Samuel Rochette @Saxamos
Discover how to train and deploy a convolutional neural net on a raspberry pi. We will talk about application development, handling and labelling the data to enhance the model, some viz to understand the model and a special method based on entropy to get automatic threshold for black and white pictures.
In this talk you will learn a type of attack for convolutional neural net. This consists in modifying slightly your input data in order to fool the inference of your neural net. this this a hot topic in IA security.
Many models give a lot more information during the inference process that we usually know. We'll begin with an intrinsic estimation of all the distribution with random forest algorithm. Then we will extend those "prediction intervals" to almost every regression models thanks to the quantile loss. Eventually we'll discuss about probability calibration to measure uncertainty in classification.