|
|
|||||
|
|
|||||
Multilayer Perceptron and Fuzzy Systems C code Generator (1992)
[Deep Learning on Embedded Devices]
NeurFuzz™ 2.0 was released in 1992 as a Multilayer Perceptron type
neural network training tool with generation of trained C source code usable
for embedded solutions. The application was developed to run in a DOS
environment and allowed you to train Multilayer Perceptron neural networks with
500 inputs, 500 outputs and up to 20 hidden layers. Therefore this application
allowed the creation of Deep Learning solutions in a DOS environment for
embedded applications.
Neurfuzz™ was released in a stripped down version 1.0 as freeware,
with the number of hidden layers limited to 2 , only 100 input/output and no
DSP coprocessors support.
NeurFuzz™ 2.0 allowed you to train a Multilayer Perceptron neural
network by generating the source code in C language with all the trained
synaptic weights included in a header file.
Among the advanced features of the tool were:
1) Genetic Algorithm Training that could start automatically when
the Error Back Propagation-based gradient descent algorithm entered a local
minimum.
2) Simulated Annealing technology that prevented entry into local minima
during gradient descent.
3) Automatic correction of Epsilon and Momentum when learning with
Error Back Propagation.
4) Generation of Input and Output interfaces based on Fuzzy Logic
with trigonometric membership functions (or conventional triangular/trapezoidal
functions).
5) Generation of trained C code for applications on any HW
platform and any operating system.
6) Support for the most popular ISA PC cards based on DSP (Digital
Signal Processing) for acceleration of the learning process with Error Back
Propagation.
|
|
|
©2024_Luca_Marchese_All_Rights_Reserved Aerospace_&_Defence_Machine_Learning_Company VAT:_IT0267070992 NATO_CAGE_CODE:_AK845 Email:_luca.marchese@synaptics.org |
|
Contacts_and_Social_Media
|
