Noise-Robust Semi-Supervised Learning by Large-Scale Sparse Coding
Authors: Zhiwu Lu, Xin Gao, Liwei Wang, Ji-Rong Wen, Songfang Huang
AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The experimental results on several benchmark image datasets show the promising performance of the proposed algorithm.In this section, we evaluate the proposed LSSC algorithm on four large image datasets listed in Table 1 |
| Researcher Affiliation | Collaboration | 1School of Information, Renmin University of China, Beijing 100872, China 2Computer, Electrical and Mathematical Sciences and Engineering Division, King Abdullah University of Science and Technology (KAUST), Thuwal, Jeddah 23955, Saudi Arabia 3Key Laboratory of Machine Perception (MOE), School of EECS, Peking University, Beijing 100871, China 4IBM China Research Lab, Beijing, China |
| Pseudocode | Yes | Algorithm 1 Large-Scale Sparse Coding (LSSC) |
| Open Source Code | No | The paper does not explicitly state that source code for the described methodology is available, nor does it provide any links to a code repository. |
| Open Datasets | Yes | In this section, we evaluate the proposed LSSC algorithm on four large image datasets listed in Table 1, where MNIST HALF and NUS HALF are derived from MNIST1 and NUS WIDE2 (Chua et al. 2009), respectively. 1http://yann.lecun.com/exdb/mnist/ 2http://lms.comp.nus.edu.sg/research/NUS-WIDE.htm |
| Dataset Splits | Yes | We follow the strategy of parameter selection used in (Chapelle and Zien 2005) and select the rest parameters (i.e. σ, r, and m) for our LSSC algorithm by five-fold crossvalidation over initial labeled images (to generate the labeled set and validation set, with the other images forming the unlabeled set). In our experiments, the fold generation process is repeated randomly two times for a total of 10 splits of each dataset. |
| Hardware Specification | Yes | All these methods are implemented in MATLAB 7.12 and run on a 3.40 GHz, 32GB RAM Core 2 Duo PC. |
| Software Dependencies | Yes | All these methods are implemented in MATLAB 7.12 and run on a 3.40 GHz, 32GB RAM Core 2 Duo PC. |
| Experiment Setup | Yes | We find that our LSSC algorithm is not sensitive to λ in our experiments, and thus fix this parameter at λ = 0.01 for all the four datasets. Meanwhile, we uniformly set k = 5000 by considering a tradeoff of running efficiency and effectiveness. We follow the strategy of parameter selection used in (Chapelle and Zien 2005) and select the rest parameters (i.e. σ, r, and m) for our LSSC algorithm by five-fold crossvalidation over initial labeled images (to generate the labeled set and validation set, with the other images forming the unlabeled set). For example, we set σ = 0.4, r = 3, and m = 18 for the two handwritten digit datasets. |