Structured Sparsity with Group-Graph Regularization

Authors: Xin-Yu Dai, Jian-Bing Zhang, Shu-Jian Huang, Jia-Jun Chen, Zhi-Hua Zhou

AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on both synthetic and real data show that, enforcing group-graph sparsity lead to better performance than using group sparsity or graph sparsity only.
Researcher Affiliation Academia National Key Laboratory for Novel Software Technology Nanjing University, Nanjing 210023, China {daixinyu,zjb,huangsj,chenjj,zhouzh}@nju.edu.cn
Pseudocode Yes Algorithm 1 The g2-regularization method
Open Source Code No A modified open-source software named SPAMS from http://spams-devel.gforge.inria.fr/ is used to implement our algorithm.
Open Datasets Yes MEMset Dataset (1) Experiment Setup This dataset is available at http://genes.mit.edu/burgelab/maxent/ssdata/. and NN269 Dataset (1) Experiment Setup We use the NN269 dataset for more real-world data evaluation (Reese et al. 1997), which is available at http://www.fruitfly.org/data/seqtools/datasets/Human/GENIE96/splicesets/.
Dataset Splits Yes In addition, we randomly choose another 600 true and 600 false splice sites as validation data in 5 case and 3 case, respectively. and We randomly partition the data into the training and test sets for 10 times, and report the average results as well as standard deviations over the 10 repetitions.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies No A modified open-source software named SPAMS from http://spams-devel.gforge.inria.fr/ is used to implement our algorithm.
Experiment Setup Yes The control parameter of λ in Eq.1 is tuned on the validation data. and The Group and Graph Structures: Based on the priori properties, we generate a graph where the first 10 groups are connected as a path and the cost of each edge on this path is 0.05. Other groups are isolated in the graph. For the edges of the source node s and the sink node t, we set {csu = 0 |u V } and {cut = 1 |u V }.