Joint Inference Walkthrough
Joint inference uses the Cyclades algorithm, which is an algorithm for parallel stochastic optimization. Cyclades maintains serializability (avoids conflicting read/writes to the parameters in a multicore setting) so optimizing with 8 threads produces the same result as optimizing with 1 thread. Serializability is important as conflicts may lead to non-convergence or poorer results.
Cyclades is a partitioning algorithm that creates a schedule of non-conflicting stochastic updates.