When it comes to time series forecasting, we usually resort to time-proven, reliable, strikingly elegant in her simplicity ARIMA. In R, with auto.arima, you can get a decent forecast in a single line of code. In a world where deep learning breaks new records almost every week (it seems), you might ask, how well are deep neural networks doing here? Can we keep up with ARIMA, perhaps even do some things better? In this session, we’ll find out. We’ll have ARIMA and a deep network play against each other, each providing forecasts for systematic artificial benchmarks as well as real world datasets!
Sigrid Keydana is a data scientist with the DACH-based IT consulting company Trivadis.
In the field of data science and machine learning, she focuses on deep learning (concepts and frameworks), statistical learning and statistics, natural language processing and software development using R.
She has a broad background in software development (esp. Java and functional programming languages like Scheme and Haskell), database administration, IT architecture and performance optimization.
She writes a blog (http://recurrentnull.wordpress.com) and is active on Twitter as @zkajdan.