Research Themes

Main research themes

The activities within the INFER project have been organized around a number of main research themes. To see how the themes map onto the outputs please visit the Publications page.


Soft Sensors and the Process Industry

The focus of this theme was to align the research conducted within the INFER project with requirements, expetatctions and constraints of the deployment environment of the platform. To this end the following activities have been pursued:

  • Requirement analysis and engineering,
  • Development of soft-sensors for a number of case studies provided by Evonik,
  • Usability study.

In September 2013 the INFER team has organised a workshop within the above theme titled INFER Workshop on Data Mining in Process Industry.


Adavanced Adaptation Mechanisms

This theme has focused on developing novel adaptation mechanisms for hierarchical, multi-level, multi-component predictive systems. These mechanisms enable effective support of model self-adaptation and thus prolong their productive life-time. Apart from the low-level adaptation techniques, i.e. incremental learning and forgetting, the emphasis has been put on:

  • Change propagation through the hierarchical structure of a multi-level, multi-component predictive systems,
  • Adaptive pre-processing methods and the effects of their interdependence,
  • Study and development of different adaptive combination methods based on the responses from individual prediction methods from the bottom level of the platform,
  • Investigation and development of novel methods for building and continuous evolution of flexible combination hierarchies.

The above research directions are being further pursued by Mr Rashid Bakirov and Mr Manuel Martin Salvador as the main topics of thier PhD projects.

In August 2011 the INFER team has organised a workshop within the above theme titled INFER Workshop on Adaptive Prediction Systems.


Meta-learning and Methodologies of Predictive Modelling

This theme focused on researching and developing techniques and methodologies of buidling predictive models, which could be automated and hence would allow to greatly simplfy and streamline the modelling process (or search through the space of possible models), and in particular:

  • Meta-learning perceived as a way to constraint the space of models and guide the search. This research direction being further pursued by Mr Abbas Raza Ali as the main topic of his PhD project,
  • Hyper Parameter Optimisation i.e. background optimisation of non-trainable/user-specified parameters of predictive models through various search strategies exploiting the exploration v. exploitation paradigm,
  • Novel methodologies of validating the predictive models which are more robust and/or more computationally efficient than tradiational methods,
  • Flexible and robust system architecture allowing to automate the methodology of building and then maintaning the predictive models, stemming from the previous research within the group (including parts of Dr Kadlec's and Dr Budka's PhD dissetations).

In April 2012 the INFER team has organised a workshop within the above theme titled INFER Workshop on Meta-learning for Complex, Adaptive Prediction Systems.


Complexity Science and Complex Systems

The research in this theme has been focused on widely understood complexity of various relevant types of systems, and in particular:

  • Complexity management of highly flexible, multi-component, multi-level evolving predictive systems and related mechanisms  (diversity, pool sizes, dynamics of adaptation, interactions, competition v. collaboration etc.),
  • Complexity control and trade-offs in the context of  software engineering practices, software architectures, and user interfaces,
  • Emergent behaviour in complex dynamic network systems,
  • Multiobjective optimisation, wihich is a research direction being further pursued by Ms Bassma Al-Jubouri as the main topic of her PhD project.