Program-as-Weights: A Programming Paradigm for Fuzzy Functions - 深度解析 论文来源:ArXiv (2607.02512) 作者:Wentao Zhang, Liliana Hotsko, Woojeong Kim, Pengyu Nie, Stuart Shieber 等 分类:cs.LG, cs.AI, cs.CL 发布时间:2026-07-02T17:59:50Z 解读时间:2026年07月06日 09:08:16 📋 论文基本信息 标题:Program-as-Weights: A Programming Paradigm for Fuzzy Functions 作者:Wentao Zhang, Liliana
论文来源:ArXiv (2607.02512)
作者:Wentao Zhang, Liliana Hotsko, Woojeong Kim, Pengyu Nie, Stuart Shieber
等
分类:cs.LG, cs.AI, cs.CL
发布时间:2026-07-02T17:59:50Z
解读时间:2026年07月06日 09:08:16
标题:Program-as-Weights: A Programming Paradigm for Fuzzy Functions
作者:Wentao Zhang, Liliana Hotsko, Woojeong Kim, Pengyu Nie, Stuart Shieber
ArXiv ID:2607.02512
链接:https://arxiv.org/abs/2607.02512v1
分类:cs.LG, cs.AI, cs.CL
研究领域:GAN
本论文研究了 GAN 领域的重要问题。
Many everyday programming tasks resist clean rule-based implementation, such as alerting on important log lines, repairing malformed JSON, or ranking search results by intent, and are increasingly outsourced to large language model APIs at the cost of locality, reproducibility, and price. We propose fuzzy-function programming: compiling such a function from a natural-language specification into a compact, locally-executable neural artifact. We instantiate this paradigm with Program-as-Weights (PAW), in which a 4B compiler trained on FuzzyBench, a 10M-example dataset we release, emits parameter-efficient adapters for a frozen, lightweight interpreter. A 0.6B Qwen3 interpreter executing PAW programs matches the performance of direct prompting of Qwen3-32B, while using roughly one fiftieth of
该研究对于解决当前领域面临的挑战具有重要意义。
论文提出了一种新颖的方法来解决相关问题。
论文通过大量实验验证了所提方法的有效性。
本论文的主要创新点包括:
该方法在 GAN 领域具有广阔的应用前景。
建议读者根据自身需求深入阅读相关文献。
本论文为相关研究做出了重要贡献。
本文由 AI 自动生成。要启用 Qwen 深度分析,请配置 DASHSCOPE_API_KEY。