Theory · Hardness of Learning and Approximations

TitleAuthors
Approximation Ratios of Graph Neural Networks for Combinatorial ProblemsRyoma Sato · Makoto Yamada · Hisashi Kashima
Deep ReLU Networks Have Surprisingly Few Activation PatternsBoris Hanin · David Rolnick
Efficient Approximation of Deep ReLU Networks for Functions on Low Dimensional ManifoldsMinshuo Chen · Haoming Jiang · Wenjing Liao · Tuo Zhao
Efficient Deep Approximation of GMMsShirin Jalali · Carl Nuzman · Iraj Saniee
Universal Invariant and Equivariant Graph Neural NetworksNicolas Keriven · Gabriel Peyré