Function estimation by feedforward sigmoidal networks with bounded weights
- Oak Ridge National Lab., TN (United States). Center for Engineering Systems Advanced Research
- Fort Valley State Coll., GA (United States). Dept. of Mathematics and Physics
The authors address the problem of PAC (probably and approximately correct) learning functions f : [0, 1]{sup d} {r_arrow} [{minus}K, K] based on iid (independently and identically distributed) sample generated according to an unknown distribution, by using feedforward sigmoidal networks. They use two basic properties of the neural networks with bounded weights, namely: (a) they form a Euclidean class, and (b) for hidden units of the form tanh ({gamma}z) they are Lipschitz functions. Either property yields sample sizes for PAC function learning under any Lipschitz cost function. The sample size based on the first property is tighter compared to the known bounds based on VC-dimension. The second estimate yields a sample size that can be conveniently adjusted by a single parameter, {gamma}, related to the hidden nodes.
- Research Organization:
- Oak Ridge National Lab., TN (United States)
- Sponsoring Organization:
- USDOE, Washington, DC (United States)
- DOE Contract Number:
- AC05-96OR22464
- OSTI ID:
- 238556
- Report Number(s):
- CONF-9606113--2; ON: DE96008788
- Country of Publication:
- United States
- Language:
- English
Similar Records
On PAC learning of functions with smoothness properties using feedforward sigmoidal networks
On PAC learning of functions with smoothness properties using feedforward sigmoidal networks
Nearest neighbor rules PAC-approximate feedforward networks
Technical Report
·
Sun Mar 31 23:00:00 EST 1996
·
OSTI ID:217721
On PAC learning of functions with smoothness properties using feedforward sigmoidal networks
Journal Article
·
Tue Oct 01 00:00:00 EDT 1996
· Proceedings of the IEEE
·
OSTI ID:417981
Nearest neighbor rules PAC-approximate feedforward networks
Conference
·
Wed May 01 00:00:00 EDT 1996
·
OSTI ID:228504