# Optimization of Convolutional Neural NetworkPrecision with statistical approach
_by Milot Quentin - 2021.09.28_

###### tags: `VAADER` `Seminar`


## Video
{%youtube na6nAk5NzRk %}

## Abstract

Convolutional neural networks are artificial intelligence techniques used to detect an object in an image using a given image bank or used for image classification. It's a really powerful tool by its precision but the price of this precision is an expanding complexity. Reducing the number of bits used to store data is a way of solving that problem. Nevertheless, this method requires a test to assure that the information lost by the reduction of bits is satisfying. The idea explored in this paper is to reduce the number of inputs for this test and by extension to limit the complexity of bits number test. For that, a possibility is to add a distortion of the input of the tested CNN in a way that each images are attributed a weight according to their difficulty to be analyzed. This method will allow to check the number of bits of the neural network by working with more harder images but less images in total.