UAI 2016 Inference Evaluation

Introduction

We are pleased to announce an upcoming evaluation of probabilistic approximate inference algorithms, as part of the UAI 2016 Conference. All researchers working on inference in graphical models and statistical relational models are encouraged to participate.

The evaluation will focus on the following computational tasks:

• Computing the partition function and probability of evidence (PR inference)

• Computing the marginal probability distribution over a variable given evidence (MAR inference)

• Computing the most likely assignment to all variables given evidence (MAP inference)

• Computing the most likely assignment to a subset of variables given evidence (Marginal MAP inference)

Each task will be evaluated under three different time constraints: 1 minute, 20 minutes and 1 hour

New this year

Two new problem formats for each of the computational tasks:

• Sparse Model Format

• Binary Model Format

Submitted algorithms can participate in some or all of these tasks.

Sample Problems

UAI Model Format

• 1.uai 1.uai.evid

• 2.uai 2.uai.evid

• 3.uai 3.uai.evid

Sparse Model Format

• 1.suai 1.suai.evid

• 2.suai 2.suai.evid

• 3.suai 3.suai.evid

Binary Model Format

• 1.buai 1.buai.evid

• 2.buai 2.buai.evid

• 3.buai 3.buai.evid

File Format Descriptions

• UAI Model Format

• Sparse Model Format

• Binary Model Format

• Evidence Format

• Query variables, only applies to the marginal MAP task

• Result/Output format