A review on evolutionary algorithms in Bayesian network learning and inference tasks

Larrañaga, Pedro; Karshenas, Hossein; Bielza, Concha; Santana, Roberto

doi:10.1016/j.ins.2012.12.051

Cited by 141 publications

(81 citation statements)

References 65 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…[42], [43] and [44]), the focus of such works has been to create the network and the links therein. For instance, [45] uses a multi-objective genetic algorithm (MOGA) to evolve dynamic Bayesian networks.…”

Section: F Previous Work and Design Decisionsmentioning

confidence: 99%

Explaining quantum correlations through evolution of causal models

et al. 2017

View full text Add to dashboard Cite

We propose a framework for the systematic and quantitative generalization of Bell's theorem using causal networks. We first consider the multi-objective optimization problem of matching observed data while minimizing the causal effect of nonlocal variables and prove an inequality for the optimal region that both strengthens and generalizes Bell's theorem. To solve the optimization problem (rather than simply bound it), we develop a novel genetic algorithm treating as individuals causal networks. By applying our algorithm to a photonic Bell experiment, we demonstrate the trade-off between the quantitative relaxation of one or more local causality assumptions and the ability of data to match quantum correlations.While it seems conceptually obvious that causality lies at the heart of physics, its exact nature has been the subject of constant debate. The fundamental implications of quantum theory shed new light on this debate. It is thought these implications may lead to new insights into the foundations of quantum theory, and possibly even quantum theories of gravity [1][2][3][4][5][6][7][8][9][10].These realizations have their roots in the EinsteinPodolski-Rosen thought experiment [11] and the fundamental theorems of Bell [12] and of Kochen and Specker [13]. A cornerstone of modern physics, Bell's theorem, rigorously excludes classical concepts of causality. Roughly speaking Bell's theorem states that the following concepts are mutually inconsistent: (1) reality; (2) locality; (3) measurement independence; and (4) quantum mechanics.In philosophical discussions, typically one rejects (1) or (2), which together are often referred to as local causality, though the other options have been considered as well. In studies with an operational bent, however, one often considers relaxations of (2) or (3) which is what we concern ourselves with here. These relaxations have been addressed from different perspectives, but only regarding specific causal influences in isolation [14][15][16][17][18][19][20][21][22][23], whereas here we wish to study all possible relaxations of the causal assumptions implied by (2) and (3) simultaneously.The framework of causal networks [24,25] is wildly successful within the field of machine learning and has led some physicists to utilize them to elucidate the tension between causality and Bell's theorem. Recently, Wood and Spekkens have shown that existing principles behind causal discovery algorithms (namely, the absence of fine tuning) still cannot be reconciled with entanglement induced quantum correlations even if one admits nonlocal models [9]. However, such results only hold for the exact distributions, and would not necessarily apply to exper- * These authors contributed equally to this work.imental data due to measurement noise, or a relaxation of the demand of reproducing exactly the quantum correlations. Clearly, the further away from the quantum correlations one is allowed to stray, the more likely a locally causal model can be found.Here we propose a framework for systematic and qua...

show abstract

Section: F Previous Work and Design Decisionsmentioning

confidence: 99%

Explaining quantum correlations through evolution of causal models

et al. 2017

View full text Add to dashboard Cite

show abstract

“…There are four general categories of algorithms for learning Bayesian networks: search-and-score, constraint-based, hybrid [19] and evolutionary algorithms [20]. Search-and-score methods such as K2 [9] and Greedy Equivalence Search (GES) [8] rely on heuristics to sequentially add, remove, or change the direction of the edges in the graph, G, to which a scoring method is applied.…”

Section: Learning and Scoring Bayesian Networkmentioning

confidence: 99%

Widened Learning of Bayesian Network Classifiers

Sampson

Berthold

2016

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. We demonstrate the application of Widening to learning performant Bayesian Networks for use as classifiers. Widening is a framework for utilizing parallel resources and diversity to find models in a hypothesis space that are potentially better than those of a standard greedy algorithm. This work demonstrates that widened learning of Bayesian Networks, using the Frobenius Norm of the networks' graph Laplacian matrices as a distance measure, can create Bayesian networks that are better classifiers than those generated by popular Bayesian Network algorithms.

