Sequential streams of information are pervasive; things happen and are recorded. These streams can be regular with all channels updating at once like sound. Alternatively, channels can update one at a time and maybe not at all, as things happen. An example of this is an electronic health record – which might capture hospital admission, a blood test, or perhaps a continuing ECG measurement.
Managing this heterogeneous stream of data is a challenge. Often there is important information in the order of events that links the channel behaviour together. In this case smoothing the data out channel by channel, like binning data, is damaging. A powerful unifying approach is to regard the data to be the input, and let it control a dynamical system. Different behaviour can be distinguished via the different responses of the system. One can modify the system so that it is not affected by the aspects of the stream that are of little interest.
With five years funding from UKRI, the DataSıg programme looks to address this key challenge of data science – to better understand multimodal data streams. It seeks to do this by developing mathematical descriptions of these streams, using ‘rough path’ (RP) theory; RP theory allows for the direct capture of the order in which events happen and in many cases can better model the effects of these data streams via a top down signature description of the stream that summarises the data effectively without exposing the individual data points to direct analysis.
This one day workshop highlighted a number of exciting research activities and outlined some of the successful collaborations within the DataSıg programme.