HSD CNN using class specific filter sensitivity analysis - reusable CNN without retraining

What is it about?

With semi supervised training as background, HSD CNN technique can convert any sequential CNNs to hierarchical structure with reusable model for various applications without retrain and finetune of the model. It can also be used for coarse to fine category feature extraction of classes. Filter sensitivity analysis applied behind is very useful in finding out the significant and insignificant filters for feature analysis.

Featured Image

Photo by Wilhelm Gunkel on Unsplash

Photo by Wilhelm Gunkel on Unsplash

Why is it important?

HSD CNN architecture make it easier to reuse the model with retraining. The Subnetworks of HSDCNN used for deployment has higher accuracy and faster inference time and make it easier for federated learning.