Knowledge

Belief propagation

Source đź“ť

2807:, or loops. The initialization and scheduling of message updates must be adjusted slightly (compared with the previously described schedule for acyclic graphs) because graphs might not contain any leaves. Instead, one initializes all variable messages to 1 and uses the same message definitions above, updating all messages at every iteration (although messages coming from known leaves or tree-structured subgraphs may no longer need updating after sufficient iterations). It is easy to show that in a tree, the message definitions of this modified procedure will converge to the set of message definitions given above within a number of iterations equal to the 27: 3917:). Later, Su and Wu established the necessary and sufficient convergence conditions for synchronous GaBP and damped GaBP, as well as another sufficient convergence condition for asynchronous GaBP. For each case, the convergence condition involves verifying 1) a set (determined by A) being non-empty, 2) the spectral radius of a certain matrix being smaller than one, and 3) the singularity issue (when converting BP message into belief) does not occur. 4721: 2202: 3322:
It can then be shown that the points of convergence of the sum-product algorithm represent the points where the free energy in such a system is minimized. Similarly, it can be shown that a fixed point of the iterative belief propagation algorithm in graphs with cycles is a stationary point of a free
2814:
The precise conditions under which loopy belief propagation will converge are still not well understood; it is known that on graphs containing a single loop it converges in most cases, but the probabilities obtained might be incorrect. Several sufficient (but not necessary) conditions for convergence
2779:
In the first step, messages are passed inwards: starting at the leaves, each node passes a message along the (unique) edge towards the root node. The tree structure guarantees that it is possible to obtain messages from all other adjoining nodes before passing the message on. This continues until the
4442: 2473:
In a typical run, each message will be updated iteratively from the previous value of the neighboring messages. Different scheduling can be used for updating the messages. In the case where the graphical model is a tree, an optimal scheduling converges after computing each message exactly once (see
4380: 3317: 1970: 3671: 2740: 5390: 3796:
Convergence of the GaBP algorithm is easier to analyze (relatively to the general BP case) and there are two known sufficient convergence conditions. The first one was formulated by Weiss et al. in the year 2000, when the information matrix
3525: 1625: 416: 2765:, the belief propagation algorithm will compute the exact marginals. Furthermore, with proper scheduling of the message updates, it will terminate after two full passes through the tree. This optimal scheduling can be described as follows: 2849:, but also known as a special case of the max-product or min-sum algorithm, which solves the related problem of maximization, or most probable explanation. Instead of attempting to solve the marginal, the goal here is to find the values 1389:. These messages contain the "influence" that one variable exerts on another. The messages are computed differently depending on whether the node receiving the message is a variable node or a factor node. Keeping the same notation: 6357:
Linear Detection via Belief Propagation. Danny Bickson, Danny Dolev, Ori Shental, Paul H. Siegel and Jack K. Wolf. In the 45th Annual Allerton Conference on Communication, Control, and Computing, Allerton House, Illinois, 7 Sept..
5219: 2461:
As shown by the previous formula: the complete marginalization is reduced to a sum of products of simpler terms than the ones appearing in the full joint distribution. This is the reason that belief propagation is sometimes called
4716:{\displaystyle \forall x_{v}\in Dom(v),\;\mu _{a\to v}(x_{v})=\sum _{\mathbf {x} '_{a}:x'_{v}=x_{v}}\delta ({\text{syndrome}}({\mathbf {x} }'_{v})={\mathbf {s} })\prod _{v^{*}\in N(a)\setminus \{v\}}\mu _{v^{*}\to a}(x'_{v^{*}}).} 3347:
Improvements in the performance of belief propagation algorithms are also achievable by breaking the replicas symmetry in the distributions of the fields (messages). This generalization leads to a new kind of algorithm called
1335: 3331:
Belief propagation algorithms are normally presented as message update equations on a factor graph, involving messages between variable nodes and their neighboring factor nodes and vice versa. Considering messages between
2591: 5083: 4963: 2938: 3753: 3085: 5510: 984: 1905: 1450: 490: 3184: 6337:
Gaussian belief propagation solver for systems of linear equations. By O. Shental, D. Bickson, P. H. Siegel, J. K. Wolf, and D. Dolev, IEEE Int. Symp. on Inform. Theory (ISIT), Toronto, Canada, July 2008.
4190: 5608: 3336:
in a graph is one way of generalizing the belief propagation algorithm. There are several ways of defining the set of regions in a graph that can exchange messages. One method uses ideas introduced by
2477:
Upon convergence (if convergence happened), the estimated marginal distribution of each node is proportional to the product of all messages from adjoining factors (missing the normalization constant):
3195: 2815:
of loopy belief propagation to a unique fixed point exist. There exist graphs which will fail to converge, or which will oscillate between multiple states over repeated iterations. Techniques like
2783:
The second step involves passing the messages back out: starting at the root, messages are passed in the reverse direction. The algorithm is completed when all leaves have received their messages.
2745:
In the case where the factor graph is acyclic (i.e. is a tree or a forest), these estimated marginal actually converge to the true marginals in a finite number of iterations. This can be shown by
573: 3904: 2415: 2247: 1670: 792: 3920:
The GaBP algorithm was linked to the linear algebra domain, and it was shown that the GaBP algorithm can be viewed as an iterative algorithm for solving the linear system of equations
3559: 1495: 2602: 2596:
Likewise, the estimated joint marginal distribution of the set of variables belonging to one factor is proportional to the product of the factor and the messages from the variables:
1841: 1387: 5930: 6377:
Distributed large scale network utility maximization. D. Bickson, Y. Tock, A. Zymnis, S. Boyd and D. Dolev. In the International symposium on information theory (ISIT), July 2009.
699: 3368: 1809: 5227: 2474:
next sub-section). When the factor graph has cycles, such an optimal scheduling does not exist, and a typical choice is to update all messages simultaneously at each iteration.
2337: 2197:{\displaystyle \mu _{a\to v}(x_{v})=\sum _{\mathbf {x} '_{a}:x'_{v}=x_{v}}\left(f_{a}(\mathbf {x} '_{a})\prod _{v^{*}\in N(a)\setminus \{v\}}\mu _{v^{*}\to a}(x'_{v^{*}})\right)} 1760: 211: 4840: 819: 600: 4772: 1016: 4726:
This syndrome-based decoder doesn't require information on the received bits, thus can be adapted to quantum codes, where the only information is the measurement syndrome.
1217: 1144: 4182: 2869: 2455: 3392: 4418: 3341: 4047: 3990: 4139: 3791: 726: 647: 517: 295: 268: 2276: 1699: 2837:, which is simply belief propagation on a modified graph guaranteed to be a tree. The basic premise is to eliminate cycles by clustering them into single nodes. 6397:
Dave, Maulik A. (1 December 2006). "Review of "Information Theory, Inference, and Learning Algorithms by David J. C. MacKay", Cambridge University Press, 2003".
4438: 4107: 4087: 4067: 4010: 2296: 1965: 1945: 1925: 1719: 1490: 1470: 1355: 1257: 1237: 1184: 1164: 1111: 1091: 1071: 1036: 903: 883: 839: 746: 620: 237: 6565: 5091: 3349: 6533: 1262: 652:
Computing marginal distributions using this formula quickly becomes computationally prohibitive as the number of variables grows. For example, given 100
303: 3337: 2483: 1046:
can be represented as a factor graph by using a factor for each node with its parents or a factor for each node with its neighborhood respectively.
4968: 4848: 2881: 3682: 3936:
is the shift vector. Empirically, the GaBP algorithm is shown to converge faster than classical iterative methods like the Jacobi method, the
3003: 1947:
is defined to be the product of the factor with messages from all other nodes, marginalized over all variables except the one associated with
5399: 911: 1851: 1396: 6206:
Weiss, Yair; Freeman, William T. (October 2001). "Correctness of Belief Propagation in Gaussian Graphical Models of Arbitrary Topology".
3100: 3955:
The previous description of BP algorithm is called the codeword-based decoding, which calculates the approximate marginal probability
2943:
An algorithm that solves this problem is nearly identical to belief propagation, with the sums replaced by maxima in the definitions.
4375:{\displaystyle \forall x_{v}\in Dom(v),\;\mu _{v\to a}(x_{v})=P(X_{v})\prod _{a^{*}\in N(v)\setminus \{a\}}\mu _{a^{*}\to v}(x_{v}).} 3367:
The cluster variational method and the survey propagation algorithms are two different improvements to belief propagation. The name
6611: 6681: 5982: 5883: 3312:{\displaystyle F=U-H=\sum _{\mathbf {X} }P(\mathbf {X} )E(\mathbf {X} )+\sum _{\mathbf {X} }P(\mathbf {X} )\log P(\mathbf {X} ).} 114:
for each unobserved node (or variable), conditional on any observed nodes (or variables). Belief propagation is commonly used in
5518: 3549:
Equivalently, it can be shown that using the Gaussian model, the solution of the marginalization problem is equivalent to the
6658: 6635: 6548: 5790: 5761: 6381: 6362: 6342: 6267: 2871:
that maximizes the global function (i.e. most probable values in a probabilistic setting), and it can be defined using the
424: 3815: 2819:
can provide an approximate visualization of the progress of belief propagation and an approximate test for convergence.
2342: 2207: 1630: 751: 162: 70: 48: 6145:
Pelizzola, Alessandro (2005). "Cluster variation method in statistical physics and probabilistic graphical models".
41: 6458:
Liu, Ye-Hua; Poulin, David (22 May 2019). "Neural Belief-Propagation Decoders for Quantum Error-Correcting Codes".
2994: 3666:{\displaystyle {\underset {x}{\operatorname {argmax} }}\ P(x)={\frac {1}{Z}}\exp(-{\tfrac {1}{2}}x^{T}Ax+b^{T}x).} 3944:, and others. Additionally, the GaBP algorithm is shown to be immune to numerical problems of the preconditioned 2735:{\displaystyle p_{X_{a}}(\mathbf {x} _{a})\propto f_{a}(\mathbf {x} _{a})\prod _{v\in N(a)}\mu _{v\to a}(x_{v}).} 522: 123: 153:. While the algorithm is not exact on general graphs, it has been shown to be a useful approximate algorithm. 1814: 1360: 5385:{\displaystyle a\to v:L_{a}=(-1)^{s_{a}}2\tanh ^{-1}\prod _{v^{*}\in N(a)\setminus \{v\}}\tanh(l_{v^{*}}/2)} 6208: 5838: 2823: 2796: 905:, with edges between variables and the factors in which they appear. We can write the joint mass function: 841:
factors in a convenient way, belief propagation allows the marginals to be computed much more efficiently.
658: 1765: 6625: 6296: 2301: 1724: 170: 5629:
Braunstein, A.; MĂ©zard, M.; Zecchina, R. (2005). "Survey propagation: An algorithm for satisfiability".
4777: 3941: 6294:
Su, Qinliang; Wu, Yik-Chung (March 2015). "On convergence conditions of Gaussian belief propagation".
4732: 992: 6734: 6729: 6441: 3945: 2982: 217: 131: 5836:
Weiss, Yair (2000). "Correctness of Local Probability Propagation in Graphical Models with Loops".
3535: 1189: 1116: 797: 578: 35: 6695: 6222: 5996: 5881:
Mooij, J; Kappen, H (2007). "Sufficient Conditions for Convergence of the Sum–Product Algorithm".
