#37: MLDublin meets Optum

Besides providing critical clinical care, hospitals also need to pay the bills to keep the lights on. In the average US hospital, significant effort is involved in reviewing claims before submission to insurance providers. This is effort that could be better be assigned to providing critical care.
This work describes a Deep Learning solution that can review hospitals’ admission charts which are too complex to manage with traditional rules-based systems and recommends the category under which the hospital can submit their claims to their insurance provider. This system can review more admission charts than the traditional human-based process without the fatigue that impacts humans. This ensures as little time as possible is spent reviewing cases for insurance payments and more time is spent treating patients.

A short but practical introduction to using CNN-based auto-encoders for document processing including a comparison of different architectures. Potential uses include identifying clusters of similar documents within unstructured data sets which can be used for RPA and work queue optimization.

This talk discusses our recent work, in collaboration with Huawei Ireland Research Centre, in creating a next-generation advert-creation system for product placement and embedded marketing. We will introduce the several components in our advert-creation system -- detecting adverts, localizing them within frames, and integrating new adverts, to create a personalised augmented video.