Community Research and Development Information Service - CORDIS

Next-generation machine learning in Europe

Machine learning is the technology that enables computers to get smarter and more personal, provided that techniques can keep up with the times. An EU initiative developed machine learning tools and algorithms to address this challenge.
Next-generation machine learning in Europe
Current machine learning methods struggle with the growing range and complexity of large numbers of tasks and objectives. As this technology advances and learning algorithms become key components, they will be expected to cope with this added complexity.

With this in mind, the EU-funded HERL (Large scale machine learning for simultaneous heterogeneous tasks) project set out to devise a tool that examines the performance of new and existing algorithms for solving multitask reinforcement learning problems. It also aimed at creating algorithms to handle many and simultaneous objectives.

Project partners first developed a random problem generator for learning problems known as Merlin. It is unique in its ability to generate multitask learning problems with control over the structure of the inter-task relationships. Merlin is now freely available and continually being improved to provide a broader range of problem structures.

The HERL team built a software platform to apply multitask reinforcement learning algorithms. It also analysed data concerning the performance of various learning algorithms that were developed and tested on generated and actual problems. To achieve this, researchers first produced known problem structures and then studied these problems in order to design more effective learning algorithms.

Thanks to HERL, Merlin enables reinforcement learning researchers to assess how their algorithms are able to perform across a greater variety of problems than those currently in use by their peers. The new algorithms can tackle problems with more than two or three simultaneous objectives. Tomorrow's machine learning systems will be more effective, all with less direct human intervention and computational burden.

Related information


Machine learning, algorithms, heterogeneous tasks, reinforcement learning
Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top