Improving the Factor Graph Implementation on GPU Problem description:

A Factor Graph is a bipartite probabilistic graphical model representing the factorization of a function. It can be used to represent other probabilistic models such as Markov networks and Bayesian networks. It is also frequently used for performing inference using belief propagation. Such a factor graph can be used to model brain-style information processing by emulating the behaviour of cognitive systems. The basic framework of factor graphs using a GPU has been implemented at NST. However, this initial implementation does not exploit the inter-thread communication in the GPU yet. In this practical course, the student is supposed to improve the current GPU implementation of the Factor Graph. It is required that the student has knowledge in computer programming in C/C++. The GPU programming itself will be learned during the project. Tasks: • Learn the existing Factor Graph framework in NST. • Learn the GPU programming and improve the current Factor Graph implementation. • Develop additional test scenario and evaluate the performance. Bibliography: [1] Sugiarto, I., Conradt, J. ”Factor Graph Inference Engine on the SpiNNaker Neural Computin