Although neural networks have been around for quite a while now, deep learning really just took of a few years ago. It pretty much all started when Alex Krizhevsky and Geoffrey Hinton utterly crushed classic image recognition in the 2012 ImageNet Large Scale Visual Recognition Challenge by implementing a deep neural network with CUDA on graphics cards. A lot has changed since that time: The toolchain to do deep learning has rapidly evolved into API’s with a very high level of abstraction. Nowadays everyone can train complex neural networks with billions of free parameters. Just last year RStudio announced the Keras for R package. Keras is a high level neural network API that makes it really easy to define the architecture of a neural network. In this talk we will rush through an explanation of convolutional neural networks for image recognition, learn how easy it has become to do production ready deep learning with the use of docker and why R’s syntax is even better suited to define a neural network than python’s. (Hint: Probably this is not a pipe 😉
Kai Lichtenberg is a PhD student in the Bosch PhD Program working on models to predict the reliability of components in drive trains by leveraging the ever more available high dimensional data in the era of the Internet of Things. Coming from the field of mechanical engineering and technical reliability (which has a lot to do with stochastic processes) he found his destination in data science. As a long time computer enthusiast he is always keen to use the newest technologies. To bear the pain of getting a toolchain up and running is probably his super power.