Learning Expected Hitting Time Distance
Authors: De-Chuan Zhan, Peng Hu, Zui Chu, Zhi-Hua Zhou
AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we evaluate the proposed LED approaches by comparing them with 5 state-of-the-art distance metric learning methods. Datasets used in our empirical investigations are from image and text classification problems. |
| Researcher Affiliation | Academia | De-Chuan Zhan and Peng Hu and Zui Chu and Zhi-Hua Zhou National Key Laboratory for Novel Software Technology Collaborative Innovation Center of Novel Software Technology and Industrialization Nanjing University, Nanjing 210023, China {zhandc, hup, chuz, zhouzh}@lamda.nju.edu.cn |
| Pseudocode | Yes | Procedures for stochastic version of LED are listed in Algorithm 1. |
| Open Source Code | No | The paper does not provide any statements or links regarding the public availability of its source code. |
| Open Datasets | Yes | Datasets used in our empirical investigations are from image and text classification problems. In detail, 3 image and 2 text datasets are used in our experiments, and the brief datesets descriptions are summarized in Table 1. For PASCAL-VOC2009, the goal is to recognize objects from 20 visual object classes in realistic scenes. Cal Tech101 collects pictures of objects from 102 categories and contains 40 to 800 images per category. Reuters-21578 is a collection of documents that published on Reuters Newswire in 1987, from which a subset of 8 classes according to (Cachopo 2007) is used in our experiments. 20Newsgroups is a collection of newsgroup documents from 20 different newsgroups, each of which corresponds to a different topic. |
| Dataset Splits | No | The paper states '2/3 instances are used as training examples while remains are used for testing' and 'Experiments on each dataset are performed 30 times with random splits', but does not mention a validation split or strategy. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware used for running its experiments. |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers. |
| Experiment Setup | Yes | 5-NN is utilized as the final classifier, while all other parameters of compared methods are tuned according to the original reports respectively. For LED and LEDSGD, the probability transition matrix T is initialized randomly and projected to a simplex described by the constraints in Eq. 4. Step size is tuned with line search, and λ is simply configured to 1 for default. |