<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Scientific Discovery | Web IR NLP Group @ NUS</title><link>https://wing.comp.nus.edu.sg/tag/scientific-discovery/</link><atom:link href="https://wing.comp.nus.edu.sg/tag/scientific-discovery/index.xml" rel="self" type="application/rss+xml"/><description>Scientific Discovery</description><generator>Hugo Blox Builder (https://hugoblox.com)</generator><language>en-us</language><lastBuildDate>Thu, 01 Oct 2026 00:00:00 +0000</lastBuildDate><image><url>https://wing.comp.nus.edu.sg/media/icon_hu_ce731d16d7c8719.png</url><title>Scientific Discovery</title><link>https://wing.comp.nus.edu.sg/tag/scientific-discovery/</link></image><item><title>Cost-Efficient LLM-Driven Scientific Discovery</title><link>https://wing.comp.nus.edu.sg/project/llm-scientific-discovery/</link><pubDate>Thu, 01 Oct 2026 00:00:00 +0000</pubDate><guid>https://wing.comp.nus.edu.sg/project/llm-scientific-discovery/</guid><description>&lt;p&gt;LLM-driven evolutionary search has become a powerful tool for scientific discovery: models repeatedly propose candidate solutions, which are evaluated and refined over many iterations. Relying on frontier models for every step is expensive, while lightweight models are cheaper but less capable on their own. This project studies how to coordinate models of different strengths so that discovery stays both effective and affordable.&lt;/p&gt;
&lt;p&gt;In &lt;a href="https://wing.comp.nus.edu.sg/publication/hieu-etal-2026-specevo/"&gt;SpecEvo: Speculative Evolution with Large Language Models for Cost-Efficient Scientific Discovery&lt;/a&gt; (arXiv preprint), we explore with lightweight models and consolidate with frontier reasoning. &lt;strong&gt;SpecEvo&lt;/strong&gt; lets parallel lightweight Speculators propose candidates, an Advisor distill the search trajectory into reusable guidance, and a frontier Navigator intervene only at sparse checkpoints, so that model coordination adapts as the search progresses. Across 176 tasks in four domains, it is competitive with frontier-model baselines on mathematical discovery and systems optimization at up to 80% lower API cost, achieves a 3.8% higher mean normalized score than the strongest baseline on CO-Bench combinatorial optimization, and attains the lowest NMSE in all eight LSR-Synth equation discovery settings among the main evolutionary baselines. The &lt;a href="https://github.com/langkhachhoha/SpecEvo" target="_blank" rel="noopener"&gt;SpecEvo repository&lt;/a&gt; provides the implementation.&lt;/p&gt;</description></item><item><title>SpecEvo: Speculative Evolution with Large Language Models for Cost-Efficient Scientific Discovery</title><link>https://wing.comp.nus.edu.sg/publication/hieu-etal-2026-specevo/</link><pubDate>Thu, 01 Oct 2026 00:00:00 +0000</pubDate><guid>https://wing.comp.nus.edu.sg/publication/hieu-etal-2026-specevo/</guid><description/></item></channel></rss>