For example, at the beginning there are 10 classes images, and each image’s feature vector is 100 dimension vector, so the classifier weight matrix is 100*10.
Then I get 5 new classes images, so the weight matrix would be 100*15.
So the question is how can I inherit and fix the previous 100*10 matrix, which means that we just only fine-tune the 100*5 matrix.
Can this be implemented easily? Thanks so much!