Overlap-Robust Decision Boundary Learning for Within-Network Classification

Authors: Sharad Nandanwar, M. N. Murty

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through extensive comparative study on different real-world datasets, we found that our method improves over the state-of-the-art approaches.
Researcher Affiliation Academia Sharad Nandanwar Indian Institute of Science Bangalore,India sharadnandanwar@csa.iisc.ernet.in M. N. Murty Indian Institute of Science Bangalore,India mnm@csa.iisc.ernet.in
Pseudocode Yes Algorithm 1 Training Re SLMin using SGD.
Open Source Code Yes We make our code and datasets publicly available at https://github.com/sharadnandanwar/Re SLMin.
Open Datasets Yes We make our code and datasets publicly available at https://github.com/sharadnandanwar/Re SLMin. Datasets used: Pub Med, Co RA, Amazon, Wikipedia. Table 1 summarizes some of the statistics of datasets used. For studying the robustness of the proposed approach with increase in class overlap, we synthesize a network of 10000 nodes using LFR benchmark (Lancichinetti, Fortunato, and Radicchi 2008).
Dataset Splits Yes For each experiment, we perform 20 runs using different realizations of the train and test sets, having 10% and 90% of nodes respectively. To determine the optimal values of regularization parameter, we perform 10-fold cross-validation with the given set of training nodes.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies No The paper does not provide specific software dependencies (e.g., library names with version numbers like Python 3.8, PyTorch 1.9) needed to replicate the experiment.
Experiment Setup Yes To determine the optimal values of regularization parameter, we perform 10-fold cross-validation with the given set of training nodes. The parameter values are chosen using grid search technique where λ and μ both were in the range {2 5, 2 6, . . . , 2 15}. Also, we set the value of parameter α, responsible for neighborhood effect, as 0.5.