4144: 3520:{\displaystyle P(x_{i})={\frac {1}{Z}}\int _{j\neq i}\exp(-{\tfrac {1}{2}}x^{T}Ax+b^{T}x)\,dx_{j}} 2852: 1620:{\displaystyle \mu _{v\to a}(x_{v})=\prod _{a^{*}\in N(v)\setminus \{a\}}\mu _{a^{*}\to v}(x_{v})} 3937: 3379:
Gaussian belief propagation is a variant of the belief propagation algorithm when the underlying
2834: 2808: 2420: 115: 6739: 6690: 6217: 5991: 2792: 2746: 52: 6071:"A Theory of Cooperative Phenomena. III. Detailed Discussions of the Cluster Variation Method" 4387: 3676:
This problem is also equivalent to the following minimization problem of the quadratic form:
1337:, whose domain is the set of values that can be taken by the random variable associated with 240: 111: 95: 4015: 3958: 6583: 6304: 6164: 6119: 6110:
Kikuchi, Ryoichi; Brush, Stephen G. (1967). "Improvement of the Cluster-Variation Method".
6082: 6043: 5939: 4112: 3764: 3550: 2804: 854: 704: 625: 495: 273: 246: 3383:. The first work analyzing this special model was the seminal work of Weiss and Freeman. 2252: 1675: 849:
Variants of the belief propagation algorithm exist for several types of graphical models (
8: 5717:"A computational model for combined causal and diagnostic reasoning in inference systems" 3802: 3380: 2955: 2762: 1043: 146: 107: 6587: 6308: 6262: 6168: 6123: 6086: 6047: 5943: 5214:{\displaystyle v\to a:l_{v}=l_{v}^{(0)}+\sum _{a^{*}\in N(v)\setminus \{a\}}(L_{a^{*}})} 6708: 6677:"Constructing free-energy approximations and generalized belief propagation algorithms" 6599: 6501: 6467: 6422: 6320: 6243: 6188: 6154: 6009: 5978:"Constructing free-energy approximations and generalized belief propagation algorithms" 5955: 5910: 5892: 5863: 5716: 5678: 5656: 5638: 4423: 4092: 4072: 4052: 3995: 2827: 2281: 1950: 1930: 1910: 1704: 1475: 1455: 1340: 1242: 1222: 1169: 1149: 1096: 1076: 1056: 1021: 888: 868: 824: 731: 605: 222: 214: 119: 6176: 5689: 3805:. The second convergence condition was formulated by Johnson et al. in 2006, when the 2958:) in a graphical model. More precisely, the marginalization problem defined above is 6654: 6631: 6544: 6493: 6485: 6414: 6235: 6180: 5855: 5786: 5757: 2846: 850: 6712: 6505: 6426: 6324: 6247: 6013: 5867: 5813:
Wainwright, M. J.; Jordan, M. I. (2007). "2.1 Probability Distributions on Graphs".
6700: 6603: 6591: 6481: 6477: 6406: 6312: 6227: 6172: 6127: 6090: 6051: 6001: 5959: 5947: 5902: 5847: 5818: 5724:
Proceedings of the Eighth International Joint Conference on Artificial Intelligence
5660: 5648: 3539: 3371:(GSP) is waiting to be assigned to the algorithm that merges both generalizations. 2970: 2959: 1330:{\displaystyle \mu _{v\to a},\mu _{a\to v}:\operatorname {Dom} (v)\to \mathbb {R} } 1039: 411:{\displaystyle p_{X_{i}}(x_{i})=\sum _{\mathbf {x} ':x'_{i}=x_{i}}p(\mathbf {x} ')} 103: 6523: 6385: 6366: 6346: 5914: 3806: 3091: 862: 165: 99: 6378: 6359: 6339: 6676: 6646: 6231: 6192: 5977: 5851: 5778: 5754:
Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference
3361: 3357: 2986: 2586:{\displaystyle p_{X_{v}}(x_{v})\propto \prod _{a\in N(v)}\mu _{a\to v}(x_{v}).} 135: 6723: 6616: 6595: 6489: 6418: 6316: 6184: 5951: 5078:{\displaystyle L_{a}=\log {\frac {u_{a\to v}(x_{v}=0)}{u_{a\to v}(x_{v}=1)}}} 4958:{\displaystyle l_{v}=\log {\frac {u_{v\to a}(x_{v}=0)}{u_{v\to a}(x_{v}=1)}}} 2969:
The memory usage of belief propagation can be reduced through the use of the
2933:{\displaystyle \operatorname {*} {\arg \max }_{\mathbf {x} }g(\mathbf {x} ).} 122:, and has demonstrated empirical success in numerous applications, including 6704: 6410: 6005: 5906: 6497: 6239: 6055: 5859: 3090:(as per the factor graph representation) can be viewed as a measure of the 2758: 858: 6261:
Malioutov, Dmitry M.; Johnson, Jason K.; Willsky, Alan S. (October 2006).
5727: 5679:"Reverend Bayes on inference engines: A distributed hierarchical approach" 2795:
graphical models, the Belief Propagation algorithm can be used in general
2772:; any non-root node which is connected to only one other node is called a 6159: 5817:. Foundations and Trends in Machine Learning. Vol. 1. pp. 5–9. 5749: 5712: 5674: 3748:{\displaystyle {\underset {x}{\operatorname {min} }}\ 1/2x^{T}Ax-b^{T}x.} 3353: 2963: 653: 142: 127: 5686:
Proceedings of the Second National Conference on Artificial Intelligence
4774:, those messages can be simplified to cause an exponential reduction of 3080:{\displaystyle P(\mathbf {X} )={\frac {1}{Z}}\prod _{f_{j}}f_{j}(x_{j})} 6034:
Kikuchi, Ryoichi (15 March 1951). "A Theory of Cooperative Phenomena".
5822: 2816: 6528:—Webpage containing recent publications as well as Matlab source code. 6131: 6095: 6070: 5652: 5505:{\displaystyle l_{v}^{(0)}=\log(P(x_{v}=0)/P(x_{v}=1))={\text{const}}} 4141:. This variation only changes the interpretation of the mass function 2768:
Before starting, the graph is oriented by designating one node as the
979:{\displaystyle p(\mathbf {x} )=\prod _{a\in F}f_{a}(\mathbf {x} _{a})} 2947: 1900:{\displaystyle \mu _{a\to v}:\operatorname {Dom} (v)\to \mathbb {R} } 1445:{\displaystyle \mu _{v\to a}:\operatorname {Dom} (v)\to \mathbb {R} } 91: 2833:
One method of exact marginalization in general graphs is called the
6472: 5897: 5643: 150: 2822:
There are other approximate methods for marginalization including
6574:
Löliger, Hans-Andrea (2004). "An Introduction to Factor Graphs".
5928:
Löliger, Hans-Andrea (2004). "An Introduction to Factor Graphs".
5815:
Graphical Models, Exponential Families, and Variational Inference
3386:
The GaBP algorithm solves the following marginalization problem:
3179:{\displaystyle E(\mathbf {X} )=-\log \prod _{f_{j}}f_{j}(x_{j}).} 2951: 2872: 857:
in particular). We describe here the variant that operates on a
6263:"Walk-sums and belief propagation in Gaussian graphical models" 1018:
is the vector of neighboring variable nodes to the factor node
145:
in 1982, who formulated it as an exact inference algorithm on
1053:
along the edges between the hidden nodes. More precisely, if
2780:
root has obtained messages from all of its adjoining nodes.
1049:
The algorithm works by passing real valued functions called
3758:
Which is also equivalent to the linear system of equations
2981:
The sum-product algorithm is related to the calculation of
2786: 844: 6647:"Understanding Belief Propagation and Its Generalizations" 6069:
Kurata, Michio; Kikuchi, Ryoichi; Watari, Tatsuro (1953).
5779:"Understanding Belief Propagation and Its Generalizations" 5603:{\displaystyle l_{v}=l_{v}^{(0)}+\sum _{a\in N(v)}(L_{a})} 2840: 5976:
Yedidia, J.S.; Freeman, W.T.; Weiss, Y.; Y. (July 2005).
6645:
Yedidia, J.S.; Freeman, W.T.; Weiss, Y. (January 2003).
5628: 6651:
Exploring Artificial Intelligence in the New Millennium
6260: 5783:
Exploring Artificial Intelligence in the New Millennium
5515:
The posterior log-likelihood ratio can be estimated as
3326: 16:
Algorithm for statistical inference on graphical models
5692:. Menlo Park, California: AAAI Press. pp. 133–136 3614: 3457: 6675:
Yedidia, J.S.; Freeman, W.T.; Weiss, Y. (July 2005).
