最近几年非常热门的GNN，其实也有很悠久的历史。非常在意深度学习等研究出处的Jürgen Schmidhuber大佬的长篇考据文章里给出了这样的答案：

"deep learning architectures that can manipulate structured data, such as graphs" were proposed by Sperduti, Goller, and Küchler in the 1990s [SP93-97][GOL][KU] ... See also Pollack's even earlier relevant work [PO87-90]; compare the important work of Baldi and colleagues [BA96-03]

“可以操作结构化数据比如图的深度学习架构”是由Sperduti, Goller, 和 Küchler在1990年代提出的[SP93-97][GOL][KU]……另外参见Pollack更早的相关工作[PO87-90]；比较Baldi和同事的重要工作 [BA96-03] 。

里面提到的人物今安在呢？查了一下：

  • Alessandro Sperduti（Google Scholar），现为意大利帕多瓦大学教授。
  • Christopher Goller慕尼黑工业大学博士毕业后进了产业界，现在是一家搜索技术公司的研究总监。
  • Pierre Baldi（Google Scholar）生于意大利人，本硕都是巴黎大学，在加州理工拿了博士。现为UC Irvine的教授，是AAAI、ACM和IEEE的Fellow。H-Index 117。近年的工作偏生物信息学、AI4Science。
  • Jordan Pollack（Google Scholar）是Brandeis大学的教授。近年研究偏复杂系统和人工生命。

相关论文列表如下：

[SP93] A. Sperduti (1993). Encoding Labeled Graphs by Labeling RAAM. NIPS 1993: 1125-1132 One of the first papers on graph neural networks.

[SP94] A. Sperduti (1994). Labelling Recursive Auto-associative Memory. Connect. Sci. 6(4): 429-459 (1994)

[SP95] A. Sperduti (1995). Stability properties of labeling recursive auto-associative memory. IEEE Trans. Neural Networks 6(6): 1452-1460 (1995)

[SPG95] A. Sperduti, A. Starita, C. Goller (1995). Learning Distributed Representations for the Classification of Terms. IJCAI 1995: 509-517

[SPG96] A. Sperduti, D. Majidi, A. Starita (1996). Extended Cascade-Correlation for Syntactic and Structural Pattern Recognition. SSPR 1996: 90-99

[SPG97] A. Sperduti, A. Starita (1997). Supervised neural networks for the classification of structures. IEEE Trans. Neural Networks 8(3): 714-735, 1997.

[GOL] C. Goller & A. Küchler (1996). Learning task-dependent distributed representations by backpropagation through structure. Proceedings of International Conference on Neural Networks (ICNN'96). Vol. 1, p. 347-352 IEEE, 1996. Based on TR AR-95-02, TU Munich, 1995.

[KU] A. Küchler & C. Goller (1996). Inductive learning in symbolic domains using structure-driven recurrent neural networks. Lecture Notes in Artificial Intelligence, vol 1137. Springer, Berlin, Heidelberg.

[PO87] J. B. Pollack. On Connectionist Models of Natural Language Processing. PhD thesis, Computer Science Department, University of Illinois, Urbana, 1987.

[PO90] J. B. Pollack. Recursive Distributed Representations. Artificial Intelligence, 46(1-2):77-105, 1990.

[BA96] P. Baldi and Y. Chauvin. Hybrid Modeling, HMM/NN Architectures, and Protein Applications, Neural Computation, Vol. 8, 7, 1541-1565, (1996). One of the first papers on graph neural networks.

[BA99] P. Baldi, S. Brunak, P. Frasconi, G. Soda, and G. Pollastri. Exploiting the Past and the Future in Protein Secondary Structure Prediction, Bioinformatics, Vol. 15, 11, 937-946, (1999).

[BA03] P. Baldi and G. Pollastri. The Principled Design of Large-Scale Recursive Neural Network Architectures-DAG-RNNs and the Protein Structure Prediction Problem. Journal of Machine Learning Research, 4, 575-602, (2003).

内容中包含的图片若涉及版权问题，请及时与我们联系删除