in 2010 I started to get interested in machine learning and artificial intelligence. This led me to start reading some books about neural networks which was the topic which fascinated me most. My first book was from the university library. I do not remember the name anymore, but it was a practical guide and was easy to read. After a few weeks reading was not enough anymore and I started to code my first neural network in VisualBasic. It could was a perceptron with one hidden layer. I successfully trained it on recognizing the letters A-Z, which where drawn on the input layer.
Doing OCR with Convolutional Neural Networks
Parallel to my internship at EADS where I worked on vehicle detection with neural networks my interest in convolutional neural networs as proposed by LeCun was rising. Together with a colleague we started developing an OCR app based on CNN for windows phones.
Unfortunatelly the development stopped after the internship ended for both of us and he went back to czech republic.
Here is a screenshot of the ongoing learning process of the cnn:
Attending the online course „Introduction to artificial intelligence“ from Prof. Thrun at the Standord University
Since I was still fascinated by machine learning and AI I took part in the online course of Sebastian Thrun. It was a one semester course about Bayes nets, Probability, reinforcement learning and basics of machine learning.
After these first steps I worked on following topics:
- Image classification with Random Forests
- Face detection with CNN
- create neural networks inspired by the Hierarchical Temporal Memory Architecture introduced by Jeff Hawkins and Numenta (Details here)
- Handwriting synthesis with recurrent neural networks
detailed desciptions will follow…