Implementing Variational Inference for Probabilistic Reasoning


thumbnail

Suitable for:
Student, Bachelor or Master project

Description:

Bayesian Networks and other graphical models for probabilistic reasoning are usually intuitive to use since their graphical nature allows easier interpretation. Probabilistic reasoning in general can be used for a wide range of scenarios and humans are argued to process information more or less in a probabilistic Bayesian way (subject to approximations of course). Our own inference implementation PRIMO (https://gitlab.ub.uni-bielefeld.de/scs/PRIMO) already allows the use of different kinds of networks as well as inference methods, however it does not yet make use of Variational Infer- ence, an approximate inference method that finds the desired distributions via optimization.
This is primarily intended as a Bachelor Thesis or student project, but could be extended for a Master Thesis as well.


Contact:
Jan Pöppel