Startseite // FHSE // News & E... // How HPC can help predict and interpret happiness

How HPC can help predict and interpret happiness

twitter linkedin facebook email this page
Veröffentlicht am Dienstag, den 21. Dezember 2021

Contemporary research requires significant computing power and generates large amounts of data. High-Performance Computers – also called supercomputers – allow to run fast complex and large-scale calculations to create simulations or models of physical systems and test the most advanced large-scale artificial intelligence applications that require massive amounts of calculations per second.

The University hosts two on-site HPC supercomputers (called Iris and Aion), totalling a computing capacity of 2,76 PetaFlops and a shared storage capacity of 10,68 PetaBytes, enabling to sustain research excellence while achieving faster time-to-solution.

The boundless power of supercomputers, using data analytics and artificial intelligence, has become a necessary tool in all disciplines that need to process huge amounts of data.

Niccolo Gentile is a doctoral researcher in behavioural sciences and individual wellbeing, and uses the HPC for psychological computational modelling and forecasting.

He uses the clusters to analyse the wellbeing of individuals in a data-driven manner, based on information shaping the concept of happiness.

Wellbeing can be measured by life satisfaction, happiness or other indices of psychological functioning, via modern nonlinear machine-learning techniques. “Who we are is determined by a myriad of factors” (demographic, socio-economic, biological, psychological and environmental, etc.), and “we are trying to model individuals using a set of characteristics,” Gentile explains. 

Gentile works on Data Science and the Economics of WELLbeing (DSEWELL), a project funded by the University’s Institute for Advanced Studies (IAS) bringing together machine learning approaches, physics-inspired descriptors and the economics of wellbeing to address questions broadly related to predicting wellbeing of individuals in a data-driven manner.

The project focuses on finding the right data descriptors using models inspired by physics and social sciences to identify appropriate metrics to describe individuals and their relationships with others. To that purpose, he uses modern nonlinear machine-learning techniques to analyse data on individual and social wellbeing.

Yet, when using machine-learning techniques, long computing times can easily become a problem even when dealing with only a few data points.

HPC clusters have become an invaluable tool for processing computations on very large and rich datasets, more complex computations on small datasets and to treat files which would be impossible to accommodate on a single computer.

Together with his team, under the supervision of Professors Conchita D’Ambrosio and Alexandre Tkatchenko, Niccolo Gentile works on datasets based on existing studies covering various aspects of life, including health; physical, cognitive and social development; parenting; career paths; and economic circumstances. They also collect information from a variety of sources, including clinical records, medical examinations, nutrition and activity diaries, parents, teachers and cohort members themselves.

These studies include the 1970 British Cohort Study (BCS), that monitored the lives of around 17,000 people born in England, Scotland and Wales in a single week of 1970, and the German Socio-Economic Panel (SOEP), a longitudinal survey of approximately 11,000 private households in the Federal Republic of Germany from 1984 to today and in the Länder of former Eastern Germany from 1990 to today.

The results are very promising. In BCS, with a richer set of controls, machine learning methods yield a notable improvement in predictive accuracy, i.e. the possibility of correctly predicting – or getting close to – the real value of a target variable. Although marital status and emotional and physical health continue to be the most important predictors of life satisfaction, as in the existing literature, gender becomes insignificant in the non-linear analysis, meaning that women are not found to be more satisfied than men, using these methods.

In SOEP, with machine learning the accuracy of the predictions of life satisfaction is increased of about 30%. The improvement is mostly among the least satisfied individuals: we can now better predict who is at risk of depression.

The research studies conducted by the DSEWELL project are part of a new approach in the broader context of improving individuals’ life and predicting life satisfaction and health in a data-driven manner, with substantial potential scientific and social impact.