Managing sensitive data
The successful application of data science to law enforcement and other operational data requires collaborative effort across the workflow pipeline for the development, training, testing, debugging, and integration of models and their outputs.
One of the challenges for this collaboration is the constraint on data access due to sensitivities in the data or models themselves for a variety of legal, security, ethical, or practical reasons. Another is the requirement to use data from multiple organisations to increase the quality of modelling as well as to mitigate against forms of bias due to narrow datasets.
In response, one of our major research themes is the design of infrastructure for facilitating ‘eyes-off’ access to restricted data in the pursuit of collaborative data science research and development. Originally conceived for the development of classsifation models for CSAM in conjunction with law enforcement, we are discovering applications for thsi approach across a wide variety of similar domains.