show abstract

“…Metode ini bertujuan untuk memodelkan variabel yang terkait dengan kategori tingkat IPM di provinsi Jawa Barat. Hasil pemodelan berupa struktur graf yang mudah untuk dipahami, terutama dalam melihat adanya keterkaitan antar variabel [10]. Variabel IPM yang berupa variabel kontinu akan dikategorikan pada dua skenario, yaitu eksperimen 1 dengan tiga kategori dan eksperimen 2 dengan empat kategori.…”

Section: Pendahuluanunclassified

Identifikasi Keterkaitan Variabel dan Prediksi Indeks Pembangunan Manusia (IPM) Provinsi Jawa Barat Menggunakan Dynamic Bayesian Networks

Sari

Priati

2016

INFOTEL

View full text Add to dashboard Cite

Abstrak -Indeks Pembangunan Manusia (IPM) adalah indeks yang menggambarkan kualitas pembangunan manusia di suatu daerah secara menyeluruh, yaitu sebagai konsekuensi logis dari aktivitas pembangunan di daerah tersebut. Membangun sebuah model prediksi tingkat IPM sangat dibutuhkan untuk mengetahui faktorfaktor apa saja yang berpengaruh dominan dan dijadikan sasaran fokus pengembangan dan perbaikan di daerah untuk dapat meningkatkan IPM. Penelitian ini menerapkan Dynamic Bayesian Networks (DBN) untuk memprediksi dan memodelkan lima belas variabel yang mempengaruhi IPM. Struktur Dynamic Bayesian Networks dibangun dengan software CaMML lalu dievaluasi dengan mengukur tingkat akurasi. Skenario eksperimen pada penelitian ini dibagi menjadi 4 skenario, yaitu perbedaan proposi data latih, data uji serta kategori tingkat IPM. Eksperimen menggunakan 75% data latih dan 25% data uji pada prediksi tiga kategori tingkat IPM menghasilkan tingkat akurasi yang terbaik, yaitu sebesar 88,461%. Struktur graf Dynamic Bayesian Networks bisa menunjukkan adanya hubungan antar variabel. Struktur graf Dynamic Bayesian Networks yang terbangun bisa dijadikan model prediksi di mana dapat memberikan rekomendasi faktor utama apa yang perlu diperhatikan untuk meningkatkan kategori tingkat IPM di kabupaten/kota yang ada di provinsi Jawa Barat, yaitu faktor pendidikan, faktor kependudukan, dan faktor kesehatan. Kata kunci -IPM, keterkaitan variabel, prediksi, dynamic bayesian networksAbstract -Human Development Index (HDI) is an index describing the quality of human development in a region as a whole, which is a logical consequence of the development activities in the area. Building a predictive model of HDI level is needed to determine what factors are dominant and the subject of focused development and improvement in the area in order to improve the HDI. This study applies Dynamic Bayesian Networks (DBN) to predict and model the fifteen variables that affect the HDI level. Structural Dynamic Bayesian Networks built with software CaMML then evaluated by measuring the level of accuracy. Scenario experiments in this study were divided into four scenarios, the differences in the proportion of training data, test data and levels of HDI category. Experiments using 75% and 25% training data test data to the prediction of three-level category HDI produces the best accuracy rate, which amounted to 88.461%. Dynamic Bayesian Networks graph structure may indicate a relationship between variables. Dynamic Bayesian Networks graph structure that is built can be used as a predictive model which could provide recommendations what key factors need to be considered to improve the level of HDI category in the districts/cities in West Java province, namely educational factors, demographic factors, and health factors.

show abstract

A review on evolutionary algorithms in Bayesian network learning and inference tasks

Cited by 141 publications

References 65 publications

Explaining quantum correlations through evolution of causal models

Explaining quantum correlations through evolution of causal models

Widened Learning of Bayesian Network Classifiers

Identifikasi Keterkaitan Variabel dan Prediksi Indeks Pembangunan Manusia (IPM) Provinsi Jawa Barat Menggunakan Dynamic Bayesian Networks

Contact Info

Product

Resources

About