From Stochastic Planning to Marginal MAP
Authors: Hao(Jackson) Cui, Radu Marinescu, Roni Khardon
NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | An experimental evaluation illustrates the potential of AGS in solving difficult MMAP problems. |
| Researcher Affiliation | Collaboration | Hao Cui Department of Computer Science Tufts University Medford, MA 02155, USA hao.cui@tufts.edu Radu Marinescu IBM Research Dublin, Ireland radu.marinescu@ie.ibm.com Roni Khardon Department of Computer Science Indiana University Bloomington, IN, USA rkhardon@iu.edu |
| Pseudocode | Yes | AGS Algebraic Gradient Based Solver for MMAP: 1. Given a MMAP problem G1 with evidence E = e, decision nodes D and sum nodes S use the reduction to obtain a MEU problem G2 with utility node V and decision nodes Dout. 2. Generate the SOGBOFA graph GSOG from the MEU problem where decision nodes are treated as action nodes and V is the Q node of the planning problem. 3. Use the gradient based optimizer in SOGBOFA (gradient ascent with random restarts) to optimize the marginal probabilities of variables {Dout i }. 4. Extract a discrete solution from the marginal probabilities by thresholding: Dout i = 1 iff p(Dout i 0.5). |
| Open Source Code | No | The paper states 'We use our own implementation of MPBP,' and for AGS, but does not provide a link or explicit statement about the availability of their source code. |
| Open Datasets | Yes | For the evaluation we use several problems from the UAI competition 2008. |
| Dataset Splits | No | The paper describes generating MMAP problems and evaluating them but does not specify standard training, validation, and test dataset splits with percentages or counts. |
| Hardware Specification | No | The paper mentions 'Some of the experiments in this paper were performed on the Tufts Linux Research Cluster supported by Tufts Technology Services' but does not specify any hardware details like CPU, GPU models, or memory. |
| Software Dependencies | No | The paper mentions using 'our own implementation' for AGS and MPBP, and the 'implementation of [14]' for AAOBF, but it does not specify any software names with version numbers (e.g., Python 3.x, PyTorch 1.x). |
| Experiment Setup | Yes | MPBP keeps iterating over updates to nodes until it runs out of time or the maximal change of the messages becomes smaller than 0.0001. ... we modified the UAI instances changing any 0 probability to 0.0001 (and 1 to 0.9999). ... The implementation of AAOBF replaces every 0 with 0.000001 for similar reasons. ... for run time of 1,5,10 seconds in corresponding columns. |