TF Labs is a dedicated research and development function set up to help our clients benefit from rapid technological advancement.
How do we achieve this?
- A culture that encourages leading-edge techniques in analytics and actuarial analysis – a tradition since our earliest days.
- An approach that recognises our talented and innovative thinkers at all levels of the business. Each year, we rotate staff into pure research tasks to develop and apply new techniques.
- Research priorities driven by our client needs. Whether it’s finding ways to incorporate new types of data, speed up modelling or turn technical output into useful management tools, we’re determined as industry leaders to ensure our clients get the best possible advice.
We highlight some of our current TF Labs projects below ...
Deep learning for processing image and video
A project to monitor live image and video feeds using a combination of open source technology and internal code. This can be used to monitor road congestion or retail foot-traffic.
More generally, the ability to process and predict from image data is increasingly important in understanding a diverse range of outcomes, including insurance risk, claim triage and health diagnosis.
Governments and business are paying increased attention to the ethical implications of machine learning models. Not only do models need to be well designed to avoid unintended unfairness, but management need the ability to ‘prove’ that a model is behaving appropriately.
We have developed our offering around a range of techniques that can understand model bias, report on fairness implications, provide interpretations of model behaviour and provide governance around model implementations.
Using the cloud for the entire analytic pipeline
One of the big challenges in modern analytics is ensuring the right data is moved to the right place at the right time. Cloud services such as those provided by Amazon, Microsoft and Google offer significant flexibility in setting up pipelines so that services can be run automatically and reliably.
The project looks at the variety of ways to deliver a full suite of service from data collection and model scoring, through to visualisation reporting.
Speeding up microsimulation
We build complex microsimulation models to help many of our clients understand both long-term dynamics, and cohort-level insights. These models can be time-consuming to both fit and run.
We use new statistical techniques and powerful parallel processing to significantly reduce the time these models take to produce insights.