Explainable and scalable statistical models for complex data
Contact person: Ingrid Hob?k Haff
Keywords: Parametric models, explainable models, algorithms, conflict, sensitivity, hierarchical models
Research group: Statistics and Data Science
Department of Mathematics
There is currently an exponential growth of data collected in different areas. Data sources can both be huge and of different types (e.g. images, text, videos, table data, knowledge graphs). Although black box machine learning techniques may be applicable for modelling in some settings, more direct use of statistical modelling will be useful when knowledge about the problem and data sources are available, and are more inherently explainable. Traditional statistical models may however not be directly applicable to the multiple and complex data modalities. Developing new models and methodology to appropriately handle the specific structures and assumptions of these data is necessary. Additionally, conducting statistical inference, model checking and hypothesis testing with the new data modalities requires innovative approaches. Potential applications are within ecology, healthcare, environmental applications, industry and technology.
Methodological research topics:
- Combining data of different modalities through hierarchical/graphical structured modelling
- Complex dependencies in high dimension through scalable copula modelling
- Handling high-dimensional data with regularisation or dimension reduction
- Efficient and scalable algorithms for statistical inference
- Spatio-temporal modelling towards massive ecological data
?
External partners:
- Norwegian Computing Center (NR)
Mentoring and internship will be offered by a relevant external partner