Abstract
Approximation capabilities are important and primary properties of neural networks and fuzzy neural networks (FNNs). Neural networks have been successfully applied in many fields since they can work as approximators in nature. Many scholars research FNNs’ approximation abilities for continuous fuzzy functions. It is concluded that FNNs can work as approximators for continuous fuzzy functions if the fuzzy functions satisfy some specified conditions. However, the problem whether FNNs can work as approximators for discontinuous fuzzy functions is not solved completely until now. In this work, we focus on the approximation of polygonal FNN for discontinuous fuzzy functions in sense of Choquet integral norms. We first introduce the Choquet integral norms in sub-additive fuzzy measure. Then the universal approximation of polygonal FNNs for fuzzy valued functions in sense of Choquet integral norms is analyzed in this paper. It is proved that the polygonal FNNs can work as approximators for fuzzy valued functions in the sense of Choquet integral norms with respect to sub-additive fuzzy measure.
We’re sorry, something doesn't seem to be working properly.
Please try refreshing the page. If that doesn't work, please contact support so we can address the problem.
References
Zheng H, Wang H (2012) Improving pattern discovery and visualisation with self-adaptive neural networks through data transformations. Int J Mach Learn Cybern 3(3):173–182
Sarlin P (2012) Visual tracking of the millennium development goals with a fuzzified self-organizing neural network. Int J Mach Learn Cybern 3(3):233–245
Buckley JJ, Hayashi Y (1994) Can fuzzy neural nets approximate continuous fuzzy functions? Fuzzy Sets Syst 61:43–52
Liu PY, Li HX (2005) Approximation analysis of feed forward regular fuzzy neural networks with two hidden layers. Fuzzy Sets Syst 150:373–396
Liu PY (2002) Fuzzy neural networks theory and application research, Department of Mathematics, Beijing Normal University mathematics research Institute
Huang H, Wu CX (2009) Approximation capabilities of multilayer fuzzy neural networks on the set of fuzzy-valued functions. Inf Sci 179(16):2762–2773
Huang H, Wu CX (2009) Approximation of level continuous fuzzy-valued functions by multilayer regular fuzzy neural networks. Math Comput Model 49:1311–1318
Huang H, Wu CX (2011) Approximation of fuzzy functions by regular fuzzy neural networks. Fuzzy Sets Syst 177(1):60–79
Zhao FX, Li HX (2006) Approximation of regular fuzzy neural networks to seguno-integrable functions. Acta Mathematicae Appli- Catae Sinica 19(1):39–45
Zhao FX, Li HX (2004) Approximation of regular fuzzy neural networks to L-integrable functions. Prog Nat Sci 14(9):1025–1031
Li J, Gao SH, Chen B (2010) Approximation of fuzzy neural networks to fuzzy-valued measurable function. Adv Intell Soft Comput 82:351–358
Lovassy R, László TK, Gál L (2010) Function approximation performance of fuzzy neural networks. Acta Polytechnica Hungarica 7(4):25–38
Gál L, Lovassy R, Kóczy L T, Function approximation performance of fuzzy neural networks based on frequently used fuzzy operations and a pair of new trigonometric norms. IEEE International Conference on Fuzzy Systems, 2010, pp.1–8
Liu PX (2002) A new fuzzy neural network and its approximation capabilities. Sci China (Series E) 32(1):76–86
He CM, Ye YP, Xu WH (2009) Universal approximation of fuzzy functions by polygonal fuzzy neural networks with general inputs. Pattern Recogn Artif Intell 22(3):481–487
Wang GJ, Li XP (2011) Universal approximation of polygonal fuzzy neural networks in sense of K-integral norms. Sci China Inf Sci 54(11):2307–2323
Choquet G (1955) Theory of capacities. Ann Inst Fouries 5:131–295
Murofushi T, Sugeno M (1989) An interpretation of fuzzy measures and the Choquet integral as an integral with respect to a fuzzy measure. Fuzzy Set Syst 129:201–227
Song J, Li J (2003) Regularity of null-additive fuzzy measure on metric space. Int J Gen Syst 33(3):271–279
Puyin L, Hongxi L (2007) Symmetric polygonal fuzzy number space. J Fuzzy Math 15(1):27–42
Báez-Sánchez AD, Morettib AC, Rojas- Medar MA (2012) On polygonal fuzzy sets and numbers. Fuzzy Sets Syst 209:54–65
Acknowledgments
This work was supported by the National Natural Science Foundation of China (Grant No. 61163040 and 61261005),the Science Foundation of Jiangxi Provincial Department of Education (Grant No. GJJ12323) and the Foundation of East China Jiaotong University (Grant No. 09102018).
