Deep Learning in text and audio

2 talks:

1. Extracting information from documents with deep learning

Scanned invoice in, digital semantic invoice out. Easy? Not so much. Especially if you want it to work for any invoice, with no setup or pre-annotation required. We'll present our iterations over the years on trying to create the 'perfect' machine learning model that works for every weird invoice layout, and under some hard constraints on the data quality. This being research, we take some detours into solving Sudokus and predicting the end of the world.

Rasmus Berg Palm
Rasmus joined Tradeshift in 2012. He started and lead ML team until 2016, during that time he built the CloudScan machine learning engine. Right now Rasmus is an industrial PhD student at DTU/TS.

2. End-to-end medical augmentation from audio

Lars Maaløe
Co-founder and Chief Technology Officer at Corti with a background within machine learning research. Lars has mainly focussed his research within unsupervised and semi-supervised learning on images, text and audio.