Jul 22, 2021
No image
Service categories
Service Lines
Artificial Intelligence
Software Development
Domain focus
Banking & Financial Services
Other
Programming language
Python
R
Challenge
University of Pavia and Catholic University of Milan were applying for a European project aiming at improving rehabilitation after invasive surgery.
We saw the opportunity to apply the same research to different fields. A global consulting firm entered the project providing funds and business expertise.
An interdisciplinary project was born, with the ambitious goal of teaching machines to recognize the psychological traits of individuals, and create optimized interaction policies for each individual.
University of Pavia and Catholic University of Milan were applying for a European project aiming at improving rehabilitation after invasive surgery.
We saw the opportunity to apply the same research to different fields. A global consulting firm entered the project providing funds and business expertise.
An interdisciplinary project was born, with the ambitious goal of teaching machines to recognize the psychological traits of individuals, and create optimized interaction policies for each individual.
Solution
The first topic where DeepPsyMe was applied was insurance. The social psychology team of Catholic University of Milan provided theoretically sound models for insurance awareness.
We created PsyMe, a mobile app to collect data from volunteers and identify the parameters of the model. Then, in collaboration with University of Pavia, we created a Bayesian network, a probabilistic predictor simulating the effects of different interaction policies on different individuals.
A Deep Reinforcement Learning agent was trained to choose the best interaction policy while asking the least questions to the user.
An algorithm to improve the model and the policy from collected data was designed and integrated into the system.
The first topic where DeepPsyMe was applied was insurance. The social psychology team of Catholic University of Milan provided theoretically sound models for insurance awareness.
We created PsyMe, a mobile app to collect data from volunteers and identify the parameters of the model. Then, in collaboration with University of Pavia, we created a Bayesian network, a probabilistic predictor simulating the effects of different interaction policies on different individuals.
A Deep Reinforcement Learning agent was trained to choose the best interaction policy while asking the least questions to the user.
An algorithm to improve the model and the policy from collected data was designed and integrated into the system.
Results
The research was presented in several papers appearing on international peer-reviewed journals. Explainability and ethics played a key role in the development: we discussed the topics in several public events.
The system sparked the interest of several major financial institutes, yet the prototypes are still not ready to be put into production and more research is required before commercial go-live.
The research was presented in several papers appearing on international peer-reviewed journals. Explainability and ethics played a key role in the development: we discussed the topics in several public events.
The system sparked the interest of several major financial institutes, yet the prototypes are still not ready to be put into production and more research is required before commercial go-live.
No image
No image
