Structured Prediction Energy Networks
Authors: David Belanger, Andrew McCallum
ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments provide impressive performance on a variety of benchmark multi-label classification tasks, demonstrate that our technique can be used to provide interpretable structure learning, and illuminate fundamental trade-offs between feedforward and iterative structured prediction. |
| Researcher Affiliation | Academia | David Belanger BELANGER@CS.UMASS.EDU Andrew Mc Callum MCCALLUM@CS.UMASS.EDU College of Information and Computer Sciences, University of Massachusetts Amherst |
| Pseudocode | No | The paper mentions "Appendix A.2 provides a computation graph for this architecture," but this is a graph, not pseudocode or an algorithm block. |
| Open Source Code | No | The paper does not include any explicit statements about releasing source code or links to a code repository. |
| Open Datasets | Yes | Table 1 compares SPENs to a variety of high-performing baselines on a selection of standard multi-label classification tasks. Dataset sizes, etc. are described in Table 4. ... In Table 2, we consider the 14-label yeast dataset (Elisseeff & Weston, 2001)... |
| Dataset Splits | Yes | For Bibtex and Delicious, we tune hyperparameters by pooling the train and test data and sampling without replacement to make a split of the same size as the original. For Bookmarks, we use the same train-dev-test split as Lin et al. (2014). ... We report hamming error, using 10-fold cross validation. |
| Hardware Specification | No | The paper states: "Prediction, both at train and test time, is performed in parallel in large minibatches on a GPU." This mentions a type of hardware (GPU) but lacks specific details such as the model, memory, or other processor specifications. |
| Software Dependencies | No | The paper describes the methods and architectures used, but does not provide specific version numbers for software libraries, frameworks, or programming languages. |
| Experiment Setup | Yes | We seleced 15 linear measurements (rows of C1 in (5)) for Bookmarks and Bibtex, and 5 for Delicious. Section A.5 describes additional choices of hyperparameters. ... In Table 3 we compare: a linear classifier, a 3-Layer Re LU MLP with hidden units of size 64 and 16, and a SPEN with a simple linear local energy network and a 2layer global energy network with Hard Tanh activations and 4 hidden units. |