5975: 5521: 5402: 5230: 5094: 4971: 4851: 4780: 4735: 4445: 4426: 4390: 4193: 4147: 4115: 4095: 4075: 4055: 4018: 3998: 3961: 3818: 3767: 3685: 3562: 3395: 3374: 3340:
in the physics literature, and is known as Kikuchi's
3198: 3103: 3006: 2884: 2855: 2605: 2486: 2423: 2345: 2304: 2284: 2255: 2210: 1973: 1953: 1933: 1913: 1854: 1817: 1768: 1727: 1707: 1678: 1633: 1498: 1478: 1458: 1399: 1363: 1343: 1265: 1245: 1225: 1192: 1172: 1152: 1119: 1099: 1079: 1059: 1024: 995: 914: 891: 871: 827: 800: 754: 734: 707: 661: 628: 608: 581: 525: 498: 485:{\displaystyle \mathbf {x} '=(x'_{1},\ldots ,x'_{n})} 427: 306: 276: 249: 225: 173: 6442:"Simplification of the Belief propagation algorithm" 5756:(2nd ed.). San Francisco, CA: Morgan Kaufmann. 821:. If it is known that the probability mass function 6674: 6644: 6068: 3899:{\displaystyle \rho (I-|D^{-1/2}AD^{-1/2}|)<1\,} 2950:problems like marginalization and maximization are 2845:A similar algorithm is commonly referred to as the 5602: 5504: 5384: 5213: 5077: 4957: 4834: 4766: 4715: 4432: 4412: 4374: 4176: 4133: 4109:is the decoded error. The decoded input vector is 4101: 4081: 4061: 4041: 4004: 3984: 3898: 3785: 3747: 3665: 3519: 3311: 3178: 3079: 2932: 2863: 2734: 2585: 2449: 2409: 2331: 2290: 2270: 2241: 2196: 1959: 1939: 1919: 1899: 1835: 1803: 1754: 1713: 1693: 1664: 1619: 1484: 1464: 1444: 1381: 1349: 1329: 1251: 1231: 1211: 1178: 1158: 1138: 1105: 1085: 1065: 1030: 1010: 978: 897: 877: 833: 813: 786: 740: 720: 693: 641: 614: 594: 567: 511: 484: 410: 289: 262: 231: 205: 6649:. In Lakemeyer, Gerhard; Nebel, Bernhard (eds.). 6620:. 9 July 2005. Issue 2507 (Registration required) 5781:. In Lakemeyer, Gerhard; Nebel, Bernhard (eds.). 5777:Yedidia, J.S.; Freeman, W.T.; Y. (January 2003). 2954:to solve exactly and approximately (at least for 2410:{\displaystyle \mu _{a\to v}(x_{v})=f_{a}(x_{v})} 748:and the above formula would involve summing over 6721: 5812: 5776: 3352:(SP), which have proved to be very efficient in 2901: 2242:{\displaystyle x_{v}\in \operatorname {Dom} (v)} 1665:{\displaystyle x_{v}\in \operatorname {Dom} (v)} 787:{\displaystyle 2^{99}\approx 6.34\times 10^{29}} 5710: 4012:. There is an equivalent form, which calculate 6147:Journal of Physics A: Mathematical and General 2278:is the set of neighboring (variable) nodes to 6567:A Tutorial Introduction to Belief Propagation 6379:http://www.cs.huji.ac.il/labs/danss/p2p/gabp/ 6360:http://www.cs.huji.ac.il/labs/danss/p2p/gabp/ 6340:http://www.cs.huji.ac.il/labs/danss/p2p/gabp/ 5806: 3950: 6205: 6109: 5704: 5340: 5334: 5183: 5177: 4797: 4791: 4761: 4749: 4656: 4650: 4325: 4319: 2752: 2326: 2320: 2135: 2129: 1749: 1743: 1573: 1567: 865:containing nodes corresponding to variables 6612:Communication Speed Nears Terminal Velocity 5971: 5969: 5880: 4420:is the prior error probability on variable 5770: 4483: 4231: 2976: 1701:is the set of neighboring factor nodes of 568:{\displaystyle \mathbf {x} ':x'_{i}=x_{i}} 6694: 6623: 6525:Gaussian Belief Propagation Resource Page 6471: 6457: 6287: 6221: 6158: 6144: 6094: 5995: 5896: 5667: 5642: 4069:is the syndrome of the received codeword 3895: 3503: 2799:. The algorithm is then sometimes called 1893: 1438: 1323: 71:Learn how and when to remove this message 6541:Pattern Recognition and Machine Learning 6371: 6351: 5966: 2791:Although it was originally designed for 2787:Approximate algorithm for general graphs 1811:is set to the uniform distribution over 845:Description of the sum-product algorithm 34:This article includes a list of general 6682:IEEE Transactions on Information Theory 6573: 6331: 6254: 6033: 5983:IEEE Transactions on Information Theory 5927: 5884:IEEE Transactions on Information Theory 5742: 2841:Related algorithm and complexity issues 1836:{\displaystyle \operatorname {Dom} (v)} 1382:{\displaystyle \operatorname {Dom} (v)} 1113:in the factor graph, then the messages 575:means that the sum is taken over those 492:is a vector of possible values for the 6722: 6531: 6439: 3189:The free energy of the system is then 2973:(at a small cost in time complexity). 6653:. Morgan Kaufmann. pp. 239–269. 5835: 5785:. Morgan Kaufmann. pp. 239–236. 5748: 5673: 5624: 5622: 3530:where Z is a normalization constant, 694:{\displaystyle X_{1},\ldots ,X_{100}} 6396: 6293: 6268:Journal of Machine Learning Research 6199: 3327:Generalized belief propagation (GBP) 1804:{\displaystyle \mu _{v\to a}(x_{v})} 141:The algorithm was first proposed by 20: 2803:, because graphs typically contain 2332:{\displaystyle N(a)\setminus \{v\}} 1755:{\displaystyle N(v)\setminus \{a\}} 206:{\displaystyle X_{1},\ldots ,X_{n}} 13: 6516: 6440:Filler, Tomas (17 November 2009). 5631:Random Structures & Algorithms 5619: 4835:{\displaystyle 2^{|\{v\}|+|N(v)|}} 4446: 4194: 3538:(inverse covariance matrix a.k.a. 3375:Gaussian belief propagation (GaBP) 239:, a common task is to compute the 40:it lacks sufficient corresponding 14: 6751: 5331: 5174: 4647: 4316: 3094:present in a system, computed as 2317: 2126: 1740: 1564: 4767:{\displaystyle x_{i}\in \{0,1\}} 4609: 4587: 4526: 3299: 3279: 3266: 3250: 3236: 3223: 3111: 3014: 2920: 2907: 2857: 2659: 2628: 2080: 2016: 1011:{\displaystyle \mathbf {x} _{a}} 998: 963: 922: 803: 584: 528: 430: 397: 350: 25: 6576:IEEE Signal Processing Magazine 6532:Bishop, Christopher M. (2006). 6451: 6433: 6390: 6138: 6112:The Journal of Chemical Physics 6103: 6075:The Journal of Chemical Physics 6062: 6027: 5931:IEEE Signal Processing Magazine 5730:. Vol. 1. pp. 190–193 4184:. Explicitly, the messages are 6630:. Cambridge University Press. 6543:. Springer. pp. 359–418. 6482:10.1103/physrevlett.122.200501 5921: 5874: 5829: 5597: 5584: 5579: 5573: 5551: 5545: 5491: 5488: 5469: 5458: 5439: 5433: 5419: 5413: 5379: 5351: 5328: 5322: 5266: 5256: 5234: 5208: 5188: 5171: 5165: 5136: 5130: 5098: 5069: 5050: 5042: 5029: 5010: 5002: 4949: 4930: 4922: 4909: 4890: 4882: 4826: 4822: 4816: 4809: 4801: 4787: 4707: 4684: 4676: 4644: 4638: 4614: 4601: 4581: 4573: 4513: 4500: 4492: 4477: 4471: 4407: 4394: 4366: 4353: 4345: 4313: 4307: 4283: 4270: 4261: 4248: 4240: 4225: 4219: 4171: 4158: 4036: 4029: 4022: 3979: 3972: 3965: 3932:is the information matrix and 3886: 3882: 3832: 3822: 3657: 3607: 3585: 3579: 3500: 3450: 3412: 3399: 3369:generalized survey propagation 3303: 3295: 3283: 3275: 3254: 3246: 3240: 3232: 3170: 3157: 3115: 3107: 3074: 3061: 3018: 3010: 2924: 2916: 2726: 2713: 2705: 2692: 2686: 2669: 2654: 2638: 2623: 2577: 2564: 2556: 2543: 2537: 2517: 2504: 2404: 2391: 2375: 2362: 2354: 2314: 2308: 2265: 2259: 2236: 2230: 2186: 2163: 2155: 2123: 2117: 2093: 2075: 2003: 1990: 1982: 1889: 1886: 1880: 1863: 1830: 1824: 1798: 1785: 1777: 1737: 1731: 1688: 1682: 1659: 1653: 1614: 1601: 1593: 1561: 1555: 1528: 1515: 1507: 1434: 1431: 1425: 1408: 1376: 1370: 1319: 1316: 1310: 1293: 1274: 1201: 1128: 1093:is a factor node connected to 973: 958: 926: 918: 701:, computing a single marginal 479: 441: 405: 392: 337: 324: 124:low-density parity-check codes 1: 6534:"Chapter 8: Graphical models" 5612: 2997:. A probability distribution 1212:{\displaystyle \mu _{a\to v}} 1139:{\displaystyle \mu _{v\to a}} 814:{\displaystyle \mathbf {x} '} 595:{\displaystyle \mathbf {x} '} 156: 5728:IJCAI-83: Karlsruhe, Germany 4845:Define log-likelihood ratio 4177:{\displaystyle f_{a}(X_{a})} 2864:{\displaystyle \mathbf {x} } 7: 6297:IEEE Trans. Signal Process. 6177:10.1088/0305-4470/38/33/R01 2464:sum-product message passing 2450:{\displaystyle x_{v}=x_{a}} 270:. The marginal of a single 88:sum–product message passing 10: 6756: 6232:10.1162/089976601750541769 5852:10.1162/089976600300015880 3992:, given received codeword 3951:Syndrome-based BP decoding 3942:successive over-relaxation 3381:distributions are Gaussian 1259:are real-valued functions 622:th coordinate is equal to 6627:Iterative Receiver Design 6610:Mackenzie, Dana (2005). " 6564:Coughlan, James. (2009). 