Efficient Neural Network Compression via Transfer Learning for Industrial Optical Inspection

In this paper, we investigate learning the deep neural networks for automated optical 1 inspection in industrial manufacturing. Our preliminary result has shown the stun2 ning performance improvement by transfer learning from the completely dissimilar 3 source domain: ImageNet. Further study for demystifying this improvement shows 4 that the transfer learning produces a highly compressible network, which was not 5 the case for the network learned from scratch. The experimental result shows that 6 there is a negligible accuracy drop in the network learned by transfer learning until 7 it is compressed to 1/128 reduction of the number of convolution filters. This result 8 is contrary to the compression without transfer learning which loses more than 5% 9 accuracy at the same compression rate. 10