MASH logo
Login | Register

This platform is still under development and may suffer from service interruptions for maintenance.

Massive Sets of Heuristics

The MASH project investigates the design of complex learning systems to increase the performance of artificial intelligence. To create such systems, we want to involve large groups of contributors from as many fields and backgrounds as possible, to program large and heterogeneous families of feature extraction modules.

Performance will be measured on action selection with a robotic arm and in a simulated 3d environment, and object detection.

Research activity will be organized on this web platform which will be fully operational in the coming weeks. Anybody with some knowledge in C++ is strongly encouraged to participate, whatever her or his background and expertise are. You can not hurt the system by getting involved but only improve it, so do not hesitate to contribute or contact us.

This project is funded by the Information and Communication Technologies division of the European Commission, Cognitive Systems and Robotics unit, under the 7th Research Framework Program. The sub-project VELASH on object detection is funded by the Swiss National Science Foundation.

Read more >>

Latest news

MASH SDK 0.6.1 Released

2010-04-29 16:38:39 by Kanma

This is a maintenance release which fix some problems under Linux.

MASH SDK 0.6.0 Released

2010-04-16 12:07:49 by Kanma

New version of the SDK! Existing heuristics don't need to be rewritten.

The heuristics can now implement the following optional methods:
  • init()
  • terminate()
  • prepareForSequence()
  • finishForSequence()
The latter two are mainly usefull when the system is working on a sequence of images (this is the case for goal-planning experiments). Using those methods, a heuristic prepare itself to that case if necessary. In the case of image classification and object detection, each image is considered as a sequence.

The Image class has a new method, view(), that returns the index of the view this image was taken from. In the case of image classification, there is only one view. For the goal-planning, there might be several views.

MASH SDK 0.5.1 Released

2010-03-04 14:45:44 by Kanma

This is a maintenance release which fix some compilation problems when using a 64bits Linux operating system.