Marginal Structured SVM with Hidden Variables
Authors: Wei Ping, Qiang Liu, Alex Ihler
ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we compare our MSSVM with other state-of-art methods on both simulated and real-world datasets. We demonstrate that the MSSVM significantly outperforms LSSVM, max-margin min-entropy (M3E) model (Miller et al., 2012) and loss-based learning by modeling latent variable(Mod Lat) (Kumar et al., 2012), especially when the uncertainty over hidden variables is high. Our method also largely outperforms HCRFs in all experiments, especially with a small training sample size. |
| Researcher Affiliation | Academia | Wei Ping WPING@ICS.UCI.EDU Qiang Liu QLIU1@ICS.UCI.EDU Alexander Ihler IHLER@ICS.UCI.EDU Department of Computer Science, University of California, Irvine |
| Pseudocode | Yes | Algorithm 1 Sub-gradient Descent for MSSVM; Algorithm 2 CCCP Training of MSSVM |
| Open Source Code | No | The paper does not include any explicit statement about making its source code available or provide a link to a code repository. |
| Open Datasets | Yes | We use the Microsoft Research Cambridge data set (Winn et al., 2005), consisting of 240 images with 213 320 pixels and their partial pixel-level labelings. |
| Dataset Splits | No | The paper specifies training and test splits (e.g., "20 training instances and 100 test instances" or "2-fold cross validation for testing"), but it does not mention a separate validation dataset split. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions using specific algorithms and methods (e.g., "mixed-product belief propagation", "sum-product belief propagation", "SGD", "CCCP") but does not specify any software dependencies with version numbers. |
| Experiment Setup | Yes | We set σx = σy = σh = 0.1, σyh = σyx = σhx = 2. Then, we train our MSSVM, LSSVM and HCRF models using both SGD and CCCP. Hamming loss is used in both training and evaluation. In our experiments, we always set the regularization weight C = 1. |