Introduction to the topic of the workshop
Multi-target prediction (MTP) is concerned with the simultaneous prediction of multiple target variables of diverse type, such as binary, nominal, ordinal, real-valued or even mixed. Often, these multiple target variables are related either explicitly, for example they could represent a ranking, be nodes of a graph or have a spatial, temporal or spatiotemporal relationship, or implicitly, for example via hidden mutual exclusion or parent-child relationships.
Recent years have witnessed a proliferation of novel and revamped MTP applications due to our society's increased ability to collect and store large amounts of data from everywhere and at any-time. Example MTP applications range from social media monitoring, where semantic indexing of media content is performed according to multiple binary variables, to energy-related forecasting, such as wind/solar energy production forecasting and load/price forecasting, where numerical prediction at multiple sites is required, to smart city monitoring, where prediction of multiple environmental variables such as humidity, luminosity, temperature, noise and traffic is needed. Moreover, several important biomedical MTP applications have recently emerged too. In the fields of computational biology and bioinformatics, example MTP applications include multi-species protein function prediction and model organism phenotype prediction, while in the fields of medicine and pharmacology, example MTP applications include discovery of gene-disease associations and genetic variants of diseases, drug repositioning and drug-target prediction problems. MTP problems have literally become ubiquitous.
Objectives and targeted audience
While some progress in developing efficient and effective MTP methods has been achieved, we are still far from being able to successfully apply them to big data, an area which is expected to bring new and smart growth opportunities for Europe . Big data are often described via a number of "Vs", the most important ones for research being Volume, Velocity, Variety and Veracity. Complementary to existing big data efforts that address the "Vs" of input variables, the BigTargets workshop will focus on the "Vs" of the target (output) variables.
Despite the encouraging progress that has been made in the last decade, the current understanding of multi-target learning tasks and methods remains shallow. Further communication and education on the fundamental insights for this type of problems is still required. To date, it remains unclear which of the numerous approaches recently proposed performs better and under what assumptions. Therefore, the workshop intends to cover an overview of existing methods, while focussing on general learning principles instead of algorithm-specific details. With the workshop we aim to attract both researchers that are already active in one of the above domains, as well as researchers with little or no prior experience in multi-target prediction. As such, we believe that the workshop will attract ECML attendees from diverse subfields of machine learning and with different backgrounds.