2015/11/13 by Mazen Melibari, Pascal Poupart, Melibari, Mazen +5
Computer Science · #Artificial Intelligence (cs.AI) #Bayesian Modeling and Causal Inference #Data Management and Algorithms #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Rough Sets and Fuzzy Logic
paper · pdf · doi:10.48550/arxiv.1511.04412
openalex publication_date 2015/11/13 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Sum-Product Networks (SPN) have recently emerged as a new class of tractable\nprobabilistic graphical models. Unlike Bayesian networks and Markov networks\nwhere inference may be exponential in the size of the network, inference in\nSPNs is in time linear in the size of the network. Since SPNs represent\ndistributions over a fixed set of variables only, we propose dynamic sum\nproduct networks (DSPNs) as a generalization of SPNs for sequence data of\nvarying length. A DSPN consists of a template network that is repeated as many\ntimes as needed to model data sequences of any length. We present a local\nsearch technique to learn the structure of the template network. In contrast to\ndynamic Bayesian networks for which inference is generally exponential in the\nnumber of variables per time slice, DSPNs inherit the linear inference\ncomplexity of SPNs. We demonstrate the advantages of DSPNs over DBNs and other\nmodels on several datasets of sequence data.\n