Structure Aware L1 Graph for Data Clustering
Authors: Shuchu Han, Hong Qin
AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental Results To evaluate the performance of our proposed algorithm, we exam it through spectral clustering applications and compare it to different graphs: Gaussian similarity (GS) graph and L1 graph. Six UCI datasets are selected. The clustering performance is measured by Normalized Mutual Information(NMI) and Accuracy(AC). In our experiment setting, we select α = 0.99 for manifold ranking, and K equals to 10%,20% and 30% percent of total number of data samples. Our experiment results show that SA-L1 graph has better clustering performance than L1 graph generally. |
| Researcher Affiliation | Academia | Shuchu Han, Hong Qin Computer Science Department Stony Brook University Stony Brook, NY 11790 |
| Pseudocode | Yes | Algorithm 1: SA-L1 graph Input :Data samples X = [x1, x2, , xn], where xi X; Parameter K; Output:Adjacency matrix W of sparse graph. 1 Calculate the manifold ranking score matrix F; 2 Normalize the data sample xi with xi 2 = 1; 3 for xi X do 4 Select top K atoms from F(i), and build ˆΦ i ; 5 Solve: min αi αi 1, s.t. xi = ˆΦ iαi, αi 0; 6 W(i, :) = αi; 8 return W; |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. |
| Open Datasets | Yes | Six UCI datasets are selected. |
| Dataset Splits | No | The paper mentions 'Six UCI datasets are selected' but does not provide specific data split information for training, validation, or test sets. |
| Hardware Specification | No | The paper does not provide specific hardware details (like GPU/CPU models or memory) used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details with version numbers. |
| Experiment Setup | Yes | In our experiment setting, we select α = 0.99 for manifold ranking, and K equals to 10%,20% and 30% percent of total number of data samples. |