<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Projects | Sara Candussio</title><link>https://gaoithee.github.io/saracandussio.github.io/project/</link><atom:link href="https://gaoithee.github.io/saracandussio.github.io/project/index.xml" rel="self" type="application/rss+xml"/><description>Projects</description><generator>Hugo Blox Builder (https://hugoblox.com)</generator><language>en-us</language><lastBuildDate>Thu, 05 Mar 2026 00:00:00 +0000</lastBuildDate><image><url>https://gaoithee.github.io/saracandussio.github.io/media/icon_hu7729264130191091259.png</url><title>Projects</title><link>https://gaoithee.github.io/saracandussio.github.io/project/</link></image><item><title>Distilling Formal Logic into Neural Spaces</title><link>https://gaoithee.github.io/saracandussio.github.io/project/stlenc/</link><pubDate>Thu, 05 Mar 2026 00:00:00 +0000</pubDate><guid>https://gaoithee.github.io/saracandussio.github.io/project/stlenc/</guid><description>&lt;p>Continuous neural representations of STL specifications via kernel distillation. Accepted at &lt;strong>NeSy 2026&lt;/strong>.&lt;/p></description></item><item><title>STLDec: Decoding Temporal Logic Embeddings via Transformers</title><link>https://gaoithee.github.io/saracandussio.github.io/project/stldec/</link><pubDate>Thu, 10 Jul 2025 00:00:00 +0000</pubDate><guid>https://gaoithee.github.io/saracandussio.github.io/project/stldec/</guid><description>&lt;p>Transformer-based decoder for inverting semantic embeddings of Signal Temporal Logic (STL) formulae. Published at &lt;strong>ECML-PKDD 2025&lt;/strong>.&lt;/p></description></item><item><title>OverRef: Studying Over-Refusal in Large Language Models</title><link>https://gaoithee.github.io/saracandussio.github.io/project/overref/</link><pubDate>Wed, 01 Jan 2025 00:00:00 +0000</pubDate><guid>https://gaoithee.github.io/saracandussio.github.io/project/overref/</guid><description>&lt;p>Ongoing project on &lt;strong>over-refusal&lt;/strong> in LLMs: studying when and why models refuse legitimate user queries, with benchmarking and dataset resources.&lt;/p></description></item></channel></rss>