ReLU 的布尔表达能力
The Boolean Power of ReLU
- Pontifical Catholic University(宗座天主教大学)
- IMFD
- CENIA Chile(智利CENIA)
- University of Antwerp(安特卫普大学)
- TU Wien(维也纳技术大学)
- Universiteit Hasselt(哈塞尔特大学)
机构由 AI 辅助整理,请以论文原文为准。
AI总结:
该研究证明在布尔特征有限简单无向图的布尔查询表达上,ReLU-MPLang 表达能力强于 Σ-MPLang,解决了 ReLU-MPLang 是否比 trReLU-MPLang 更强大的开放问题,表明 ReLU-GNN 比 {TrReLU,id}-GNN 表达能力更强。
AI中文摘要:
我们证明,在配备单个布尔节点特征的有限简单无向图上,对于任意由最终恒定激活函数构成的集合 Σ 及任意实系数,Σ-MPLang 中可表达的布尔查询是 ReLU-MPLang 中可表达的布尔查询的严格子类。我们由此解决了最近提出的开放问题:在布尔查询方面,ReLU-MPLang 是否比 trReLU-MPLang 更强大。特别地,这意味着在布尔特征图的布尔查询上,ReLU-GNN 比 {TrReLU,id}-GNN 具有更强的表达能力。
英文摘要:
We prove that, on finite simple undirected graphs equipped with a single Boolean node feature, the Boolean queries expressible in $Σ$-MPLang, for any collection $Σ$ of eventually constant activation functions and with arbitrary real coefficients, form a strict subclass of the Boolean queries expressible in ReLU-MPLang. We thereby settle a recently posed open problem: whether ReLU-MPLang is more powerful than trReLU-MPLang when it comes to Boolean queries. In particular, this implies that ReLU-GNNs are strictly more expressive than {TrReLU,id}-GNNs with respect to Boolean queries on Boolean-featured graphs.