<?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>Geometry on SciDraft | A Hugo Theme</title><link>https://www.zhumengyao.com/scidraft/tags/geometry/</link><description>Recent content in Geometry on SciDraft | A Hugo Theme</description><generator>Hugo</generator><language>en</language><lastBuildDate>Mon, 21 Sep 2026 15:12:00 +0900</lastBuildDate><atom:link href="https://www.zhumengyao.com/scidraft/tags/geometry/index.xml" rel="self" type="application/rss+xml"/><item><title>Rotary Position Embedding (formula)</title><link>https://www.zhumengyao.com/scidraft/notes/demo-note-formula-rope/</link><pubDate>Mon, 21 Sep 2026 15:12:00 +0900</pubDate><guid>https://www.zhumengyao.com/scidraft/notes/demo-note-formula-rope/</guid><description>&lt;p&gt;&lt;strong&gt;Demo note.&lt;/strong&gt; This note demonstrates what the &lt;strong&gt;SciDraft&lt;/strong&gt; Hugo theme can do
for academic and scientific publishing — for research institutions, researchers
and students. Its content is demonstration material only: readers should ignore
whether the demo content is correct.&lt;/p&gt;
&lt;p&gt;Position information has to enter attention somewhere, and one way is to rotate the query
and key vectors by a position-dependent angle. The formula is appealing because the effect
on the score depends only on the difference of the two positions.&lt;/p&gt;</description></item><item><title>Features Are Directions (conjecture)</title><link>https://www.zhumengyao.com/scidraft/notes/demo-note-conjecture-features-directions/</link><pubDate>Mon, 21 Sep 2026 14:32:00 +0900</pubDate><guid>https://www.zhumengyao.com/scidraft/notes/demo-note-conjecture-features-directions/</guid><description>&lt;p&gt;&lt;strong&gt;Demo note.&lt;/strong&gt; This note demonstrates what the &lt;strong&gt;SciDraft&lt;/strong&gt; Hugo theme can do
for academic and scientific publishing — for research institutions, researchers
and students. Its content is demonstration material only: readers should ignore
whether the demo content is correct.&lt;/p&gt;
&lt;p&gt;The linear representation hypothesis conjectures that the concepts a network has
learned are encoded as directions in activation space, so that adding a concept vector to
an activation moves the computation in a meaningful way.&lt;/p&gt;</description></item><item><title>A Bound on the Attention Output Norm (corollary)</title><link>https://www.zhumengyao.com/scidraft/notes/demo-note-corollary-attention-norm/</link><pubDate>Mon, 21 Sep 2026 13:02:00 +0900</pubDate><guid>https://www.zhumengyao.com/scidraft/notes/demo-note-corollary-attention-norm/</guid><description>&lt;p&gt;&lt;strong&gt;Demo note.&lt;/strong&gt; This note demonstrates what the &lt;strong&gt;SciDraft&lt;/strong&gt; Hugo theme can do
for academic and scientific publishing — for research institutions, researchers
and students. Its content is demonstration material only: readers should ignore
whether the demo content is correct.&lt;/p&gt;
&lt;p&gt;Reading a bound off a proposition is the standard use of a corollary, and the one worth
having here is a ceiling on how large an attention output can be. Since the output is a
convex combination of value vectors, its length can never exceed the longest of them.&lt;/p&gt;</description></item><item><title>Equivariance as a Design Principle (principle)</title><link>https://www.zhumengyao.com/scidraft/notes/demo-note-principle-equivariance/</link><pubDate>Mon, 21 Sep 2026 11:52:00 +0900</pubDate><guid>https://www.zhumengyao.com/scidraft/notes/demo-note-principle-equivariance/</guid><description>&lt;p&gt;&lt;strong&gt;Demo note.&lt;/strong&gt; This note demonstrates what the &lt;strong&gt;SciDraft&lt;/strong&gt; Hugo theme can do
for academic and scientific publishing — for research institutions, researchers
and students. Its content is demonstration material only: readers should ignore
whether the demo content is correct.&lt;/p&gt;
&lt;p&gt;Architectures are usually chosen for what they can compute, but the sharper criterion is
what they cannot distinguish: matching the symmetries of the problem removes the need to
learn them from data. This principle is why convolution appears in vision and why attention
appears in sequences.&lt;/p&gt;</description></item></channel></rss>