3946:conjugate gradient method 3938:Gauss–Seidel method 2753:Exact algorithm for trees 218:probability mass function 6624:Wymeersch, Henk (2007). 6596:10.1109/MSP.2004.1267047 6522:Bickson, Danny. (2009). 6317:10.1109/TSP.2015.2389755 5952:10.1109/msp.2004.1267047 4413:{\displaystyle P(X_{v})} 3536:positive definite matrix 3342:cluster variation method 2946:It is worth noting that 2801:loopy belief propagation 6705:10.1109/TIT.2005.850085 6460:Physical Review Letters 6411:10.1145/1189056.1189063 6006:10.1109/TIT.2005.850085 5907:10.1109/TIT.2007.909166 5690:AAAI-82: Pittsburgh, PA 2977:Relation to free energy 2835:junction tree algorithm 1073:is a variable node and 116:artificial intelligence 90:, is a message-passing 55:more precise citations. 6056:10.1103/PhysRev.81.988 5604: 5506: 5386: 5215: 5079: 4959: 4836: 4768: 4717: 4434: 4414: 4376: 4178: 4135: 4103: 4083: 4063: 4043: 4042:{\displaystyle P(e|s)} 4006: 3986: 3985:{\displaystyle P(x|X)} 3900: 3787: 3749: 3667: 3521: 3323:energy approximation. 3313: 3180: 3081: 2934: 2865: 2747:mathematical induction 2736: 2587: 2451: 2411: 2333: 2292: 2272: 2243: 2198: 1961: 1941: 1921: 1901: 1837: 1805: 1756: 1715: 1695: 1666: 1621: 1486: 1466: 1446: 1383: 1351: 1331: 1253: 1233: 1213: 1180: 1160: 1140: 1107: 1087: 1067: 1032: 1012: 980: 899: 879: 861:. A factor graph is a 835: 815: 788: 742: 722: 695: 643: 616: 596: 569: 513: 486: 412: 291: 264: 241:marginal distributions 233: 207: 161:Given a finite set of 5605: 5507: 5387: 5216: 5080: 4960: 4837: 4769: 4718: 4435: 4415: 4377: 4179: 4136: 4134:{\displaystyle x=X+e} 4104: 4084: 4064: 4044: 4007: 3987: 3901: 3788: 3786:{\displaystyle Ax=b.} 3750: 3668: 3546:is the shift vector. 3522: 3314: 3181: 3082: 2935: 2866: 2757:In the case when the 2737: 2588: 2468:sum-product algorithm 2452: 2417:, since in this case 2412: 2334: 2293: 2273: 2244: 2199: 1962: 1942: 1922: 1902: 1838: 1806: 1757: 1716: 1696: 1667: 1622: 1487: 1467: 1452:from a variable node 1447: 1384: 1352: 1332: 1254: 1234: 1214: 1181: 1161: 1141: 1108: 1088: 1068: 1033: 1013: 981: 900: 880: 836: 816: 789: 743: 723: 721:{\displaystyle X_{i}} 696: 644: 642:{\displaystyle x_{i}} 617: 597: 570: 514: 512:{\displaystyle X_{i}} 487: 413: 292: 290:{\displaystyle X_{i}} 265: 263:{\displaystyle X_{i}} 234: 208: 112:marginal distribution 6384:14 June 2011 at the 6365:14 June 2011 at the 6345:14 June 2011 at the 5519: 5400: 5228: 5092: 4969: 4849: 4778: 4733: 4729:In the binary case, 4443: 4424: 4388: 4191: 4145: 4113: 4093: 4073: 4053: 4016: 3996: 3959: 3816: 3765: 3683: 3560: 3553:assignment problem: 3393: 3196: 3101: 3004: 2962:and maximization is 2882: 2853: 2603: 2484: 2421: 2343: 2302: 2282: 2271:{\displaystyle N(a)} 2253: 2208: 1971: 1951: 1931: 1911: 1852: 1815: 1766: 1725: 1705: 1694:{\displaystyle N(v)} 1676: 1631: 1496: 1476: 1456: 1397: 1361: 1341: 1263: 1243: 1223: 1190: 1170: 1150: 1117: 1097: 1077: 1057: 1022: 993: 912: 889: 869: 855:Markov random fields 825: 798: 794:possible values for 752: 732: 705: 659: 626: 606: 579: 523: 496: 425: 304: 274: 247: 223: 171: 149:, later extended to 110:. It calculates the 108:Markov random fields 6588:2004ISPM...21...28L 6309:2015ITSP...63.1144S 6169:2005JPhA...38R.309P 6124:1967JChPh..47..195K 6087:1953JChPh..21..434K 6048:1951PhRv...81..988K 5944:2004ISPM...21...28L 5555: 5423: 5140: 4706: 4600: 4554: 4538: 3803:diagonally dominant 2828:Monte Carlo methods 2824:variational methods 2185: 2092: 2044: 2028: 1927:to a variable node 1907:from a factor node 1044:Markov random field 551: 519:, and the notation 478: 456: 373: 134:approximation, and 6209:Neural Computation 5839:Neural Computation 5823:10.1561/2200000001 5600: 5583: 5535: 5502: 5403: 5382: 5344: 5211: 5187: 5120: 5075: 4955: 4842:in the complexity 4832: 4764: 4713: 4687: 4660: 4584: 4569: 4542: 4524: 4430: 4410: 4372: 4329: 4174: 4131: 4099: 4079: 4059: 4039: 4002: 3982: 3896: 3783: 3745: 3694: 3663: 3623: 3571: 3517: 3466: 3350:survey propagation 3309: 3271: 3228: 3176: 3146: 3077: 3050: 2995:partition function 2930: 2861: 2732: 2696: 2583: 2547: 2447: 2407: 2329: 2288: 2268: 2239: 2194: 2166: 2139: 2078: 2059: 2032: 2014: 1957: 1937: 1917: 1897: 1833: 1801: 1752: 1711: 1691: 1662: 1617: 1577: 1482: 1462: 1442: 1379: 1347: 1327: 1249: 1229: 1209: 1176: 1156: 1136: 1103: 1083: 1063: 1028: 1008: 976: 947: 895: 875: 831: 811: 784: 738: 718: 691: 639: 612: 592: 565: 539: 509: 482: 466: 444: 408: 388: 361: 287: 260: 229: 203: 120:information theory 84:Belief propagation 6660:978-1-55860-811-5 6637:978-0-521-87315-4 6550:978-0-387-31073-2 6216:(10): 2173–2200. 6153:(33): R309–R339. 6132:10.1063/1.1711845 6096:10.1063/1.1698926 5891:(12): 4422–4437. 5792:978-1-55860-811-5 5763:978-1-55860-479-7 5653:10.1002/rsa.20057 5559: 5500: 5301: 5144: 5073: 4953: 4617: 4579: 4519: 4433:{\displaystyle v} 4286: 4102:{\displaystyle e} 4082:{\displaystyle X} 4062:{\displaystyle s} 4005:{\displaystyle X} 3698: 3687: 3622: 3599: 3575: 3564: 3465: 3426: 3260: 3217: 3130: 3034: 3032: 2888: 2847:Viterbi algorithm 2672: 2523: 2291:{\displaystyle a} 2096: 2009: 1960:{\displaystyle v} 1940:{\displaystyle v} 1920:{\displaystyle a} 1714:{\displaystyle v} 1534: 1485:{\displaystyle a} 1472:to a factor node 1465:{\displaystyle v} 1350:{\displaystyle v} 1252:{\displaystyle v} 1232:{\displaystyle a} 1186:and the messages 1179:{\displaystyle a} 1159:{\displaystyle v} 1106:{\displaystyle v} 1086:{\displaystyle a} 1066:{\displaystyle v} 1031:{\displaystyle a} 932: 898:{\displaystyle F} 878:{\displaystyle V} 851:Bayesian networks 834:{\displaystyle p} 741:{\displaystyle p} 615:{\displaystyle i} 343: 297:is defined to be 232:{\displaystyle p} 104:Bayesian networks 81: 80: 73: 6747: 6735:Graphical models 6730:Graph algorithms 6716: 6698: 6689:(7): 2282–2312. 6671: 6669: 6667: 6641: 6607: 6561: 6559: 6557: 6538: 6510: 6509: 6475: 6455: 6449: 6448: 6446: 6437: 6431: 6430: 6394: 6388: 6375: 6369: 6355: 6349: 6335: 6329: 6328: 6303:(5): 1144–1155. 6291: 6285: 6284: 6282: 6280: 6258: 6252: 6251: 6225: 6203: 6197: 6196: 6162: 6160:cond-mat/0508216 6142: 6136: 6135: 6107: 6101: 6100: 6098: 6066: 6060: 6059: 6031: 6025: 6024: 6022: 6020: 5999: 5990:(7): 2282–2312. 5973: 5964: 5963: 5925: 5919: 5918: 5900: 5878: 5872: 5871: 5833: 5827: 5826: 5810: 5804: 5803: 5801: 5799: 5774: 5768: 5767: 5746: 5740: 5739: 5737: 5735: 5721: 5708: 5702: 5701: 5699: 5697: 5683: 5671: 5665: 5664: 5646: 5626: 5609: 5607: 5606: 5601: 5596: 5595: 5582: 5554: 5543: 5531: 5530: 5511: 5509: 5508: 5503: 5501: 5498: 5481: 5480: 5465: 5451: 5450: 5422: 5411: 5391: 5389: 5388: 5383: 5375: 5370: 5369: 5368: 5367: 5343: 5315: 5314: 5297: 5296: 5281: 5280: 5279: 5278: 5252: 5251: 5220: 5218: 5217: 5212: 5207: 5206: 5205: 5204: 5186: 5158: 5157: 5139: 5128: 5116: 5115: 5084: 5082: 5081: 5076: 5074: 5072: 5062: 5061: 5049: 5048: 5032: 5022: 5021: 5009: 5008: 4992: 4981: 4980: 4964: 4962: 4961: 4956: 4954: 4952: 4942: 4941: 4929: 4928: 4912: 4902: 4901: 4889: 4888: 4872: 4861: 4860: 4841: 4839: 4838: 4833: 4831: 4830: 4829: 4812: 4804: 4790: 4773: 4771: 4770: 4765: 4745: 4744: 4722: 4720: 4719: 4714: 4702: 4701: 4700: 4683: 4682: 4675: 4674: 4659: 4631: 4630: 4613: 4612: 4596: 4591: 4590: 4580: 4577: 4568: 4567: 4566: 4550: 4534: 4529: 4512: 4511: 4499: 4498: 4458: 4457: 4439: 4437: 4436: 4431: 4419: 4417: 4416: 4411: 4406: 4405: 4381: 4379: 4378: 4373: 4365: 4364: 4352: 4351: 4344: 4343: 4328: 4300: 4299: 4282: 4281: 4260: 4259: 4247: 4246: 4206: 4205: 4183: 4181: 4180: 4175: 4170: 4169: 4157: 4156: 4140: 4138: 4137: 4132: 4108: 4106: 4105: 4100: 4088: 4086: 4085: 4080: 4068: 4066: 4065: 4060: 4048: 4046: 4045: 4040: 4032: 4011: 4009: 4008: 4003: 3991: 3989: 3988: 3983: 3975: 3905: 