Function approximation using adaptive and overlapping intervals
A problem common to many disciplines is to approximate a function given only the values of the function at various points in input variable space. A method is proposed for approximating a function of several to one variable. The model takes the form of weighted averaging of overlapping basis functions defined over intervals. The number of such basis functions and their parameters (widths and centers) are automatically determined using given training data and a learning algorithm. The proposed algorithm can be seen as placing a nonuniform multidimensional grid in the input domain with overlapping cells. The non-uniformity and overlap of the cells is achieved by a learning algorithm to optimize a given objective function. This approach is motivated by the fuzzy modeling approach and a learning algorithms used for clustering and classification in pattern recognition. The basics of why and how the approach works are given. Few examples of nonlinear regression and classification are modeled. The relationship between the proposed technique, radial basis neural networks, kernel regression, probabilistic neural networks, and fuzzy modeling is explained. Finally advantages and disadvantages are discussed.
- Research Organization:
- Los Alamos National Laboratory (LANL), Los Alamos, NM (United States)
- Sponsoring Organization:
- USDOE, Washington, DC (United States)
- DOE Contract Number:
- W-7405-ENG-36
- OSTI ID:
- 64149
- Report Number(s):
- LA-UR-95-1224; CONF-9503135-2; ON: DE95010865; TRN: 95:004362
- Resource Relation:
- Conference: APIC `95, El Paso, TX (United States), Mar 1995; Other Information: PBD: [1995]
- Country of Publication:
- United States
- Language:
- English
Similar Records
Neural-network algorithms and architectures for pattern classification
Convolutional Neural Networks for Hydrometeor Classification using Dual Polarization Doppler Radars