VISIONS AND OBJECTIVES
Fulfilling the promise of the big data revolution, the center produces analytical tools to extract knowledge from complex data and delivers BigInsight. Despite extraordinary advances in the collection and processing of information, much of the potential residing in contemporary data sources remains unexploited. The value does not reside in the data alone, but in the methods to extract knowledge from them.
Digitalisation means producing data, organizing and storing data, accessing data and analyzing data. BigInsight works in this last direction. There is a dramatic scope for industries, companies, and nations – including Norway – to create value from employing novel ways of analysing complex data. The complexity, diversity and dimensionality of the data, and our partner’s innovation objectives, pose fundamentally new challenges to statistics and machine learning. We develop original, cutting-edge statistical, mathematical and machine learning methods, produce high-quality algorithms implementing these approaches and thereby deliver new, powerful, operational solutions.
“”Kunstig intelligens som er klar om tre–fire år vil hjelpe oss å gjøre helse og utdanning mye bedre.””
BigInsight’s research converges on two central innovation themes:
personalised solutions: to move away from operations based on average and group behaviour towards individualised actions
predicting transient phenomena: to forecast the evolution of unstable phenomena for system or populations, which are not in equilibrium, and to design intervention strategies for their control
Our solutions are courageous and creative, exploit knowledge and structure in complex data and integrate these from various sources.
Our research is open: we publish generic methodology and their new applications in international scientific journals.
Through training, capacity building and outreach, BigInsight contributes to growth and progress in the private and public sector, in science and society at large, preparing a new generation of statisticians and machine learners ready for the knowledge based economy of the future.
““Statistics is the science of learning from data, and accounting for relevant uncertainties. As such, it permeates the physical, natural, and social sciences, as well as public health, medicine, technology, business, and policy.””
Innovation themes and BigInsight’s objectives
The industrial, business and public partners of BigInsight have different core activities, yet they shall unite in the centre to attack together a set of common challenges, the solution of which will shape their coming years and their enterprise identities:
The first common theme for our partners is to offer a radically new collection of products, services and instruments that adapt to and target individual needs and conditions, thus providing dramatically improved quality and efficacy. Identifying highly specific segments allows tailoring products and services more precisely. From each partner’s perspective, individual stands for customer, user, patient, citizen, ship, company, sensor, smart power meter, tax payer, etc
The second common objective for our partners is to empower their own decisions with precise predictions of critical quantities, which are unstable and in dynamic transition, in order to enable intervention and control. Again, the future quantities for each partner are different –customer retention probabilities, cancer survival, electricity prices, probability of success of a new product, recovered taxes, service reliability, etc.
Therefore BigInsight identifies two Central Innovation Themes that mirror these key challenges and are supported by high quality data of unprecedented availability, at the needed scales.
• Predicting transient phenomena
Modern measurement instruments, the new demands of markets and society and a widespread focus on data acquisition, is often producing high frequency time series data. As never before, we are able to measure processes evolving while they are not in a stable situation, not in equilibrium. A patient receiving treatment, a sensor on a ship on sea, a customer offered products from several providers, a worker who lost his job, the price of an asset in a complex market – all examples of systems in a transient phase. Our partners are interested in the prediction of certain behaviours of their customers and service users, predicting churn or fraud activities. In the health area, the availability of real time monitoring of patients and healthcare institutions allows completely new screening protocols and treatment monitoring, real time prevention and increased safety. High dimensional times series are generated by sensors monitoring a ship, with the purpose of predicting operational drifts or failures and redesigning inspection and maintenance protocols. The objective is to predict the dynamics, the future performance, and the next events. Importantly, real time monitoring of such transient behaviour and a causal understanding of the factors which affect the process, allow optimal interventions and prevention. While the concrete objectives are diverse, we exploit very clear parallels:
systems operate in a transient phase, out of equilibrium and exposed to external forces;
in some cases, there are many time series which are very long and with high frequency; in other cases, short and with more irregular measurements;
there is a complex dependence structure between time series;
there can be unknown and complex causes of observed abnormal behavior;
there is possibilities to intervene to retain control.
BigInsight develops new statistical methodology that allow our partners to produce new and more precise predictions in unstable situations, in order to make the right decisions and interventions.
• Personalised solutions
The core operation of our partners involves interacting with many individual units: customers, users, patients, but also sensors, vessels, wind-turbines, etc. Beside their obvious differences, there are many common characteristics:
the high number of units / individuals / sensors under consideration;
in some cases, massive data for each unit; in other cases, more limited information per unit;
complex dependence structure between units;
new data types, new technologies, new regulations make their use innovative;
in most cases, units have their own intelligence, their own strategies, and are exposed to their specific environment.
Each partner has specific objectives for and with their units, but they share the goal to fundamentally innovate the management of their units, by recognising similarities and exploiting diversity between units. This will allow personalised marketing, personalised products, personalised prices, personalised recommendations, personalised risk assessments, personalised fraud assessment, personalised screening, personalised therapy, sensor based condition monitoring, individualised maintenance schemes, individualised power production and more – each providing value to our partner, to the individuals and to society: better health, reduced churn, strengthened competitiveness, reduced tax evasion, improved fraud detection and optimised maintenance plans.
“”As a first step, industrial leaders could gain a better understanding of AI technology and how it can be used to solve specific business problems.
They will then be better positioned to begin experimenting with new applications.””
Methods
We solve innovation challenges of our partners by developing new or applying state-of-the-art statistical, mathematical, and machine learning methods in these fields:
Probabilistic Bayesian models and forecasting
Complex dependence models
Latent variable models
Data integration and knowledge incorporation
Physics informed machine learning
Knowledge based machine learning
Multi-type and multi scale models
Scalable approximation algorithms
High dimensional data and time series
Change and anomaly detection and prediction
Local and global explanations of black box models
Networks and graphical models
Preference learning
Mechanistic multi-type models in oncology
Time-to-event models
