Exact MAP-Inference by Confining Combinatorial Search With LP Relaxation
Authors: Stefan Haller, Paul Swoboda, Bogdan Savchynskyy
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show efficacy of our method on publicly available benchmarks from computer vision, machine learning and bio-imaging. Experimental Evaluation Algorithms In this section we compare our proposed algorithm with other related methods. As baselines we use CPLEX 12.6.2 (CPLEX, IBM 2014) and Toul Bar2 0.9.8.0 (Cooper et al. 2010) |
| Researcher Affiliation | Academia | Stefan Haller, Paul Swoboda, Bogdan Savchynskyy, University of Heidelberg, IST Austria stefan.haller@iwr.uni-heidelberg.de |
| Pseudocode | Yes | Algorithm 1 Conceptual Dense-Combi LP Algorithm. Algorithm 2 Dense-Combi LP Algorithm. |
| Open Source Code | Yes | Code is available at github.com/fgrsnau/combilp. |
| Open Datasets | Yes | Datasets We verify performance of the algorithms on the following publicly available datasets: worms (Kainmueller et al. 2017), color-seg (Lellmann and Schn orr 2011), mrf-stereo (Scharstein and Szeliski 2002) and On Call Rostering (Stuckey et al. 2014), proteinfolding (Yanover, Schueler-Furman, and Weiss 2008). |
| Dataset Splits | No | The paper describes the datasets used (worms, color-seg, mrf-stereo, On Call Rostering, protein-folding) but does not provide specific train/validation/test splits or how the data was partitioned for experimental evaluation. It mentions solving 'instances' but not how those instances relate to specific data splits like train/validation/test sets. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory, or specific cloud instances) used for running the experiments. |
| Software Dependencies | Yes | As baselines we use CPLEX 12.6.2 (CPLEX, IBM 2014) and Toul Bar2 0.9.8.0 (Cooper et al. 2010) |
| Experiment Setup | Yes | We set the maximum number of TRW-S/SRMP iterations to 2000. Furthermore we tested the performance of a recent partial optimality technique (Shekhovtsov, Swoboda, and Savchynskyy 2015) which is denoted by popt. As this approach does not solve the whole problem, we run Toul Bar2 on the reduced model and measure the total running time (popt-tb2). We set the maximal running time for all methods to 1 hour. |