Summary: A picture is worth a thousand words. But what does it say?
An unprecedented amount of data is being generated on a daily basis. Automatic processing and analyses of these data sets therefore offer numerous benefits to decision makers in governmental and commercial arena. Due to the diverse nature of the information sources, data can come in many forms including time series data, text data or visual media such as pictures. It is therefore vital to consider these distinctive characteristics when building machine learning algorithms to extract meaningful information from an given dataset. For photographic data, one of the most popular algorithms is without a doubt convolutional neural networks. Owing to the simplicity of the implementation via programming packages, these algorithms, however, are often treated as a black box. In my talk, I will focus on demystifying convolutional networks by explaining their components and the training process in an intuitive way as well as covering examples from transfer learning to using convolutional nets as feature extractors. I will finish my presentation with a discussion on potential applications in insurance industry.