Clique Number Estimation via Differentiable Functions of Adjacency Matrix Permutations

Authors: Indradyumna Roy, Eeshaan Jain, Soumen Chakrabarti, Abir De

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on eight datasets show the superior accuracy of our approach. The code is available on Git Hub. ... 4 EXPERIMENTS We report on extensive experiments using eight datasets, comparing the performance of MXNET with other methods. We also instrument different components of MXNET to understand their impact.
Researcher Affiliation Academia Indradyumna Roy 1, Eeshaan Jain 2, Soumen Chakrabarti1, Abir De1 1IIT Bombay, 2EPFL EMAIL, EMAIL
Pseudocode Yes Algorithm 1 MSS(B) # B is binary
Open Source Code Yes The code is available on Git Hub.
Open Datasets Yes Datasets We conduct experiments on eight datasets, comprising five real-world and three synthetic datasets. Real-world datasets include (1) IMDB-BINARY (IMDB), (2) Enzymes and modular products of graph pairs from (3) PTC-MM-m, (4) AIDS, (5) Mutagenicity (MUTAG-m) datasets. We also generate three synthetic datasets from (6) DSJC, (7) Brockington (Brock), and (8) RB. ... We use modular graph products for three datasets, viz., AIDS, MUTAG, PTC-MM. We call them AIDS-m, MUTAG-m and PTC-MM-m respectively. Additional details are in in Appendix E. ... sourced from the TUDatasets repository (Morris et al., 2020): (3) PTC-MM, (4) AIDS and (5) Mutagenicity.
Dataset Splits Yes We split each dataset D = {Gi, ω(Gi) | i [I]} into 60% training, 20% eval, and 20% test folds.
Hardware Specification Yes The training of our models and the baselines was performed on servers containing AMD EPYC 7642 48-Core Processors at 2.30GHz CPUs, and Nvidia RTX A6000 GPUs.
Software Dependencies Yes We implement our models using Python 3.10 and Py Torch 2.3.0.
Experiment Setup Yes All models are trained using the Adam optimizer, with a learning rate of 10 3, and weight decay 5 10 4. ... For the design of LComposite, we set δ = 1, and search for γ, λ in {0.25, 0.75}, and {0.1, 1} respectively. Hence, the total search space for hyperparameters consists of 8 combinations ({τ} {γ} {λ}), and we select the best model out of the 8 hyperparameter configurations. ... For the early stopping criteria based on the validation MSE, we use a patience parameter as 200 epochs.