Lifted Probabilistic Inference for Asymmetric Graphical Models
Authors: Guy Van den Broeck, Mathias Niepert
AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments demonstrate that the approach outperforms existing MCMC algorithms. We conduct experiments where, for the first time, lifted inference is applied to graphical models with no exact symmetries and no color-passing symmetries, and where every random variable has distinct soft evidence. |
| Researcher Affiliation | Academia | Guy Van den Broeck Department of Computer Science KU Leuven, Belgium guy.vandenbroeck@cs.kuleuven.be Mathias Niepert Computer Science and Engineering University of Washington, Seattle mniepert@cs.washington.edu |
| Pseudocode | Yes | The general Lifted Metropolis-Hastings framework can be summarized as follows. 1. Obtain an approximate automorphism group G; 2. Run the following mixture of Markov chains: (a) With probability 0 < α < 1, apply the kernel of the base chain MB; (b) Otherwise, apply the kernel of the orbital Metropolis chain MS for G. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing source code or a link to a code repository. |
| Open Datasets | Yes | For our first experiments, we use the standard Web KB data set, consisting of web pages from four computer science departments (Craven and Slattery 2001). We also ran experiments on the Chimera model which has recently received some attention as it was used to assess the performance of quantum annealing (Boixo et al. 2013). |
| Dataset Splits | No | The paper discusses the datasets used but does not specify the exact training, validation, and test splits (e.g., percentages or sample counts). |
| Hardware Specification | No | The paper does not specify the hardware (e.g., CPU, GPU models, memory) used to conduct the experiments. |
| Software Dependencies | No | The paper mentions using 'Alchemy' but does not provide specific version numbers for software dependencies or libraries. |
| Experiment Setup | Yes | Figure 3 shows that the LMH chain, with mixing parameter α = 4/5, has a lower KL divergence than Gibbs and Lifted MCMC vs. the number of iterations. Next we construct an OSA using a rank-5 approximation of the link structure (Van den Broeck and Darwiche 2013) and group the potential weights into 6 clusters. |