Probabilistic Inference Modulo Theories
Authors: Rodrigo de Salvo Braz, Ciaran O’Reilly, Vibhav Gogate, Rina Dechter
IJCAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct a proof-of-concept experiment comparing our implementation of SGDPLL(T)-based SGVE(T) (available from the corresponding author s web page) to the state-of-the-art probabilistic inference solver variable elimination and conditioning (VEC) [Gogate and Dechter, 2011], on the election example described above. The model is simple enough for SGVE(T) to solve the query P(like Incumbent > like Challenger|new Jobs = 80000 dow = 17000) exactly in around 2 seconds on a desktop computer with an Intel E52630 processor, which results in 0.6499 for N = 108. The run time of SGVE(T) is constant in N; however, the number of values is too large for a regular solver such as VEC to solve exactly, because the tables involved will be too large even to instantiate. By decreasing the range of new Jobs to 0..100, of dow to 110..180 and N to just 500, we managed to use VEC but it still takes 51 seconds to solve the problem. |
| Researcher Affiliation | Collaboration | Rodrigo de Salvo Braz SRI International Menlo Park, CA, USA Ciaran O Reilly SRI International Menlo Park, CA, USA Vibhav Gogate U. of Texas at Dallas Dallas, TX, USA Rina Dechter U. of California, Irvine Irvine, CA, USA |
| Pseudocode | Yes | Algorithm 2 Symbolic Generalized DPLL (SGDPLL(T)), omitting pruning, heuristics and optimizations. |
| Open Source Code | Yes | We conduct a proof-of-concept experiment comparing our implementation of SGDPLL(T)-based SGVE(T) (available from the corresponding author s web page) to the state-of-the-art probabilistic inference solver variable elimination and conditioning (VEC) [Gogate and Dechter, 2011] |
| Open Datasets | No | The paper uses a custom 'election example' model described in Section 5 with variables like 'attacks Uniform(0..20)' and 'new Jobs Uniform(0..100000)'. This is a described generative model, not an external, publicly accessible dataset with a specified source or link. |
| Dataset Splits | No | The paper does not mention explicit training/test/validation dataset splits, as it primarily focuses on probabilistic inference over a defined generative model rather than a traditional machine learning dataset. |
| Hardware Specification | Yes | The model is simple enough for SGVE(T) to solve the query P(like Incumbent > like Challenger|new Jobs = 80000 dow = 17000) exactly in around 2 seconds on a desktop computer with an Intel E52630 processor |
| Software Dependencies | No | The paper does not provide specific software names with version numbers for ancillary software dependencies used in the experiment. |
| Experiment Setup | No | The paper describes the parameters of the probabilistic model used in the experiment (e.g., 'attacks Uniform(0..20)', 'new Jobs Uniform(0..100000)') and the query, but does not provide specific experimental setup details such as hyperparameters, optimizer settings, or explicit training configurations. |