Internal/MachineLearning

Categories: Category 1 Category 2 Category 3 Category 4

• Design knowledge representation models that allow for efficient learning, while being flexible enough to capture different aspects of the data simultaneously and take into account different kinds of initial domain and expert knowledge

• Models capable of reusing knowledge and adapting over time, from a generic model to a specific model. For example, by using measures of behavior change and user satisfaction as goal functions. To develop decision support systems for chronic conditions that always respect medical guidelines but that can adjust to a particular subject's needs and resources.

• Scalable machine learning algorithms suitable for handling massive streams of data in a distributed and self-organising manner. Models designed to process large amounts of data efficiently, provide end users with descriptive and explanatory analysis and can be understood by experts in the application domain.

• Methods for near-optimal combination of all relevant, often incomplete, uncertain, and contradictory information. Provide degree of confidence in fused information.

• High level functionality and visualization for increased usefulness and usability of machine learning algorithms. Current tools provide, too little analytical support and very basic visualization functionality, especially for modelling streaming data.

• Find quantitative metrics of "interestingness" of both data and extracted knowledge, they should be adaptable to different tasks/domains and work for both supervised and unsupervised learning.

• Algorithms to build and explore data dependent committees of models.

• Algorithms for exploring the relationship between objective sensor data and subjective data, obtained via questionnaires in health care for example.

• Algorithms for segmentation of time series on multiple time scales.

• Algorithms for guided search through structured and semi-structured expert and historical knowledge, often uncertain.

• Obtain deeper understanding of fundamental concepts of self-organisation and self-awareness of systems, allowing us to build a general theory on top of successful application examples.

• Develop models able to decide when to make use of either a data-driven approach or a model-driven approach (or a combination thereof). Example of research question: How could predict/detect a deviating behaviour by fusing data-driven methods such as Random Forests with neuropsychological/behavioural models?

• Design algorithms and tools for deriving simple and expressive knowledge from (less opaque) learnt models. Example of research question: How do we extract knowledge about a deviating behaviour given a sequence of observations from a smart home and a ensemble of trees? How do we communicate the derived information to a caregiver (with limited skills in computer science!).