3903: 3902: 3897: 3885: 3880: 3879: 3875: 3856: 3855: 3851: 3835: 3792: 3790: 3789: 3784: 3754: 3752: 3751: 3746: 3738: 3737: 3719: 3718: 3706: 3696: 3695: 3672: 3670: 3669: 3664: 3653: 3652: 3634: 3633: 3624: 3615: 3600: 3592: 3573: 3572: 3540:precision matrix 3526: 3524: 3523: 3518: 3516: 3515: 3496: 3495: 3477: 3476: 3467: 3458: 3443: 3442: 3427: 3419: 3411: 3410: 3318: 3316: 3315: 3310: 3302: 3282: 3270: 3269: 3253: 3239: 3227: 3226: 3185: 3183: 3182: 3177: 3169: 3168: 3156: 3155: 3145: 3144: 3143: 3114: 3086: 3084: 3083: 3078: 3073: 3072: 3060: 3059: 3049: 3048: 3047: 3033: 3025: 3017: 2971:Island algorithm 2939: 2937: 2936: 2931: 2923: 2912: 2911: 2910: 2904: 2889: 2886: 2870: 2868: 2867: 2862: 2860: 2741: 2739: 2738: 2733: 2725: 2724: 2712: 2711: 2695: 2668: 2667: 2662: 2653: 2652: 2637: 2636: 2631: 2622: 2621: 2620: 2619: 2592: 2590: 2589: 2584: 2576: 2575: 2563: 2562: 2546: 2516: 2515: 2503: 2502: 2501: 2500: 2456: 2454: 2453: 2448: 2446: 2445: 2433: 2432: 2416: 2414: 2413: 2408: 2403: 2402: 2390: 2389: 2374: 2373: 2361: 2360: 2338: 2336: 2335: 2330: 2297: 2295: 2294: 2289: 2277: 2275: 2274: 2269: 2248: 2246: 2245: 2240: 2220: 2219: 2203: 2201: 2200: 2195: 2193: 2189: 2181: 2180: 2179: 2162: 2161: 2154: 2153: 2138: 2110: 2109: 2088: 2083: 2074: 2073: 2058: 2057: 2056: 2040: 2024: 2019: 2002: 2001: 1989: 1988: 1966: 1964: 1963: 1958: 1946: 1944: 1943: 1938: 1926: 1924: 1923: 1918: 1906: 1904: 1903: 1898: 1896: 1870: 1869: 1842: 1840: 1839: 1834: 1810: 1808: 1807: 1802: 1797: 1796: 1784: 1783: 1761: 1759: 1758: 1753: 1720: 1718: 1717: 1712: 1700: 1698: 1697: 1692: 1671: 1669: 1668: 1663: 1643: 1642: 1626: 1624: 1623: 1618: 1613: 1612: 1600: 1599: 1592: 1591: 1576: 1548: 1547: 1527: 1526: 1514: 1513: 1491: 1489: 1488: 1483: 1471: 1469: 1468: 1463: 1451: 1449: 1448: 1443: 1441: 1415: 1414: 1388: 1386: 1385: 1380: 1356: 1354: 1353: 1348: 1336: 1334: 1333: 1328: 1326: 1300: 1299: 1281: 1280: 1258: 1256: 1255: 1250: 1238: 1236: 1235: 1230: 1218: 1216: 1215: 1210: 1208: 1207: 1185: 1183: 1182: 1177: 1165: 1163: 1162: 1157: 1145: 1143: 1142: 1137: 1135: 1134: 1112: 1110: 1109: 1104: 1092: 1090: 1089: 1084: 1072: 1070: 1069: 1064: 1040:Bayesian network 1037: 1035: 1034: 1029: 1017: 1015: 1014: 1009: 1007: 1006: 1001: 985: 983: 982: 977: 972: 971: 966: 957: 956: 946: 925: 904: 902: 901: 896: 884: 882: 881: 876: 840: 838: 837: 832: 820: 818: 817: 812: 810: 806: 793: 791: 790: 785: 783: 782: 764: 763: 747: 745: 744: 739: 727: 725: 724: 719: 717: 716: 700: 698: 697: 692: 690: 689: 671: 670: 654:binary variables 648: 646: 645: 640: 638: 637: 621: 619: 618: 613: 601: 599: 598: 593: 591: 587: 574: 572: 571: 566: 564: 563: 547: 535: 531: 518: 516: 515: 510: 508: 507: 491: 489: 488: 483: 474: 452: 437: 433: 417: 415: 414: 409: 404: 400: 387: 386: 385: 369: 357: 353: 336: 335: 323: 322: 321: 320: 296: 294: 293: 288: 286: 285: 269: 267: 266: 261: 259: 258: 238: 236: 235: 230: 212: 210: 209: 204: 202: 201: 183: 182: 166:random variables 100:graphical models 86:, also known as 76: 69: 65: 62: 56: 51:this article by 42:inline citations 29: 28: 21: 6755: 6754: 6750: 6749: 6748: 6746: 6745: 6744: 6720: 6719: 6665: 6663: 6661: 6638: 6555: 6553: 6551: 6536: 6519: 6517:Further reading 6514: 6513: 6456: 6452: 6444: 6438: 6434: 6399:ACM SIGACT News 6395: 6391: 6386:Wayback Machine 6376: 6372: 6367:Wayback Machine 6356: 6352: 6347:Wayback Machine 6336: 6332: 6292: 6288: 6278: 6276: 6259: 6255: 6204: 6200: 6143: 6139: 6108: 6104: 6067: 6063: 6042:(6): 988–1003. 6036:Physical Review 6032: 6028: 6018: 6016: 5974: 5967: 5926: 5922: 5879: 5875: 5834: 5830: 5811: 5807: 5797: 5795: 5793: 5775: 5771: 5764: 5747: 5743: 5733: 5731: 5719: 5709: 5705: 5695: 5693: 5681: 5672: 5668: 5627: 5620: 5615: 5591: 5587: 5563: 5544: 5539: 5526: 5522: 5520: 5517: 5516: 5497: 5476: 5472: 5461: 5446: 5442: 5412: 5407: 5401: 5398: 5397: 5371: 5363: 5359: 5358: 5354: 5310: 5306: 5305: 5289: 5285: 5274: 5270: 5269: 5265: 5247: 5243: 5229: 5226: 5225: 5200: 5196: 5195: 5191: 5153: 5149: 5148: 5129: 5124: 5111: 5107: 5093: 5090: 5089: 5057: 5053: 5038: 5034: 5033: 5017: 5013: 4998: 4994: 4993: 4991: 4976: 4972: 4970: 4967: 4966: 4937: 4933: 4918: 4914: 4913: 4897: 4893: 4878: 4874: 4873: 4871: 4856: 4852: 4850: 4847: 4846: 4825: 4808: 4800: 4786: 4785: 4781: 4779: 4776: 4775: 4740: 4736: 4734: 4731: 4730: 4696: 4692: 4691: 4670: 4666: 4665: 4661: 4626: 4622: 4621: 4608: 4607: 4592: 4586: 4585: 4576: 4562: 4558: 4546: 4530: 4525: 4523: 4507: 4503: 4488: 4484: 4453: 4449: 4444: 4441: 4440: 4425: 4422: 4421: 4401: 4397: 4389: 4386: 4385: 4360: 4356: 4339: 4335: 4334: 4330: 4295: 4291: 4290: 4277: 4273: 4255: 4251: 4236: 4232: 4201: 4197: 4192: 4189: 4188: 4165: 4161: 4152: 4148: 4146: 4143: 4142: 4114: 4111: 4110: 4094: 4091: 4090: 4074: 4071: 4070: 4054: 4051: 4050: 4028: 4017: 4014: 4013: 3997: 3994: 3993: 3971: 3960: 3957: 3956: 3953: 3881: 3871: 3864: 3860: 3847: 3840: 3836: 3831: 3817: 3814: 3813: 3807:spectral radius 3766: 3763: 3762: 3733: 3729: 3714: 3710: 3702: 3686: 3684: 3681: 3680: 3648: 3644: 3629: 3625: 3613: 3591: 3563: 3561: 3558: 3557: 3534:is a symmetric 3511: 3507: 3491: 3487: 3472: 3468: 3456: 3432: 3428: 3418: 3406: 3402: 3394: 3391: 3390: 3377: 3329: 3298: 3278: 3265: 3264: 3249: 3235: 3222: 3221: 3197: 3194: 3193: 3164: 3160: 3151: 3147: 3139: 3135: 3134: 3110: 3102: 3099: 3098: 3092:internal energy 3068: 3064: 3055: 3051: 3043: 3039: 3038: 3024: 3013: 3005: 3002: 3001: 2979: 2919: 2906: 2905: 2894: 2893: 2885: 2883: 2880: 2879: 2856: 2854: 2851: 2850: 2843: 2789: 2755: 2720: 2716: 2701: 2697: 2676: 2663: 2658: 2657: 2648: 2644: 2632: 2627: 2626: 2615: 2611: 2610: 2606: 2604: 2601: 2600: 2571: 2567: 2552: 2548: 2527: 2511: 2507: 2496: 2492: 2491: 2487: 2485: 2482: 2481: 2441: 2437: 2428: 2424: 2422: 2419: 2418: 2398: 2394: 2385: 2381: 2369: 2365: 2350: 2346: 2344: 2341: 2340: 2339:is empty, then 2303: 2300: 2299: 2283: 2280: 2279: 2254: 2251: 2250: 2215: 2211: 2209: 2206: 2205: 2175: 2171: 2170: 2149: 2145: 2144: 2140: 2105: 2101: 2100: 2084: 2079: 2069: 2065: 2064: 2060: 2052: 2048: 2036: 2020: 2015: 2013: 1997: 1993: 1978: 1974: 1972: 1969: 1968: 1952: 1949: 1948: 1932: 1929: 1928: 1912: 1909: 1908: 1892: 1859: 1855: 1853: 1850: 1849: 1816: 1813: 1812: 1792: 1788: 1773: 1769: 1767: 1764: 1763: 1726: 1723: 1722: 1706: 1703: 1702: 1677: 1674: 1673: 1638: 1634: 1632: 1629: 1628: 1608: 1604: 1587: 1583: 1582: 1578: 1543: 1539: 1538: 1522: 1518: 1503: 1499: 1497: 1494: 1493: 1477: 1474: 1473: 1457: 1454: 1453: 1437: 1404: 1400: 1398: 1395: 1394: 1362: 1359: 1358: 1342: 1339: 1338: 1322: 1289: 1285: 1270: 1266: 1264: 1261: 1260: 1244: 1241: 1240: 1224: 1221: 1220: 1197: 1193: 1191: 1188: 1187: 1171: 1168: 1167: 1151: 1148: 1147: 1124: 1120: 1118: 1115: 1114: 1098: 1095: 1094: 1078: 1075: 1074: 1058: 1055: 1054: 1023: 1020: 1019: 1002: 997: 996: 994: 991: 990: 967: 962: 961: 952: 948: 936: 921: 913: 910: 909: 890: 887: 886: 870: 867: 866: 863:bipartite graph 847: 826: 823: 822: 802: 801: 799: 796: 795: 778: 774: 759: 755: 753: 750: 749: 733: 730: 729: 712: 708: 706: 703: 702: 685: 681: 666: 662: 660: 657: 656: 633: 629: 627: 624: 623: 607: 604: 603: 583: 582: 580: 577: 576: 559: 555: 543: 527: 526: 524: 521: 520: 503: 499: 497: 494: 493: 470: 448: 429: 428: 426: 423: 422: 396: 395: 381: 377: 365: 349: 348: 347: 331: 327: 316: 312: 311: 307: 305: 302: 301: 281: 277: 275: 272: 271: 254: 250: 248: 245: 244: 224: 221: 220: 197: 193: 178: 174: 172: 169: 168: 159: 94:for performing 77: 66: 60: 57: 47:Please help to 46: 30: 26: 17: 12: 11: 5: 6753: 6743: 6742: 6737: 6732: 6718: 6717: 6672: 6659: 6642: 6636: 6621: 6608: 6571: 6562: 6549: 6529: 6518: 6515: 6512: 6511: 6466:(20): 200501. 6450: 6432: 6389: 6370: 6350: 6330: 6286: 6253: 6198: 6137: 6118:(1): 195–203. 6102: 6081:(3): 434–448. 