Ways of Seeing

The biggest difficulty that I last encountered during reading is the ability to philosophically see it. Therefore, I treat myself myself not myself but a headache. I have read lots of literature, but art is not that good at it. It is not a concept I have to agree, but the idea that images are more accurate and richer than literature may be correct. The beauty of unspecified literature is the ability to design your own image. Therefore, although the image is more accurate, I think the text based fictitious image is more abundant.

The core claim of my paper is informally, "You can learn the form of observation by examining the best set of objects in this way of observation." From RBM to VAE and interactive GAN, it is known not only in the broadest sense as "automatic encoder" but also it can learn not only from the low dimensional representation space but from every point in the input space to the input space . Points in low dimension manifolds In many cases, practitioners of machine learning in applications are most interested in this projection function during the training phase used as a substitute for low dimensional representation space. They use the distance between input and projection (on training data) as the optimization target. And when the training is over, they extract the low dimensional representation space achieved by the training.

In many ways this is similar to the way the two dimensions sense the existence of three dimensions in two dimensional space. Only a small part of 3D can be seen, and what you see is like a two dimensional object. If you feel better, immediately buy ($ 0.99), read Edwin Abbott's very short and wonderful classic "Flatland". The dimension layer used in Flatland proposes a way to convert this perceptual limit into one that can be quantified. It has shown that with the development of human beings, we have made many progressions from single cell organisms we are seeing today to global species, and these advances can not be recognized in our previous versions.