SI (Seperation Index) and SmI (Smoothness Index) are some metrics proposed by Prof. Ahmad Kalhor at university of Tehran.
These metrics can be used in a wide range of applications such as "compressing neural networks" and "extrapolating relations in data without training a model" .
This repository contains implementation and applications of some variants of these metrics.
For more technical details and results read the following paper by Prof. Kalhor and A. Karimi :
Learning Enhancement of CNNs via Separation Index Maximizing at the First Convolutional Layer