6061: 6026: 5965: 5920: 5873: 5828: 5805: 5791: 5769: 5762: 5741: 5703: 5666: 5637:(2): 201–226. 5617: 5616: 5614: 5611: 5599: 5594: 5590: 5586: 5581: 5578: 5575: 5572: 5569: 5566: 5562: 5558: 5553: 5550: 5547: 5542: 5538: 5534: 5529: 5525: 5513: 5512: 5496: 5493: 5490: 5487: 5484: 5479: 5475: 5471: 5468: 5464: 5460: 5457: 5454: 5449: 5445: 5441: 5438: 5435: 5432: 5429: 5426: 5421: 5418: 5415: 5410: 5406: 5393: 5392: 5381: 5378: 5374: 5366: 5362: 5357: 5353: 5350: 5347: 5342: 5339: 5336: 5333: 5330: 5327: 5324: 5321: 5318: 5313: 5309: 5304: 5300: 5295: 5292: 5288: 5284: 5277: 5273: 5268: 5264: 5261: 5258: 5255: 5250: 5246: 5242: 5239: 5236: 5233: 5222: 5221: 5210: 5203: 5199: 5194: 5190: 5185: 5182: 5179: 5176: 5173: 5170: 5167: 5164: 5161: 5156: 5152: 5147: 5143: 5138: 5135: 5132: 5127: 5123: 5119: 5114: 5110: 5106: 5103: 5100: 5097: 5071: 5068: 5065: 5060: 5056: 5052: 5047: 5044: 5041: 5037: 5031: 5028: 5025: 5020: 5016: 5012: 5007: 5004: 5001: 4997: 4990: 4987: 4984: 4979: 4975: 4951: 4948: 4945: 4940: 4936: 4932: 4927: 4924: 4921: 4917: 4911: 4908: 4905: 4900: 4896: 4892: 4887: 4884: 4881: 4877: 4870: 4867: 4864: 4859: 4855: 4828: 4824: 4821: 4818: 4815: 4811: 4807: 4803: 4799: 4796: 4793: 4789: 4784: 4763: 4760: 4757: 4754: 4751: 4748: 4743: 4739: 4724: 4723: 4712: 4709: 4705: 4699: 4695: 4690: 4686: 4681: 4678: 4673: 4669: 4664: 4658: 4655: 4652: 4649: 4646: 4643: 4640: 4637: 4634: 4629: 4625: 4620: 4616: 4611: 4606: 4603: 4599: 4595: 4589: 4583: 4575: 4572: 4565: 4561: 4557: 4553: 4549: 4545: 4541: 4537: 4533: 4528: 4522: 4518: 4515: 4510: 4506: 4502: 4497: 4494: 4491: 4487: 4482: 4479: 4476: 4473: 4470: 4467: 4464: 4461: 4456: 4452: 4448: 4429: 4409: 4404: 4400: 4396: 4393: 4382: 4371: 4368: 4363: 4359: 4355: 4350: 4347: 4342: 4338: 4333: 4327: 4324: 4321: 4318: 4315: 4312: 4309: 4306: 4303: 4298: 4294: 4289: 4285: 4280: 4276: 4272: 4269: 4266: 4263: 4258: 4254: 4250: 4245: 4242: 4239: 4235: 4230: 4227: 4224: 4221: 4218: 4215: 4212: 4209: 4204: 4200: 4196: 4173: 4168: 4164: 4160: 4155: 4151: 4130: 4127: 4124: 4121: 4118: 4098: 4078: 4058: 4038: 4035: 4031: 4027: 4024: 4021: 4001: 3981: 3978: 3974: 3970: 3967: 3964: 3952: 3949: 3907: 3906: 3894: 3891: 3888: 3884: 3878: 3874: 3870: 3867: 3863: 3859: 3854: 3850: 3846: 3843: 3839: 3834: 3830: 3827: 3824: 3821: 3809:of the matrix 3794: 3793: 3782: 3779: 3776: 3773: 3770: 3756: 3755: 3744: 3741: 3736: 3732: 3728: 3725: 3722: 3717: 3713: 3709: 3705: 3701: 3693: 3690: 3674: 3673: 3662: 3659: 3656: 3651: 3647: 3643: 3640: 3637: 3632: 3628: 3621: 3618: 3612: 3609: 3606: 3603: 3598: 3595: 3590: 3587: 3584: 3581: 3578: 3570: 3567: 3528: 3527: 3514: 3510: 3506: 3502: 3499: 3494: 3490: 3486: 3483: 3480: 3475: 3471: 3464: 3461: 3455: 3452: 3449: 3446: 3441: 3438: 3435: 3431: 3425: 3422: 3417: 3414: 3409: 3405: 3401: 3398: 3376: 3373: 3362:graph coloring 3358:satisfiability 3356:problems like 3328: 3325: 3320: 3319: 3308: 3305: 3301: 3297: 3294: 3291: 3288: 3285: 3281: 3277: 3274: 3268: 3263: 3259: 3256: 3252: 3248: 3245: 3242: 3238: 3234: 3231: 3225: 3220: 3216: 3213: 3210: 3207: 3204: 3201: 3187: 3186: 3175: 3172: 3167: 3163: 3159: 3154: 3150: 3142: 3138: 3133: 3129: 3126: 3123: 3120: 3117: 3113: 3109: 3106: 3088: 3087: 3076: 3071: 3067: 3063: 3058: 3054: 3046: 3042: 3037: 3031: 3028: 3023: 3020: 3016: 3012: 3009: 2987:thermodynamics 2978: 2975: 2956:relative error 2941: 2940: 2929: 2926: 2922: 2918: 2915: 2909: 2903: 2900: 2897: 2892: 2859: 2842: 2839: 2788: 2785: 2754: 2751: 2743: 2742: 2731: 2728: 2723: 2719: 2715: 2710: 2707: 2704: 2700: 2694: 2691: 2688: 2685: 2682: 2679: 2675: 2671: 2666: 2661: 2656: 2651: 2647: 2643: 2640: 2635: 2630: 2625: 2618: 2614: 2609: 2594: 2593: 2582: 2579: 2574: 2570: 2566: 2561: 2558: 2555: 2551: 2545: 2542: 2539: 2536: 2533: 2530: 2526: 2522: 2519: 2514: 2510: 2506: 2499: 2495: 2490: 2459: 2458: 2444: 2440: 2436: 2431: 2427: 2406: 2401: 2397: 2393: 2388: 2384: 2380: 2377: 2372: 2368: 2364: 2359: 2356: 2353: 2349: 2328: 2325: 2322: 2319: 2316: 2313: 2310: 2307: 2287: 2267: 2264: 2261: 2258: 2238: 2235: 2232: 2229: 2226: 2223: 2218: 2214: 2192: 2188: 2184: 2178: 2174: 2169: 2165: 2160: 2157: 2152: 2148: 2143: 2137: 2134: 2131: 2128: 2125: 2122: 2119: 2116: 2113: 2108: 2104: 2099: 2095: 2091: 2087: 2082: 2077: 2072: 2068: 2063: 2055: 2051: 2047: 2043: 2039: 2035: 2031: 2027: 2023: 2018: 2012: 2008: 2005: 2000: 1996: 1992: 1987: 1984: 1981: 1977: 1956: 1936: 1916: 1895: 1891: 1888: 1885: 1882: 1879: 1876: 1873: 1868: 1865: 1862: 1858: 1845: 1844: 1832: 1829: 1826: 1823: 1820: 1800: 1795: 1791: 1787: 1782: 1779: 1776: 1772: 1762:is empty then 1751: 1748: 1745: 1742: 1739: 1736: 1733: 1730: 1710: 1690: 1687: 1684: 1681: 1661: 1658: 1655: 1652: 1649: 1646: 1641: 1637: 1616: 1611: 1607: 1603: 1598: 1595: 1590: 1586: 1581: 1575: 1572: 1569: 1566: 1563: 1560: 1557: 1554: 1551: 1546: 1542: 1537: 1533: 1530: 1525: 1521: 1517: 1512: 1509: 1506: 1502: 1492:is defined by 1481: 1461: 1440: 1436: 1433: 1430: 1427: 1424: 1421: 1418: 1413: 1410: 1407: 1403: 1378: 1375: 1372: 1369: 1366: 1346: 1325: 1321: 1318: 1315: 1312: 1309: 1306: 1303: 1298: 1295: 1292: 1288: 1284: 1279: 1276: 1273: 1269: 1248: 1228: 1206: 1203: 1200: 1196: 1175: 1155: 1133: 1130: 1127: 1123: 1102: 1082: 1062: 1027: 1005: 1000: 987: 986: 975: 970: 965: 960: 955: 951: 945: 942: 939: 935: 931: 928: 924: 920: 917: 894: 874: 846: 843: 830: 809: 805: 781: 777: 773: 770: 767: 762: 758: 737: 715: 711: 688: 684: 680: 677: 674: 669: 665: 636: 632: 611: 590: 586: 562: 558: 554: 550: 546: 542: 538: 534: 530: 506: 502: 481: 477: 473: 469: 465: 462: 459: 455: 451: 447: 443: 440: 436: 432: 419: 418: 407: 403: 399: 394: 391: 384: 380: 376: 372: 368: 364: 360: 356: 352: 346: 342: 339: 334: 330: 326: 319: 315: 310: 284: 280: 257: 253: 228: 200: 196: 192: 189: 186: 181: 177: 158: 155: 136:satisfiability 79: 78: 33: 31: 24: 15: 9: 6: 4: 3: 2: 6752: 6741: 6740:Coding theory 6738: 6736: 6733: 6731: 6728: 6727: 6725: 6714: 6710: 6706: 6702: 6697: 6696:10.1.1.3.5650 6692: 6688: 6684: 6683: 6678: 6673: 6662: 6656: 6652: 6648: 6643: 6639: 6633: 6629: 6628: 6622: 6619: 6618: 6617:New Scientist 6613: 6609: 6605: 6601: 6597: 6593: 6589: 6585: 6581: 6577: 6572: 6569: 6568: 6563: 6552: 6546: 6542: 6535: 6530: 6527: 6526: 6521: 6520: 6507: 6503: 6499: 6495: 6491: 6487: 6483: 6479: 6474: 6469: 6465: 6461: 6454: 6443: 6436: 6428: 6424: 6420: 6416: 6412: 6408: 6404: 6400: 6393: 6387: 6383: 6380: 6374: 6368: 6364: 6361: 6354: 6348: 6344: 6341: 6334: 6326: 6322: 6318: 6314: 6310: 6306: 6302: 6299: 6298: 6290: 6274: 6270: 6269: 6264: 6257: 6249: 6245: 6241: 6237: 6233: 6229: 6224: 6223:10.1.1.44.794 6219: 6215: 6211: 6210: 6202: 6194: 6190: 6186: 6182: 6178: 6174: 6170: 6166: 6161: 6156: 6152: 6148: 6141: 6133: 6129: 6125: 6121: 6117: 6113: 6106: 6097: 6092: 6088: 6084: 6080: 6076: 6072: 6065: 6057: 6053: 6049: 6045: 6041: 6037: 6030: 6015: 6011: 6007: 6003: 5998: 5997:10.1.1.3.5650 5993: 5989: 5985: 5984: 5979: 5972: 5970: 5961: 5957: 5953: 5949: 5945: 5941: 5937: 5933: 5932: 5924: 5916: 5912: 5908: 5904: 5899: 5894: 5890: 5886: 5885: 5877: 5869: 5865: 5861: 5857: 5853: 5849: 5845: 5841: 5840: 5832: 5824: 5820: 5816: 5809: 5794: 5788: 5784: 5780: 5773: 5765: 5759: 5755: 5751: 5745: 5729: 5725: 5718: 5714: 5711:Kim, Jin H.; 5707: 5691: 5687: 5680: 5676: 5670: 5662: 5658: 5654: 5650: 5645: 5640: 5636: 5632: 5625: 5623: 5618: 5610: 5592: 5588: 5576: 5570: 5567: 5564: 5560: 5556: 5548: 5540: 5536: 5532: 5527: 5523: 5494: 5485: 5482: 5477: 5473: 5466: 5462: 5455: 5452: 5447: 5443: 5436: 5430: 5427: 5424: 5416: 5408: 5404: 5395: 5394: 5376: 5372: 5364: 5360: 5355: 5348: 5345: 5337: 5325: 5319: 5316: 5311: 5307: 5302: 5298: 5293: 5290: 5286: 5282: 5275: 5271: 5262: 5259: 5253: 5248: 5244: 5240: 5237: 5231: 5224: 5223: 5201: 5197: 5192: 5180: 5168: 