The authors would like to thank anonymous referees for their valuable comments to improve this paper.
Author information
Authors and Affiliations
Corresponding author
Appendix proof of theorems
Appendix proof of theorems
1.1 Proof of theorem 4.1
Because \( \left( {F_{0}^{c} (R),D} \right) \)is a complete disjunctive metric space, for arbitrary \( \tilde{X} \in F_{0}^{c} (R),i \in N \), there exist an enumerable dense set \( \left\{ {\tilde{A}_{i} ,i \in N} \right\} \) such that \( D(\tilde{X},\tilde{A}_{i} ) < \varepsilon_{0} \). Let
where \( \mathop \cup \limits_{i \in N} T_{i} = T,T_{i} \cap T_{j} \ne \Upphi \) (i ≠ j) and \( \mu (T) = \mu (\mathop \cup \limits_{i \in N} T_{i} ) < + \infty \). Then there exist \( d_{0} \in N \) such that \( \mu (\mathop \cup \limits_{{i > d_{0} }} T_{i} ) < \varepsilon_{0} \). Let
where \( T_{0} = \mathop \cup \limits_{{i > d_{0} }} T_{i} ,a_{0} = 0,0 < a_{1} \le a_{2} \le \cdots \le a_{n} .\{ T_{0} ,T_{1} , \cdots ,T_{{d_{0} }} \} \), is a finite dissection of T, S is the fuzzy-valued simple function on T, and
Suppose \( \varepsilon \to 0,{\text{then}}\;\Updelta \left( {F,S} \right) \to 0, \) that is, for arbitrary \( \varepsilon > 0,\;\Updelta \left( {F,S} \right) < \varepsilon \) hold. The proof is completed.
1.2 Proof of theorem 4.2
(1) Let
Where \( a_{0} = 0,0 < a_{1} \le a_{2} \le \cdots \le a_{n} . \) For \( \chi_{{A_{i} (x)}} \) is a non- negative measurable function, from lemma 4.1, we can conclude that, for arbitrary \( \varepsilon_{0} > 0, \) there exist one closed set \( L_{k} \subseteq T \) such that
and \( \chi_{{A_{i} (x)}} \) is continue on L k . Therefore there exist a Tauber–Wener function \( \sigma ,\,t_{k} \subseteq N,\,v_{k1}^{'} ,v_{k2}^{'} , \cdots ,v_{{kt_{k} }}^{'} ,\,\theta_{k1}^{'} ,\theta_{k2}^{'} , \cdots ,\theta_{{kt_{k} }}^{'} \in R, \) and \( u_{k}^{'} (1),u_{k}^{'} (2), \cdots ,\,u_{k}^{'} (t_{k} ) \in R^{n} \) such that
If let \( L = \mathop \cap \limits_{k = 1}^{q} L_{k} \), then \( T = L \cup (T - L) \), where
Because \( \mu \left( {T - L_{k} } \right) < \varepsilon_{0} \), there exist ɛ 1 > 0 such that \( \mu \left( {T - L} \right) < \varepsilon_{1} \). Suppose \( t = \sum\limits_{k = 1}^{q} {t_{k} } \), denote \( \beta_{k} = \sum\nolimits_{r = 1}^{k - 1} {t_{r} } ,\beta_{1} = 0 \) For \( k = 1,2, \cdots ,q,j = 1,2, \cdots ,t, \) define
Let
Then \( H \in H_{0} [\sigma ] \) is the fuzzy measure space and for \( \forall k = \left\{ {1,2, \cdots ,q} \right\}, \)
For \( \mu \left( {T - L} \right) < \varepsilon_{1} \), there exist ɛ 2 > 0 such that \( (C)\int_{T - L} {\left[ {D\left( {\sum\limits_{j = 1}^{t} {v_{kj} \cdot \sigma \left( { < u_{k} (j) + \theta_{kj} > } \right),\chi_{k} } } \right)} \right]d\mu < \varepsilon_{2} } \). Therefore
That is to say, there exist ɛ > 0 such that \( \Updelta \left( {H,S} \right) < \varepsilon \). The proof of (1) is completed.
(2) We can conclude (2) from theorem 4.1 and (1).
Rights and permissions
About this article
Cite this article
He, C. Approximation of polygonal fuzzy neural networks in sense of Choquet integral norms. Int. J. Mach. Learn. & Cyber. 5, 93–99 (2014). https://doi.org/10.1007/s13042-013-0154-8
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s13042-013-0154-8