FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W2283360320> ?p ?o ?g. }

• W2283360320 abstract "Markov logic networks (MLNs) [86, 24] are a powerful representation combining first-order logic and probability. An MLN attaches weights to first-order formulas and views these as templates for features of Markov networks. Learning MLN structure consists of learning both formulas and their weights. This is a challenging problem because of its super-exponential search space of formulas, and the need to repeatedly learn the weights of formulas in order to evaluate them, a process that requires computationally expensive statistical inference. This thesis presents a series of algorithms that efficiently and accurately learn MLN structure. We begin by combining ideas from inductive logic programming (ILP) and feature induction in Markov networks in our MSL system. Previous approaches learn MLN structure in a disjoint manner by first learning formulas using off-the-shelf ILP systems and then learning formula weights that optimize some measure of the data's likelihood. We present an integrated approach that learns both formulas and weights to jointly optimize likelihood. Next we present the MRC system that learns latent MLN structure by discovering unary predicates in the form of clusters. MRC forms multiple clusterings of constants and relations, with each cluster corresponding to an invented predicate. We empirically show that by creating multiple clusterings, MRC outperforms previous systems. Then we apply a variant of MRC to the long-standing AI problem of extracting knowledge from text. Our system extracts simple semantic networks in an unsupervised, domain-independent manner from Web text, and introduces several techniques to scale up to the Web. After that, we incorporate the discovery of latent unary predicates into the learning of MLN clauses in the LHL system. LHL first compresses the data into a compact form by clustering the constants into high-level concepts, and then searches for clauses in the compact representation. We empirically show that LHL is more efficient and finds better formulas than previous systems. Finally, we present the LSM system that makes use of random walks to find repeated patterns in data. By restricting its search to within such patterns, LSM is able to accurately and efficiently find good formulas, improving efficiency by 2-5 orders of magnitude compared to previous systems." @default.
• W2283360320 created "2016-06-24" @default.
• W2283360320 creator A5013720432 @default.
• W2283360320 creator A5018711331 @default.
• W2283360320 date "2010-01-01" @default.
• W2283360320 modified "2023-09-26" @default.
• W2283360320 title "Structure learning in markov logic networks" @default.
• W2283360320 cites W10678874 @default.
• W2283360320 cites W127700604 @default.
• W2283360320 cites W1493490255 @default.
• W2283360320 cites W1495683013 @default.
• W2283360320 cites W1496189301 @default.
• W2283360320 cites W1503499665 @default.
• W2283360320 cites W1514107797 @default.
• W2283360320 cites W1523949738 @default.
• W2283360320 cites W1541752891 @default.
• W2283360320 cites W1543809641 @default.
• W2283360320 cites W1551893515 @default.
• W2283360320 cites W1565236324 @default.
• W2283360320 cites W1576159843 @default.
• W2283360320 cites W1585529040 @default.
• W2283360320 cites W1588165908 @default.
• W2283360320 cites W1601974683 @default.
• W2283360320 cites W1692984784 @default.
• W2283360320 cites W1817561967 @default.
• W2283360320 cites W1877678056 @default.
• W2283360320 cites W1880262756 @default.
• W2283360320 cites W1890589545 @default.
• W2283360320 cites W1964821516 @default.
• W2283360320 cites W1972714715 @default.
• W2283360320 cites W1977970897 @default.
• W2283360320 cites W1983599491 @default.
• W2283360320 cites W1999138184 @default.
• W2283360320 cites W200645999 @default.
• W2283360320 cites W2012170877 @default.
• W2283360320 cites W2012179495 @default.
• W2283360320 cites W203049729 @default.
• W2283360320 cites W2036216970 @default.
• W2283360320 cites W2038721957 @default.
• W2283360320 cites W2066720893 @default.
• W2283360320 cites W2068431743 @default.
• W2283360320 cites W2097266862 @default.
• W2283360320 cites W2097903822 @default.
• W2283360320 cites W2098678088 @default.
• W2283360320 cites W2098704351 @default.
• W2283360320 cites W2099531122 @default.
• W2283360320 cites W2100236179 @default.
• W2283360320 cites W2103160678 @default.
• W2283360320 cites W2103574271 @default.
• W2283360320 cites W2105480138 @default.
• W2283360320 cites W2105622872 @default.
• W2283360320 cites W2105767494 @default.
• W2283360320 cites W2109646372 @default.
• W2283360320 cites W2119831128 @default.
• W2283360320 cites W2121075864 @default.
• W2283360320 cites W2121465811 @default.
• W2283360320 cites W2121743855 @default.
• W2283360320 cites W2123143128 @default.
• W2283360320 cites W2125027602 @default.
• W2283360320 cites W2128245513 @default.
• W2283360320 cites W2130416410 @default.
• W2283360320 cites W2131353734 @default.
• W2283360320 cites W2132827946 @default.
• W2283360320 cites W2134647431 @default.
• W2283360320 cites W2138180870 @default.
• W2283360320 cites W2139193890 @default.
• W2283360320 cites W2140785063 @default.
• W2283360320 cites W2144429462 @default.
• W2283360320 cites W2145677303 @default.
• W2283360320 cites W2146442558 @default.
• W2283360320 cites W2150678881 @default.
• W2283360320 cites W2151447942 @default.
• W2283360320 cites W2153190022 @default.
• W2283360320 cites W2155800811 @default.
• W2283360320 cites W2157272674 @default.
• W2283360320 cites W2158292827 @default.
• W2283360320 cites W2159080219 @default.
• W2283360320 cites W2160842254 @default.
• W2283360320 cites W2164456230 @default.
• W2283360320 cites W2166741250 @default.
• W2283360320 cites W2167044614 @default.
• W2283360320 cites W2168175751 @default.
• W2283360320 cites W2168185617 @default.
• W2283360320 cites W2169992051 @default.
• W2283360320 cites W2170112109 @default.
• W2283360320 cites W2434205482 @default.
• W2283360320 cites W2484153108 @default.
• W2283360320 cites W2605441573 @default.
• W2283360320 cites W2615248003 @default.
• W2283360320 cites W27508602 @default.
• W2283360320 cites W28766783 @default.
• W2283360320 cites W2913629715 @default.
• W2283360320 cites W2914587038 @default.
• W2283360320 cites W2963587356 @default.
• W2283360320 cites W3017143921 @default.
• W2283360320 cites W3029645440 @default.
• W2283360320 cites W3098888484 @default.
• W2283360320 cites W314565566 @default.
• W2283360320 cites W3188864409 @default.
• W2283360320 cites W33696292 @default.

• next