5162: 5159: 5154: 5150: 5145: 5141: 5133: 5125: 5121: 5117: 5112: 5108: 5104: 5101: 5095: 5088: 5087: 5086: 5066: 5063: 5058: 5054: 5045: 5039: 5035: 5026: 5023: 5018: 5014: 5005: 4999: 4995: 4988: 4985: 4982: 4977: 4973: 4946: 4943: 4938: 4934: 4925: 4919: 4915: 4906: 4903: 4898: 4894: 4885: 4879: 4875: 4868: 4865: 4862: 4857: 4853: 4843: 4819: 4813: 4805: 4794: 4782: 4758: 4755: 4752: 4746: 4741: 4737: 4727: 4710: 4703: 4697: 4693: 4688: 4679: 4671: 4667: 4662: 4653: 4641: 4635: 4632: 4627: 4623: 4618: 4604: 4597: 4593: 4570: 4563: 4559: 4555: 4551: 4547: 4543: 4539: 4535: 4531: 4520: 4516: 4508: 4504: 4495: 4489: 4485: 4480: 4474: 4468: 4465: 4462: 4459: 4454: 4450: 4427: 4402: 4398: 4391: 4383: 4369: 4361: 4357: 4348: 4340: 4336: 4331: 4322: 4310: 4304: 4301: 4296: 4292: 4287: 4278: 4274: 4267: 4264: 4256: 4252: 4243: 4237: 4233: 4228: 4222: 4216: 4213: 4210: 4207: 4202: 4198: 4187: 4186: 4185: 4166: 4162: 4153: 4149: 4128: 4125: 4122: 4119: 4116: 4096: 4076: 4056: 4033: 4025: 4019: 3999: 3976: 3968: 3962: 3948: 3947: 3943: 3939: 3935: 3931: 3927: 3923: 3918: 3916: 3912: 3892: 3889: 3876: 3872: 3868: 3865: 3861: 3857: 3852: 3848: 3844: 3841: 3837: 3828: 3825: 3819: 3812: 3811: 3810: 3808: 3804: 3800: 3780: 3777: 3774: 3771: 3768: 3761: 3760: 3759: 3742: 3739: 3734: 3730: 3726: 3723: 3720: 3715: 3711: 3707: 3703: 3699: 3691: 3688: 3679: 3678: 3677: 3660: 3654: 3649: 3645: 3641: 3638: 3635: 3630: 3626: 3619: 3616: 3610: 3604: 3601: 3596: 3593: 3588: 3582: 3576: 3568: 3565: 3556: 3555: 3554: 3552: 3547: 3545: 3541: 3537: 3533: 3512: 3508: 3504: 3497: 3492: 3488: 3484: 3481: 3478: 3473: 3469: 3462: 3459: 3453: 3447: 3444: 3439: 3436: 3433: 3429: 3423: 3420: 3415: 3407: 3403: 3396: 3389: 3388: 3387: 3384: 3382: 3372: 3370: 3365: 3363: 3359: 3355: 3351: 3345: 3343: 3339: 3335: 3324: 3306: 3292: 3289: 3286: 3272: 3261: 3257: 3243: 3229: 3218: 3214: 3211: 3208: 3205: 3202: 3199: 3192: 3191: 3190: 3173: 3165: 3161: 3152: 3148: 3140: 3136: 3131: 3127: 3124: 3121: 3118: 3104: 3097: 3096: 3095: 3093: 3069: 3065: 3056: 3052: 3044: 3040: 3035: 3029: 3026: 3021: 3007: 3000: 2999: 2998: 2996: 2992: 2988: 2984: 2974: 2972: 2967: 2965: 2961: 2957: 2953: 2949: 2944: 2927: 2913: 2898: 2895: 2890: 2878: 2877: 2876: 2874: 2848: 2838: 2836: 2831: 2829: 2825: 2820: 2818: 2812: 2811:of the tree. 2810: 2806: 2802: 2798: 2794: 2784: 2781: 2777: 2775: 2771: 2766: 2764: 2760: 2750: 2748: 2729: 2721: 2717: 2708: 2702: 2698: 2689: 2683: 2680: 2677: 2673: 2664: 2649: 2645: 2641: 2633: 2616: 2612: 2607: 2599: 2598: 2597: 2580: 2572: 2568: 2559: 2553: 2549: 2540: 2534: 2531: 2528: 2524: 2520: 2512: 2508: 2497: 2493: 2488: 2480: 2479: 2478: 2475: 2471: 2469: 2465: 2442: 2438: 2434: 2429: 2425: 2399: 2395: 2386: 2382: 2378: 2370: 2366: 2357: 2351: 2347: 2323: 2311: 2305: 2285: 2262: 2256: 2233: 2227: 2224: 2221: 2216: 2212: 2190: 2182: 2176: 2172: 2167: 2158: 2150: 2146: 2141: 2132: 2120: 2114: 2111: 2106: 2102: 2097: 2089: 2085: 2070: 2066: 2061: 2053: 2049: 2045: 2041: 2037: 2033: 2029: 2025: 2021: 2010: 2006: 1998: 1994: 1985: 1979: 1975: 1954: 1934: 1914: 1883: 1877: 1874: 1871: 1866: 1860: 1856: 1847: 1846: 1827: 1821: 1818: 1793: 1789: 1780: 1774: 1770: 1746: 1734: 1728: 1708: 1685: 1679: 1656: 1650: 1647: 1644: 1639: 1635: 1609: 1605: 1596: 1588: 1584: 1579: 1570: 1558: 1552: 1549: 1544: 1540: 1535: 1531: 1523: 1519: 1510: 1504: 1500: 1479: 1459: 1428: 1422: 1419: 1416: 1411: 1405: 1401: 1392: 1391: 1390: 1373: 1367: 1364: 1344: 1313: 1307: 1304: 1301: 1296: 1290: 1286: 1282: 1277: 1271: 1267: 1246: 1226: 1204: 1198: 1194: 1173: 1153: 1131: 1125: 1121: 1100: 1080: 1060: 1052: 1047: 1045: 1041: 1025: 1003: 968: 953: 949: 943: 940: 937: 933: 929: 915: 908: 907: 906: 892: 872: 864: 860: 856: 852: 842: 828: 807: 779: 775: 771: 768: 765: 760: 756: 735: 713: 709: 686: 682: 678: 675: 672: 667: 663: 655: 650: 634: 630: 609: 588: 560: 556: 552: 548: 544: 540: 536: 532: 504: 500: 475: 471: 467: 463: 460: 457: 453: 449: 445: 438: 434: 401: 389: 382: 378: 374: 370: 366: 362: 358: 354: 344: 340: 332: 328: 317: 313: 308: 300: 299: 298: 282: 278: 255: 251: 242: 226: 219: 216: 198: 194: 190: 187: 184: 179: 175: 167: 164: 154: 152: 148: 144: 139: 137: 133: 129: 125: 121: 117: 113: 109: 105: 101: 97: 93: 89: 85: 75: 72: 64: 54: 50: 44: 43: 37: 32: 23: 22: 19: 6686: 6680: 6664:. Retrieved 6650: 6626: 6615: 6582:(1): 28–41. 6579: 6575: 6566: 6554:. Retrieved 6540: 6524: 6463: 6459: 6453: 6435: 6405:(4): 34–36. 6402: 6398: 6392: 6373: 6353: 6333: 6300: 6295: 6289: 6277:. Retrieved 6272: 6266: 6256: 6213: 6207: 6201: 6150: 6146: 6140: 6115: 6111: 6105: 6078: 6074: 6064: 6039: 6035: 6029: 6017:. Retrieved 5987: 5981: 5938:(1): 28–41. 5935: 5929: 5923: 5888: 5882: 5876: 5843: 5837: 5831: 5814: 5808: 5796:. Retrieved 5782: 5772: 5753: 5750:Pearl, Judea 5744: 5732:. Retrieved 5723: 5713:Pearl, Judea 5706: 5694:. Retrieved 5685: 5675:Pearl, Judea 5669: 5634: 5630: 5514: 4844: 4728: 4725: 3954: 3933: 3929: 3925: 3921: 3919: 3914: 3910: 3908: 3798: 3795: 3757: 3675: 3548: 3543: 3531: 3529: 3385: 3378: 3366: 3346: 3333: 3330: 3321: 3188: 3089: 2990: 2980: 2968: 2945: 2942: 2844: 2832: 2821: 2813: 2800: 2790: 2782: 2778: 2773: 2769: 2767: 2759:factor graph 2756: 2744: 2595: 2476: 2472: 2467: 2463: 2460: 1050: 1048: 988: 885:and factors 859:factor graph 848: 651: 420: 160: 140: 87: 83: 82: 67: 58: 39: 18: 6275:: 2031–2064 5846:(1): 1–41. 3354:NP-complete 2983:free energy 2964:NP-complete 2960:#P-complete 2817:EXIT charts 143:Judea Pearl 132:free energy 128:turbo codes 53:introducing 6724:Categories 6556:2 December 6473:1811.07835 5898:cs/0504030 5644:cs/0212002 5613:References 1848:A message 1393:A message 1357:, denoted 157:Motivation 102:, such as 61:April 2009 36:references 6691:CiteSeerX 6490:0031-9007 6419:0163-5700 6218:CiteSeerX 6185:0305-4470 5992:CiteSeerX 5568:∈ 5561:∑ 5431:⁡ 5365:∗ 5349:⁡ 5332:∖ 5317:∈ 5312:∗ 5303:∏ 5299:⁡ 5291:− 5260:− 5235:→ 5202:∗ 5175:∖ 5160:∈ 5155:∗ 5146:∑ 5099:→ 5043:→ 5003:→ 4989:⁡ 4923:→ 4883:→ 4869:⁡ 4747:∈ 4698:∗ 4677:→ 4672:∗ 4663:μ 4648:∖ 4633:∈ 4628:∗ 4619:∏ 4571:δ 4521:∑ 4493:→ 4486:μ 4460:∈ 4447:∀ 4346:→ 4341:∗ 4332:μ 4317:∖ 4302:∈ 4297:∗ 4288:∏ 4241:→ 4234:μ 4208:∈ 4195:∀ 3866:− 3842:− 3829:− 3820:ρ 3727:− 3611:− 3605:⁡ 3454:− 3448:⁡ 3437:≠ 3430:∫ 3290:⁡ 3262:∑ 3219:∑ 3209:− 3132:∏ 3128:⁡ 3122:− 3036:∏ 2948:inference 2899:⁡ 2891:⁡ 2706:→ 2699:μ 2681:∈ 2674:∏ 2642:∝ 2557:→ 2550:μ 2532:∈ 2525:∏ 2521:∝ 2466:, or the 2355:→ 2348:μ 2318:∖ 2228:⁡ 2222:∈ 2177:∗ 2156:→ 2151:∗ 2142:μ 2127:∖ 2112:∈ 2107:∗ 2098:∏ 2011:∑ 1983:→ 1976:μ 1890:→ 1878:⁡ 1864:→ 1857:μ 1822:⁡ 1778:→ 1771:μ 1741:∖ 1651:⁡ 1645:∈ 1594:→ 1589:∗ 1580:μ 1565:∖ 1550:∈ 1545:∗ 1536:∏ 1508:→ 1501:μ 1435:→ 1423:⁡ 1409:→ 1402:μ 1368:⁡ 1320:→ 1308:⁡ 1294:→ 1287:μ 1275:→ 1268:μ 1202:→ 1195:μ 1129:→ 1122:μ 941:∈ 934:∏ 772:× 766:≈ 676:… 461:… 345:∑ 188:… 151:polytrees 96:inference 92:algorithm 6713:52835993 6666:30 March 6506:53959182 6498:31172756 6427:10570465 6382:Archived 6363:Archived 6343:Archived 6325:12055229 6279:28 March 6248:10624764 6240:11570995 6019:28 March 6014:52835993 5868:15402308 5860:10636932 5798:30 March 5752:(1988). 5734:20 March 5715:(1983). 5696:28 March 5677:(1982). 