Reconstructing Japanese handwritten images using auto-encoder with residual block in parallel computing
M Octaviano Pratama, Pamela Kareen · 2017
Unsupervised learning is essential for reconstructing a model without target label such as Reconstructing handwritten character that usually used as benchmarking in deep learning tasks. Auto-encoder is one of unsupervised algorithm with purpose of dimensionality reduction and data reconstruction. Auto-encoder with deep architecture layer will produce noise higher than shallow architecture. In this research, existing Auto-Encoder algorithm augmented with Residual Block is performed in parallel computing to reconstruct Japanese handwritten images. Residual block can strengthen deep layer connection. The result is compared with classical AutoEncoder.