China
Africa
Explore chapters and articles related to this topic
Advances in Radio Localization Techniques
Published in Chao Gao, Guorong Zhao, Hassen Fourati, Cooperative Localization and Navigation, 2019
Cesar Vargas-Rosales, Rafaela Villalpando-Hernandez, Mort Naraghi-Pour
In [39], the self-localization problem is addressed by generating graphs that contain the information of the topology of the network which is built with local information using distance measurements between pairs of sensors. Nonparametric belief propagation (NBP) is used and shown to perform better than belief propagation alone. Considering location errors and the estimation of sensor locations, NBP is a generalization of particle filtering. Authors use graphical models where inference algorithms are applied, where belief propagation and NBP are two of those algorithms. With the definition of the graphical model for localization, the inference algorithms can be iterative processes where sensors generate a belief to be shared with its neighbors, and then with information from the neighbors, the belief is modified and shared once again, repeating until convergence.
Signature Generation Algorithms for Polymorphic Worms
Published in Mohssen Mohammed, Al-Sakib Khan Pathan, Automatic Defense Against Zero-day Polymorphic Worms in Communication Networks, 2016
Mohssen Mohammed, Al-Sakib Khan Pathan
Belief propagation (BP) is a message-passing algorithm proposed by Judea Pearl in 1982 for performing inference on graphical models, such as BNs and Markov random fields. It is inherently a Bayesian procedure, which calculates the marginal distribution for each unobserved node, conditioned on the observed nodes. BP was supposed to work only for tree-like graphs, but it has demonstrated empirical success in networks with cycles such as error-correcting codes. It has also been shown to converge to a stationary point of an approximate free energy, known as the Bethe free energy in statistical physics. The BP algorithm follows from the rules of probability and the conditional independence properties of the graph. Whereas variable elimination focuses on finding the conditional probability of a single variable Xi given Xobs = xobs, BP can compute at once all the conditionals p(Xi | Xobs = xobs) for all i not observed.
A Tutorial on Bayesian Networks for Systems Health Management
Published in Ashok N. Srivastava, Jiawei Han, Machine Learning and Knowledge Discovery for Engineering Systems Health Management, 2016
Arthur Choi, Lu Zheng, Adnan Darwiche, Ole J. Mengshoel
“Loopy” belief propagation and other message-passing algorithms inspired by it have in the past decade been particularly successful and have enabled new and influential applications in fields as diverse as information theory [25], computer vision, satisfiability, and more broadly in artificial intelligence [1,26]. More recent generalizations of belief propagation also provide a means by which computational accuracy can be traded for computational resources. For example, the perspective given by [27] formulates belief propagation as a way to compensate for structural relaxations, which yields a spectrum of approximation. On one end, with a coarse, fully disconnected approximation, we have loopy belief propagation. On the other end, with the original unrelaxed model, we have exact inference. Given constraints on the computational resources available, we can then to try to identify an approximation along this spectrum that is as accurate as possible.
Plant stress propagation detection and monitoring with disruption propagation network modelling and Bayesian network inference
Published in International Journal of Production Research, 2022
Win P. V. Nguyen, Puwadol Oak Dusadeerungsikul, Shimon Y. Nof
The scanning protocol initialises the BayesNet, and updates the set of observations after each scanning decision allocation. Each observed node with would have its corresponding status updated in the BayesNet. Then, the probability of each unobserved node being stressed (with ) can be computed. Exact Bayesian network inference requires exponential time, but the belief propagation algorithm, which requires polynomial time, can also be used (Pearl 1982). The scanning protocol then proceeds to select an unobserved node with the highest probability of being stressed, given all observations already made. Formally, The information is already available to the system because for all , and is provided by the BayesNet.
Terrain-aware traverse planning for a Mars sample return rover
Published in Advanced Robotics, 2021
A few authors have proposed to utilize Markov Random Field (MRF) to find the maximum-likelihood field of principal directions, with features that the robot can detect in its surroundings for driving into confusing environment [26]. In addition, MRFs (also called Undirected Graphical Models or UGMs) have been considered for terrain classification in support of path planning [27]. They can also be applied to belief propagation using inference between nodes to propagate a belief and locally update a map [11], illustrated by widely used methods such as the sum-product message passing algorithm [28]. Map updates are beneficial to planning as it reduces the environmental uncertainty over time. They have been proposed using other methods as well, for example in research where the rover gathers information about its environment online to update its knowledge and plan its traverse accordingly. This has been demonstrated to work for local planning in an unknown, off-road environment [29], and for planning in hazardous environments containing radioactive materials [30]. However, the mapping and planning phases are both performed online, which might not be suitable for a planetary rover with limited on-board resources and time to perform computations.
Intelligent Collectives: Theory, Applications, and Research Challenges
Published in Cybernetics and Systems, 2018
Van Du Nguyen, Ngoc Thanh Nguyen
However, it can be seen that the anonymous members who are often diverse in their backgrounds or knowledge bases. Thus, in some sense, their solutions are unreliable or biased. This phenomenon leads to the need of how to properly aggregate the results of the diverse members, and how to reduce the bias (to increase the quality of collective prediction) such as by providing a small amount of expert knowledge. For this aim, belief propagation - a message-passing algorithm (Yedidia, Freeman, and Weiss 2003) for performing inference probabilistic graphical models, which provide a powerful framework for aggregating multiple sources of information and reasoning over large numbers of variables, will be used (Yedidia, Freeman, and Weiss 2003). Moreover, choosing members for solving a given problem also plays an important role in collective intelligence applications. Beyond criteria proposed by Surowiecki, it is often required that selected members also have relevant knowledge of the problem that needs to be solved (Simmons et al. 2011). For such tasks, machine learning methods such as Learning-to-Rank, InfluenceRank can be used for forming a collective with predefined requirements. These research problems will be the subject of future works.