4704:′ 4598:′ 4578:syndrome 4552:′ 4536:′ 4049:, where 2809:diameter 2249:, where 2183:′ 2090:′ 2042:′ 2026:′ 1672:, where 1051:messages 808:′ 589:′ 549:′ 533:′ 476:′ 454:′ 435:′ 402:′ 371:′ 355:′ 163:discrete 6604:7722934 6584:Bibcode 6305:Bibcode 6165:Bibcode 6120:Bibcode 6083:Bibcode 6044:Bibcode 5960:7722934 5940:Bibcode 5661:6601396 5085:, then 4384:where 3913:= diag( 3338:Kikuchi 3334:regions 2993:be the 2952:NP-hard 2873:arg max 2793:acyclic 243:of the 49:improve 6711:  6693:  6657:  6634:  6602:  6547:  6504:  6496:  6488:  6425:  6417:  6323:  6246:  6238:  6220:  6191:  6183:  6012:  5994:  5958:  5913:  5866:  5858:  5789:  5760:  5659:  5396:where 3928:where 3909:where 3697:  3574:  3566:argmax 3542:) and 2989:. Let 2805:cycles 2797:graphs 1038:. Any 989:where 728:using 602:whose 421:where 38:, but 6709:S2CID 6600:S2CID 6537:(PDF) 6502:S2CID 6468:arXiv 6445:(PDF) 6423:S2CID 6321:S2CID 6244:S2CID 6189:S2CID 6155:arXiv 6010:S2CID 5956:S2CID 5915:57228 5911:S2CID 5893:arXiv 5864:S2CID 5720:(PDF) 5682:(PDF) 5657:S2CID 5639:arXiv 5499:const 2761:is a 2298:. If 1721:. If 1219:from 1146:from 215:joint 213:with 147:trees 6668:2009 6655:ISBN 6632:ISBN 6558:2023 6545:ISBN 6494:PMID 6486:ISSN 6415:ISSN 6281:2009 6236:PMID 6181:ISSN 6021:2009 5856:PMID 5800:2009 5787:ISBN 5758:ISBN 5736:2016 5698:2009 5346:tanh 5287:tanh 4089:and 3890:< 3360:and 2826:and 2774:leaf 2770:root 2763:tree 2204:for 1627:for 853:and 769:6.34 118:and 106:and 6701:doi 6614:", 6592:doi 6478:doi 6464:122 6407:doi 6313:doi 6228:doi 6193:942 6173:doi 6128:doi 6091:doi 6052:doi 6002:doi 5948:doi 5903:doi 5848:doi 5819:doi 5649:doi 5428:log 4986:log 4866:log 3801:is 3689:min 3602:exp 3551:MAP 3445:exp 3287:log 3125:log 2985:in 2902:max 2896:arg 2225:Dom 1875:Dom 1819:Dom 1648:Dom 1420:Dom 1365:Dom 1305:Dom 1239:to 1166:to 1042:or 687:100 98:on 6726:: 6707:. 6699:. 6687:51 6685:. 6679:. 6598:. 6590:. 6580:21 6578:. 6539:. 6500:. 6492:. 6484:. 6476:. 6462:. 6421:. 6413:. 6403:37 6401:. 6319:. 6311:. 6301:63 6271:. 6265:. 6242:. 6234:. 6226:. 6214:13 6212:. 6187:. 6179:. 6171:. 6163:. 6151:38 6149:. 6126:. 6116:47 6114:. 6089:. 6079:21 6077:. 6073:. 6050:. 6040:81 6038:. 6008:. 6000:. 5988:51 5986:. 5980:. 5968:^ 5954:. 5946:. 5936:21 5934:. 5909:. 5901:. 5889:53 5887:. 5862:. 5854:. 5844:12 5842:. 5726:. 5722:. 5688:. 5684:. 5655:. 5647:. 5635:27 5633:. 5621:^ 4965:, 3940:, 3924:= 3922:Ax 3364:. 3344:. 2966:. 2875:: 2830:. 2776:. 2749:. 2470:. 780:29 776:10 761:99 649:. 138:. 130:, 126:, 6715:. 6703:: 6670:. 6640:. 6606:. 6594:: 6586:: 6570:. 6560:. 6508:. 6480:: 6470:: 6447:. 6429:. 6409:: 6327:. 6315:: 6307:: 6283:. 6273:7 6250:. 6230:: 6195:. 6175:: 6167:: 6157:: 6134:. 6130:: 6122:: 6099:. 6093:: 6085:: 6058:. 6054:: 6046:: 6023:. 6004:: 5962:. 5950:: 5942:: 5917:. 5905:: 5895:: 5870:. 5850:: 5825:. 5821:: 5802:. 5766:. 5738:. 5700:. 5663:. 5651:: 5641:: 5598:) 5593:a 5589:L 5585:( 5580:) 5577:v 5574:( 5571:N 5565:a 5557:+ 5552:) 5549:0 5546:( 5541:v 5537:l 5533:= 5528:v 5524:l 5495:= 5492:) 5489:) 5486:1 5483:= 5478:v 5474:x 5470:( 5467:P 5463:/ 5459:) 5456:0 5453:= 5448:v 5444:x 5440:( 5437:P 5434:( 5425:= 5420:) 5417:0 5414:( 5409:v 5405:l 5380:) 5377:2 5373:/ 5361:v 5356:l 5352:( 5341:} 5338:v 5335:{ 5329:) 5326:a 5323:( 5320:N 5308:v 5294:1 5283:2 5276:a 5272:s 5267:) 5263:1 5257:( 5254:= 5249:a 5245:L 5241:: 5238:v 5232:a 5209:) 5198:a 5193:L 5189:( 5184:} 5181:a 5178:{ 5172:) 5169:v 5166:( 5163:N 5151:a 5142:+ 5137:) 5134:0 5131:( 5126:v 5122:l 5118:= 5113:v 5109:l 5105:: 5102:a 5096:v 5070:) 5067:1 5064:= 5059:v 5055:x 5051:( 5046:v 5040:a 5036:u 5030:) 5027:0 5024:= 5019:v 5015:x 5011:( 5006:v 5000:a 4996:u 4983:= 4978:a 4974:L 4950:) 4947:1 4944:= 4939:v 4935:x 4931:( 4926:a 4920:v 4916:u 4910:) 4907:0 4904:= 4899:v 4895:x 4891:( 4886:a 4880:v 4876:u 4863:= 4858:v 4854:l 4827:| 4823:) 4820:v 4817:( 4814:N 4810:| 4806:+ 4802:| 4798:} 4795:v 4792:{ 4788:| 4783:2 4762:} 4759:1 4756:, 4753:0 4750:{ 4742:i 4738:x 4711:. 4708:) 4694:v 4689:x 4685:( 4680:a 4668:v 4657:} 4654:v 4651:{ 4645:) 4642:a 4639:( 4636:N 4624:v 4615:) 4610:s 4605:= 4602:) 4594:v 4588:x 4582:( 4574:( 4564:v 4560:x 4556:= 4548:v 4544:x 4540:: 4532:a 4527:x 4517:= 4514:) 4509:v 4505:x 4501:( 4496:v 4490:a 4481:, 4478:) 4475:v 4472:( 4469:m 4466:o 4463:D 4455:v 4451:x 4428:v 4408:) 4403:v 4399:X 4395:( 4392:P 4370:. 4367:) 4362:v 4358:x 4354:( 4349:v 4337:a 4326:} 4323:a 4320:{ 4314:) 4311:v 4308:( 4305:N 4293:a 4284:) 4279:v 4275:X 4271:( 4268:P 4265:= 4262:) 4257:v 4253:x 4249:( 4244:a 4238:v 4229:, 4226:) 4223:v 4220:( 4217:m 4214:o 4211:D 4203:v 4199:x 4172:) 4167:a 4163:X 4159:( 4154:a 4150:f 4129:e 4126:+ 4123:X 4120:= 4117:x 4097:e 4077:X 4057:s 4037:) 4034:s 4030:| 4026:e 4023:( 4020:P 4000:X 3980:) 3977:X 3973:| 3969:x 3966:( 3963:P 3934:b 3930:A 3926:b 3915:A 3911:D 3893:1 3887:) 3883:| 3877:2 3873:/ 3869:1 3862:D 3858:A 3853:2 3849:/ 3845:1 3838:D 3833:| 3826:I 3823:( 3799:A 3781:. 3778:b 3775:= 3772:x 3769:A 3743:. 3740:x 3735:T 3731:b 3724:x 3721:A 3716:T 3712:x 3708:2 3704:/ 3700:1 3692:x 3661:. 3658:) 3655:x 3650:T 3646:b 3642:+ 3639:x 3636:A 3631:T 3627:x 3620:2 3617:1 3608:( 3597:Z 3594:1 3589:= 3586:) 3583:x 3580:( 3577:P 3569:x 3544:b 3532:A 3513:j 3509:x 3505:d 3501:) 3498:x 3493:T 3489:b 3485:+ 3482:x 3479:A 3474:T 3470:x 3463:2 3460:1 3451:( 3440:i 3434:j 3424:Z 3421:1 3416:= 3413:) 3408:i 3404:x 3400:( 3397:P 3307:. 3304:) 3300:X 3296:( 3293:P 3284:) 3280:X 3276:( 3273:P 3267:X 3258:+ 3255:) 3251:X 3247:( 3244:E 3241:) 3237:X 3233:( 3230:P 3224:X 3215:= 3212:H 3206:U 3203:= 3200:F 3174:. 3171:) 3166:j 3162:x 3158:( 3153:j 3149:f 3141:j 3137:f 3119:= 3116:) 3112:X 3108:( 3105:E 3075:) 3070:j 3066:x 3062:( 3057:j 3053:f 3045:j 3041:f 3030:Z 3027:1 3022:= 3019:) 3015:X 3011:( 3008:P 2991:Z 2928:. 2925:) 2921:x 2917:( 2914:g 2908:x 2887:* 2858:x 2730:. 2727:) 2722:v 2718:x 2714:( 2709:a 2703:v 2693:) 2690:a 2687:( 2684:N 2678:v 2670:) 2665:a 2660:x 2655:( 2650:a 2646:f 2639:) 2634:a 2629:x 2624:( 2617:a 2613:X 2608:p 2581:. 2578:) 2573:v 2569:x 2565:( 2560:v 2554:a 2544:) 2541:v 2538:( 2535:N 2529:a 2518:) 2513:v 2509:x 2505:( 2498:v 2494:X 2489:p 2457:. 2443:a 2439:x 2435:= 2430:v 2426:x 2405:) 2400:v 2396:x 2392:( 2387:a 2383:f 2379:= 2376:) 2371:v 2367:x 2363:( 2358:v 2352:a 2327:} 2324:v 2321:{ 2315:) 2312:a 2309:( 2306:N 2286:a 2266:) 2263:a 2260:( 2257:N 2237:) 2234:v 2231:( 2217:v 2213:x 2191:) 2187:) 2173:v 2168:x 2164:( 2159:a 2147:v 2136:} 2133:v 2130:{ 2124:) 2121:a 2118:( 2115:N 2103:v 2094:) 2086:a 2081:x 2076:( 2071:a 2067:f 2062:( 2054:v 2050:x 2046:= 2038:v 2034:x 2030:: 2022:a 2017:x 2007:= 2004:) 1999:v 1995:x 1991:( 1986:v 1980:a 1967:, 1955:v 1935:v 1915:a 1894:R 1887:) 1884:v 1881:( 1872:: 1867:v 1861:a 1843:. 1831:) 1828:v 1825:( 1799:) 1794:v 1790:x 1786:( 1781:a 1775:v 1750:} 1747:a 1744:{ 1738:) 1735:v 1732:( 1729:N 1709:v 1689:) 1686:v 1683:( 1680:N 1660:) 1657:v 1654:( 1640:v 1636:x 1615:) 1610:v 1606:x 1602:( 1597:v 1585:a 1574:} 1571:a 1568:{ 1562:) 1559:v 1556:( 1553:N 1541:a 1532:= 1529:) 1524:v 1520:x 1516:( 1511:a 1505:v 1480:a 1460:v 1439:R 1432:) 1429:v 1426:( 1417:: 1412:a 1406:v 1377:) 1374:v 1371:( 1345:v 1324:R 1317:) 1314:v 1311:( 1302:: 1297:v 1291:a 1283:, 1278:a 1272:v 1247:v 1227:a 1205:v 1199:a 1174:a 1154:v 1132:a 1126:v 1101:v 1081:a 1061:v 1026:a 1004:a 999:x 974:) 969:a 964:x 959:( 954:a 950:f 944:F 938:a 930:= 927:) 923:x 919:( 916:p 893:F 873:V 829:p 804:x 757:2 736:p 714:i 710:X 683:X 679:, 673:, 668:1 664:X 635:i 631:x 610:i 585:x 561:i 557:x 553:= 545:i 541:x 537:: 529:x 505:i 501:X 480:) 472:n 468:x 464:, 458:, 450:1 446:x 442:( 439:= 431:x 406:) 398:x 393:( 390:p 383:i 379:x 375:= 367:i 363:x 359:: 351:x 341:= 338:) 333:i 329:x 325:( 318:i 314:X 309:p 283:i 279:X 256:i 252:X 227:p 199:n 195:X 191:, 185:, 180:1 176:X 74:) 68:( 63:) 59:( 45:.

Index

references
inline citations
improve
introducing
Learn how and when to remove this message
algorithm
inference
graphical models
Bayesian networks
Markov random fields
marginal distribution
artificial intelligence
information theory
low-density parity-check codes
turbo codes
free energy
satisfiability
Judea Pearl
trees
polytrees
discrete
random variables
joint
probability mass function
marginal distributions
binary variables
Bayesian networks
Markov random fields
factor graph
bipartite graph

Text is available under the Creative Commons Attribution-ShareAlike License. Additional terms may apply.

↑