Scalable Model Selection for Belief Networks
Authors: Zhao Song, Yusuke Muraoka, Ryohei Fujimaki, Lawrence Carin
NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | On both synthetic and real data, our experiments suggest that FABIA, when compared to state-of-the-art algorithms for learning SBNs, (i) produces a more concise model, thus enabling faster testing; (ii) improves predictive performance; (iii) accelerates convergence; and (iv) prevents overfitting. |
| Researcher Affiliation | Collaboration | Zhao Song , Yusuke Muraoka , Ryohei Fujimaki , Lawrence Carin Department of ECE, Duke University Durham, NC 27708, USA {zhao.song, lcarin}@duke.edu NEC Data Science Research Laboratories Cupertino, CA 95014, USA {ymuraoka, rfujimaki}@nec-labs.com |
| Pseudocode | No | The paper describes the FABIA algorithm and its optimization process but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement about the release of their own source code for FABIA, nor does it include a link to a code repository for their method. |
| Open Datasets | Yes | We use the publicly available MNIST dataset, which contains 60, 000 training and 10, 000 test images of size 28 28. ... The two benchmarks we used for topic modeling are Reuters Corpus Volume I (RCV1) and Wikipedia, as in Gan et al. [10], Henao et al. [14]. |
| Dataset Splits | Yes | We simulate 1250 data points, and then follow an 80/20% split to obtain the training and test sets. ... We use the publicly available MNIST dataset, which contains 60, 000 training and 10, 000 test images... RCV1 contains 794,414 training and 10,000 testing documents... Wikipedia is composed of 9,986,051 training documents, 1,000 test documents... Instead, we follow the approach of 80/20% split on the test set, with details provided in Gan et al. [10]. |
| Hardware Specification | Yes | Both FABIA and NVIL are implemented in Theano [4] and tested on a machine with 3.0GHz CPU and 64GB RAM. |
| Software Dependencies | No | The paper mentions 'Theano' but does not specify a version number for it or any other ancillary software. |
| Experiment Setup | Yes | The learning rate in Adam is set to be 0.001 and we follow the default settings of other parameters in all of our experiments. We set the threshold parameter ϵ(l) to be 0.001, l unless otherwise stated. ... The mini-batches for FABIA and NVIL are set to 100. ... For both FABIA and NVIL, we use a mini-batch of 200 documents. ... FABIA starts from a model initialized with 400 hidden units in each layer. |