divergence as the learning rate changes.</description></item><item><title>What backpropagation does before gradient descent takes a step</title><link>https://glacius.ai/blog/backpropagation-vs-gradient-descent</link><guid isPermaLink="true">https://glacius.ai/blog/backpropagation-vs-gradient-descent</guid><pubDate>Thu, 10 Sep 2026 09:54:08 GMT</pubDate><description>Follow one training step by hand to see how backpropagation calculates gradients, how gradient descent uses them, and where the learning rate enters the calculation.</description></item><item><title>Why cosine similarity and Euclidean distance can disagree</title><link>https://glacius.ai/blog/cosine-similarity-vs-euclidean-distance</link><guid isPermaLink="true">https://glacius.ai/blog/cosine-similarity-vs-euclidean-distance</guid><pubDate>Thu, 10 Sep 2026 09:54:08 GMT</pubDate><description>Compare two neighbors by hand, then normalize the vectors to see exactly when cosine similarity and Euclidean distance produce the same ranking.</description></item></channel></rss>