Combination Rules Cores

Typical combination rules that are most commonly used include:

• majority voting,
• weighted majority voting and
• behaviour knowledge spaces.

Since ensemle members can be evaluated sequentially, one at a time, or in parallel, each combination rule should have two possible versions: serial and parallel.

So-Logic offers IP cores that can be used to implement selected combination rule directly in hardware. In conjuction with appropriate ensembe evaluation core these cores can be used to create an fully functional ensemle classifier in hardware.

It should be stated that So-Logic's combination rule IP cores can be used to combine predictions of individual ensemble members, independent of their type. This means that using these cores it is possible to combine predictions of ensembles composed from artificial neural networks, decission trees, support vector machines, or even ensembles that are composed from mixture of different prediction models.

Below you can find a list of currently available ensemble combination rule IP cores:

• Behavior Knowledge Space Combination Core - Parallel Architecture
• Behavior Knowledge Space Combination Core - Serial Architecture
• Weighted Majory Voting Combination Core - Parallel Architecture
• Majory Voting Combination Core - Parallel Architecture
• Majory Voting Combination Core - Serial Architecture