Lecture Anchormen: Deep Neural Networks and Mechanical Turk

8 months ago
Thursday 13 dec 2018, 17:00 - 18:00


Deep Learning is one of the buzz words of past years in academia as well as business. How does the business integrate deep learning into successful projects? This will be the topic of today’s presentation. Anchormen is a company with many Data Scientists and Data Engineers who activate data. Robert Musters, medior Data Scientist at Anchormen, will walk you through a successful project (technical and practical). The project is about finding and reading the watermeter numbers from an image automatically. The challenging part about this project was acquiring training data and extract the reading from real-world data (instead of e.g. the MNIST dataset). We solved these challenges using crowd-sourcing (Mechanical Turk) and Deep Learning (object classification and localization) respectively. The Deep Learning model used is YOLO which builds on Convolutional Networks, but extends it with localization using regression (and some other tricks)