Machine Learning

Image Recognition:
Above MNIST data is simple data sets from 0 to 9. An objective is to estimate hand-written number with high accuracy. Finally, estimation data could be obtained with a resulting error less than 1%. It has been applied to face detection and medical diagnosis, for example. A part of deep learning algorithm is described as,

Outputs after convolution layer:
Above 16 images are the outputs after determining the weightings of convolutional layer. The vertical edge is exaggerated in one of them while horizontal edge is stressed in the other image. This characterization feature contributes to improve the accuracy of neural network.