pengzhenghao / axnet

AXNet: Approximate computing using an end-to-end trainable neural network

• 👋 Hi, I’m Zhenghao Peng.
• 👀 Please visit my homepage: https://pengzhenghao.github.io
• 📫 Feel free to contact me via: pzh AT cs.ucla.edu

axnet's Issues

accuracy metric

Hello! @pengzhenghao
I have cloned the AXNet and followed the paper "AXNet: approximate computing using an end-to-end trainable neural network". nice work!
but I have some question:
for the "inversek2j" application, the parameters used in the AXNet program https://github.com/PengZhenghao/AXNet/blob/master/example.ipynb is bench_mark = 'inversek2j' total_iteration = 50000 total_LR = 1e-4 error_bound = 0.1, but the error bound in your paper is error _bound=0.01 with corresponding prediction accuracy near 80%.
when I follow the idea, and set the error _bound=0.01, the accuracy is about 67%. I wanna find the reason
that I going wrong. Thanks!