<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>SeriesFusion | AI &amp; Machine Learning</title><description>AI &amp; Machine Learning papers published by SeriesFusion.</description><link>https://seriesfusion.com</link><item><title>A mass spectrometer loses names for most molecules it sees, so AIMe built a searchable atlas from 105.3 million possible structures.</title><link>https://seriesfusion.com/paper/cw_a83e9eb1a7220a7e635a</link><guid isPermaLink="true">https://seriesfusion.com/paper/cw_a83e9eb1a7220a7e635a</guid><description>Charting the small-molecule universe from mass spectra with neuro-symbolic AI | ML Methods. Original: 10.64898/2026.08.05.743095</description><pubDate>Fri, 07 Aug 2026 12:00:00 GMT</pubDate><category>ML Methods</category><category>ml-methods</category></item><item><title>A printed sign can hijack a robot. Across 5,670 trials, physical prompt-injection attacks fooled three vision-language planners 5.0 to 29.4 percent of the time, while simple defenses cut the risk sharply.</title><link>https://arxiv.org/abs/2608.05715</link><guid isPermaLink="true">https://arxiv.org/abs/2608.05715</guid><description>Hijacking Robots with a Piece of Paper: A Systematic Study of Physical Prompt Injection in VLM-Controlled Robots | Computer Vision. Original: 2608.05715</description><pubDate>Fri, 07 Aug 2026 12:00:00 GMT</pubDate><category>Computer Vision</category><category>computer-vision</category></item><item><title>A de-identified medical scan may still identify you. MirrorNet recovered a recognizable patient likeness from held-out image pixels, suggesting the scan itself can function like biometric data.</title><link>https://arxiv.org/abs/2608.05938</link><guid isPermaLink="true">https://arxiv.org/abs/2608.05938</guid><description>MirrorNet: Can Medical Image Anonymization Really Protect Patient Identity? | Computer Vision. Original: 2608.05938</description><pubDate>Fri, 07 Aug 2026 12:00:00 GMT</pubDate><category>Computer Vision</category><category>computer-vision</category></item><item><title>One professional&apos;s screen history shrank 86-fold in 68 milliseconds. A deterministic compiler turned 128,756 activity frames into auditable agent memory that supported 98.4 percent question-answering accuracy and zero-token replay on a matched routine.</title><link>https://arxiv.org/abs/2608.05784</link><guid isPermaLink="true">https://arxiv.org/abs/2608.05784</guid><description>Activity Frames: Deterministic Screen-Activity Compilation for Agent Memory and Replay | Natural Language Processing. Original: 2608.05784</description><pubDate>Fri, 07 Aug 2026 12:00:00 GMT</pubDate><category>Natural Language Processing</category><category>nlp</category></item></channel></rss>