Iterative Methods via Locally Evolving Set Process

Authors: Baojian Zhou, Yifan Sun, Reza Babanezhad Harikandeh, Xingzhi Guo, Deqing Yang, Yanghua Xiao

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical results confirm the efficiency of this novel framework and show up to a hundredfold speedup over corresponding standard solvers on real-world graphs. and 6 Experiments We conduct experiments over 17 graphs to solve (3) and explore the local clustering task.
Researcher Affiliation Collaboration 1 the School of Data Science, Fudan University, 2 Shanghai Key Laboratory of Data Science, School of Computer Science, Fudan University 3 Department of Computer Science, Stony Brook University, 4 Samsung SAIT AI Lab.
Pseudocode Yes Algo. 1 PUSH(u, α, p, z), Algo. 2 APPR(α, ϵ, s, G) via FIFO Queue, Algo. 3 LOCSOR(α, ϵ, s, G, ω) via FIFO Queue, Algo. 4 LOCGD(α, ϵ, s, G) via FIFO Queue
Open Source Code Yes Our code is available at https: // github. com/ baojian/ Local CH .
Open Datasets Yes Following Leskovec et al. [34], we treat all 17 graphs as undirected with unit weights. and Table 3: Dataset Statistics
Dataset Splits No The paper does not explicitly mention training, validation, and test splits for the datasets in the context of model training. The experiments involve solving linear systems on graphs rather than typical machine learning model training with data splits.
Hardware Specification Yes For our experiment, we used a server powered by an Intel(R) Xeon(R) Gold 5218R CPU, which features 40 cores (80 threads). The system is equipped with 256 GB of RAM.
Software Dependencies Yes All methods are implemented in Python 3.10 with the numba library [33].
Experiment Setup Yes To compare local solvers to their standard counterparts, we set α = 0.1, randomly select 50 nodes from each graph to serve as es in (3)... The range of ϵ is ϵ [ α 2(1+α)ds , 10 4/n]. and For the local ISTA method [13], the precision parameter is set to ˆϵ = 0.5 for all experiments. ... For LOCSOR, the parameter ω is calculated as 2(1 + α)/(1 + α)2.