Z1h. AI with neural networks: recognition of handwritten digits

Door: Gerhard Röhner, MNU

WERKVORM: workshop

DOELGROEP: iedereen

VAKKEN: alle

The topic of artificial intelligence has become a mega-topic in just a few years. AI can drive autonomously, play Go and Chess, Siri and Alexa recognize language, robots report on sporting events and post on Facebook, and ChatGPT writes school essays. But how intelligent is AI really?

Today’s AI systems are based on neural networks. Due to high computing power and large data volumes, neural networks can be trained very successfully. The large language models based on them are very useful in the private, public, commercial, and scientific sectors.

In this workshop, we will explore how neural networks work. Starting with a biological neuron, we will abstract its functionality to model a computational neuron. We will then create a neural network with neurons in the input, hidden, and output layers, each interconnected. Using this data structure, we will model forward propagation for the application of the neural network and backward propagation as a central learning method in supervised learning, which requires some mathematical derivations. The implementation of a neural network in a Java program is based on the classic example of recognizing handwritten digits. You can train and test the network and use it for recognizing handwritten digits.