I presented our work on GPU Fractional Differencing in Visa Data Summit Singapore.
Typically we attempt to achieve some form of stationarity via a transformation on our time series through common methods including integer differencing. However, integer differencing unnecessarily removes too much memory to achieve stationarity. An alternative, fractional differencing, allows us to achieve stationarity while maintaining the maximum amount of memory compared to integer differencing. While existing CPU-based implementations are inefficient for running fractional differencing on many large-scale time series, our GPU-based implementation enables rapid fractional differencing with up to 400x speed-up over a CPU implementation.