The subject application provides a semiconductor system (100) and a computer-implemented method configured to provide probabilistic inference data processing for solving probabilistic model.
The inventors have found that using a dedicated processor-coprocessor architecture enables the acceleration of sampling-based algorithms for solving probabilistic models.
In particular, the inventors propose to configure a precompiler (150) that astutely distribute between a processor and a coprocessor (120), the executable instructions associated with portions of the computer program code (30) implementing the sampling-based algorithms.
With the proposed architecture, the developer of a sampling-based algorithm does not have to explicitly code the drawing of random samples from probability distribution functions or part of a probabilistic fusion of probability distribution functions, since the corresponding executable instructions would be executed by the coprocessor (120) after a function call has been detected for the drawing of random samples from probability distribution functions or a probabilistic fusion.