Abstract
We extend the class of variational message passing algorithms to approximate fitting and inference for skew t regression models, building on recent
work concerning variational message passing on factor graph. A major advantage
of a factor graph fragment approach is that calculations only need to be done once
for the considered distribution family and can be easily adapted to accommodate
more complex model structures. A simulation study shows how posterior dependence arising in auxiliary variable representation of a skew t response model may
lead to poor performances in terms of variational message passing when using
convenient factorizations of the approximating densities
work concerning variational message passing on factor graph. A major advantage
of a factor graph fragment approach is that calculations only need to be done once
for the considered distribution family and can be easily adapted to accommodate
more complex model structures. A simulation study shows how posterior dependence arising in auxiliary variable representation of a skew t response model may
lead to poor performances in terms of variational message passing when using
convenient factorizations of the approximating densities
| Original language | English |
|---|---|
| Title of host publication | Proceedings of the 33rd International Workshop on Statistical Modelling |
| Pages | 204-208 |
| Publication status | Published - 2018 |
| Event | Proceedings of the 33rd International Workshop on Statistical Modelling - BRISTOL, United Kingdom Duration: 16 Jul 2018 → 20 Jul 2018 |
Workshop
| Workshop | Proceedings of the 33rd International Workshop on Statistical Modelling |
|---|---|
| Country/Territory | United Kingdom |
| City | BRISTOL |
| Period | 16/07/18 → 20/07/18 |