Exact Sampling with Integer Linear Programs and Random Perturbations
Authors: Carolyn Kim, Ashish Sabharwal, Stefano Ermon
AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We empirically demonstrate that our approach parallelizes well, our exact sampler scales better than alternative approaches, and our approximate sampler yields better quality samples than a Gibbs sampler and a low-dimensional perturbation method. |
| Researcher Affiliation | Collaboration | Carolyn Kim Computer Science Department Stanford University ckim@cs.stanford.edu Ashish Sabharwal Allen Institute for AI Seattle, WA ashishs@allenai.org Stefano Ermon Computer Science Department Stanford University ermon@cs.stanford.edu |
| Pseudocode | Yes | Algorithm 1 describes our sampling method, called Gumbel MIP, that takes as input an ILP Q. |
| Open Source Code | No | The paper states 'We implemented Gumbel MIP on top of the commercial ILP solver CPLEX using callbacks' but does not provide any link or explicit statement about the availability of their own source code. |
| Open Datasets | Yes | We consider synthetic Ising models with n binary variables... trained on MNIST (handwritten digits) with constrastive divergence (Carreira Perpinan and Hinton 2005). |
| Dataset Splits | No | The paper mentions using 'synthetic Ising models' and 'MNIST' for evaluation but does not specify explicit training/validation/test dataset splits with percentages, sample counts, or citations to predefined splits. |
| Hardware Specification | No | The paper states 'We experimented with up to 48 threads' and mentions 'parallel solving capabilities' and 'multi-threaded search' but does not specify any particular CPU, GPU, or other hardware model used for experiments. |
| Software Dependencies | No | The paper states 'We implemented Gumbel MIP on top of the commercial ILP solver CPLEX' but does not provide a specific version number for CPLEX or any other software dependency. |
| Experiment Setup | No | The paper describes the overall approach and some high-level experimental conditions (e.g., '4 hour timeout', '200 independent samples' for some cases), but does not provide specific hyperparameter values (e.g., learning rate, batch size, epochs for RBM training) or detailed system-level training settings. |