<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Downscaling Attack and Defense: Turning What You See Back Into What You Get · HackerPapers discussions</title><link>https://hackerpaper.kaipfister.de/p/2010.02456/</link><description>Collected HN comments and submissions linking this paper.</description><language>en</language><lastBuildDate>Mon, 02 Nov 2026 11:16:22 GMT</lastBuildDate><atom:link href="https://hackerpaper.kaipfister.de/p/2010.02456/feed.xml" rel="self" type="application/rss+xml" /><item><title>Downscaling Attack and Defense: Turning What You See Back Into What You Get — mentioned by pixl97</title><link>https://news.ycombinator.com/item?id=46054369</link><guid isPermaLink="false">urn:hackerpapers:mention:2010.02456:46054369</guid><pubDate>Wed, 26 Nov 2025 05:07:09 GMT</pubDate><description>&lt;p&gt;&amp;gt;Not sure what this technique is called or how it&amp;#x27;s done Downscaling/downsampling attack. Commonly used against AI systems either to pass filters or poison data. https://arxiv.org/abs/2010.02456 If you know the sampling rate of the sampling filter you want to pass you can do some tricky encoding to show a different image at that particular resolution bypassing AI/content recognition systems. This said if it starts happening a lot Google will take those images/videos and feed them into a learning system that looks for those patterns and preemptively marks the video.&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2010.02456/"&gt;Paper and discussion history&lt;/a&gt;. A mention is not an endorsement.&lt;/p&gt;</description><category>cs.CR</category></item></channel></rss>
