Compositional Generative Mapping for Tree-Structured Data-Part I: Bottom-Up Probabilistic Modeling of Trees

被引:33
作者
Bacciu, Davide [1 ]
Micheli, Alessio [1 ]
Sperduti, Alessandro [2 ]
机构
[1] Univ Pisa, Dipartimento Informat, I-56127 Pisa, Italy
[2] Univ Padua, Dipartimento Matemat Pura & Applicata, I-35121 Padua, Italy
关键词
Bottom-up processing; hidden recursive model; hidden tree Markov model; tree-structured data; MARKOV-MODELS; FRAMEWORK; MIXTURES;
D O I
10.1109/TNNLS.2012.2222044
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
We introduce a novel compositional (recursive) probabilistic model for trees that defines an approximated bottom-up generative process from the leaves to the root of a tree. The proposed model defines contextual state transitions from the joint configuration of the children to the parent nodes. We argue that the bottom-up context postulates different probabilistic assumptions with respect to a top-down approach, leading to different representational capabilities. We discuss classes of applications that are best suited to a bottom-up approach. In particular, the bottom-up context is shown to better correlate and model the co-occurrence of substructures among the child subtrees of internal nodes. A mixed memory approximation is introduced to factorize the joint children-to-parent state transition matrix as a mixture of pairwise transitions. The proposed approach is the first practical bottom-up generative model for tree-structured data that maintains the same computational class of its top-down counterpart. Comparative experimental analyses exploiting synthetic and real-world datasets show that the proposed model can deal with deep structures better than a top-down generative model. The model is also shown to better capture structural information from real-world data comprising trees with a large out-degree. The proposed bottom-up model can be used as a fundamental building block for the development of other new powerful models.
引用
收藏
页码:1987 / 2002
页数:16
相关论文
共 29 条
[11]   Wavelet-based statistical signal processing using hidden Markov models [J].
Crouse, MS ;
Nowak, RD ;
Baraniuk, RG .
IEEE TRANSACTIONS ON SIGNAL PROCESSING, 1998, 46 (04) :886-902
[12]   Bayesian network model for semi-structured document classification [J].
Denoyer, L ;
Gallinari, P .
INFORMATION PROCESSING & MANAGEMENT, 2004, 40 (05) :807-827
[13]  
Diligenti M, 2003, IEEE T PATTERN ANAL, V25, P519, DOI 10.1109/TPAMI.2003.1190578
[14]   Computational methods for hidden Markov tree models -: An application to wavelet trees [J].
Durand, JB ;
Gonçalvès, P ;
Guèdon, Y .
IEEE TRANSACTIONS ON SIGNAL PROCESSING, 2004, 52 (09) :2551-2560
[15]   BOTTOM-UP AND TOP-DOWN TREE TRANSFORMATIONS - COMPARISON [J].
ENGELFRIET, J .
MATHEMATICAL SYSTEMS THEORY, 1975, 9 (03) :198-231
[16]   A general framework for adaptive processing of data structures [J].
Frasconi, P ;
Gori, M ;
Sperduti, A .
IEEE TRANSACTIONS ON NEURAL NETWORKS, 1998, 9 (05) :768-786
[17]   Factorial hidden Markov models [J].
Ghahramani, Z ;
Jordan, MI .
MACHINE LEARNING, 1997, 29 (2-3) :245-273
[18]   Visualization of tree-structured data through generative topographic mapping [J].
Gianniotis, Nikolaos ;
Tino, Peter .
IEEE TRANSACTIONS ON NEURAL NETWORKS, 2008, 19 (08) :1468-1493
[19]   Recursive self-organizing network models [J].
Hammer, B ;
Micheli, A ;
Sperduti, A ;
Strickert, M .
NEURAL NETWORKS, 2004, 17 (8-9) :1061-1085
[20]   A general framework for unsupervised processing of structured data [J].
Hammer, B ;
Micheli, A ;
Sperduti, A ;
Strickert, M .
NEUROCOMPUTING, 2004, 57 :3-35