| Author | Htet Htet Wunn |
| Call Number | AIT Thesis no.CS-95-08 |
| Subject(s) | Neural networks (Computer science)
|
| Note | A thesis submitted in partial fulfillment of the requirement for the degree of Master of
Engineering |
| Publisher | Asian Institute of Technology |
| Abstract | Generation of elementary functions is not a trivial task and in hardware implementations
of neural networks it may be critical that they be generated with speed and with given precision.
The transfer function and its derivative, which are elementary functions or compositions of
them, for example, are needed extensively in backpropagation learning algorithm. We compare
two methods, piece-wise linear approximation (PLA) method and add table-lookup add (ATA)
method, and present two more polynomial interpolation methods called pl and p2, to compute
elementary functions by hardware. Although ATA yields more precision than other methods
presented, because of simple circuitry, PLA, pl and p2 methods are more suitable for neural
network hardware. Further p2 method among these three is the most precise: it offers an
average precision of 10-4 while PLA and pl's average precision is 10-3. An elementary
statistical analysis has been provided and design of components by using SIS is introduced. |
| Year | 1995 |
| Type | Thesis |
| School | School of Engineering and Technology (SET) |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Computer Science (CS) |
| Chairperson(s) | Yulu, Qi; |
| Examination Committee(s) | Sadananda, Ramakoti;Batanov, Dentcho N.; |
| Scholarship Donor(s) | Government of Finland; |
| Degree | Thesis (M.Eng.) - Asian Institute of Technology, 1995 |