C-018
Fault-tolerant MLP Learnings using Deep Learning Framework

○ Astriwindusari・Tadayoshita Horita・Masakazu Akiba（Polytechnic Univ.）

One of the author proposed a learning algorithm, called “FTL-algo”, which make 3-layered perceptrons multiple-weight-and-neuron-fault tolerant developed its CUDA C code and gave the result using GPU. The FTL-algo is an ordinary back propagation algorithm with some modification. However, the FTL-algo using deep learning frameworks have not been found.
The purpose of this study is to evaluate in time and accuracy of learning when the FTL-algo calculation is done using deep learning frameworks to compare with previous result.
Index Terms – Fault tolerance, GPGPU, Deep Learning Framework