Session: AI: Images Classification for beginners

One of the basic human abilities is to analyze their environment. This involves in most cases recognizing the elements of our field of vision: finding others, identifying cars, animals, etc. This task was difficult for a computer until the emergence of convolutional neural networks in 2012. Luckily, the approach of these networks inspired by our visual cortex has opened many applications, whether in medical imaging, or autonomous cars…ect.

CNNs receive input images, detect the features of each of them, then train a classifier on them. However, features are learned automatically. The CNNs themselves do all the tedious work of extracting and describing features: during the training phase, the classification error is minimized in order to optimize the parameters of the classifier and the features. In addition, the specific architecture of the network makes it possible to extract features of different complexities, from the simplest to the most sophisticated. One of the strengths of CNNs is the automatic extraction and hierarchy of features, which adapt to the given problem: no need to implement an extraction algorithm “by hand”, like SIFT (Scale-invariant Feature Transform) or Harris-Stephens.

First of all I will give a brief introduction on Deep Learning and Artificial Neural networks then i will explain how a convolutional neural network works. In particular, I will present the different elements of a CNN architecture (convolutions, pooling, ReLU, flattening, dense ...) and real networks used in production.

Key Takeaways

• Artificial Intelligence /Deep learning

• Artificial neural networks /Convolutional Neural Networks

• Transfer Learning

This course is for extreme beginners but anyone is welcomed to take it.

If you want more of this content; leave comments.