Thank you for your thoughts on GAN & Active learning.
Appreciate your thoughts!
1. Can GAN solve the ever growing need of realistic synthetic data for training deep neural networks?
2. Cursive handwriting is still an unsolved problem with variation in different styles. Do you feel active learning can be helpful in such problems?
Right there wasn't much loss of data, basically I have to train a model for binary images produced after scanning. Since, NIST is a rich database for hand written characters, transferring knowledge from this source can be one of the possible options, but it contains characters in grayscale (originally binary, converted to gray scale by anti aliasing resampling ). I am interested to know the motivation behind using anti aliasing by Yann le cunn to resample the binary data using anti aliasing to form mnist data, this could answer many questions.