arXiv subjects

Muhammed Fatih Balin

Publications and source records attributed to Muhammed Fatih Balin.

5 recordsLinked to original sources

arXiv2025

BLISS: Bandit Layer Importance Sampling Strategy for Efficient Training of Graph Neural Networks

Graph Neural Networks (GNNs) are powerful tools for learning from graph-structured data, but their application to large graphs is hindered by computational costs. The need to process every neighbor for each node creates memory and computational bottlenecks. To address this, we introduce BLISS, a Bandit Layer Importance Sampling Strategy. It uses multi-armed bandits to dynamically select the most informative nodes at each layer, balancing exploration and exploitation to ensure comprehensive graph coverage. Unlike existing static sampling methods, BLISS adapts to evolving node importance, leading to more informed node selection and improved performance. It demonstrates versatility by integrating with both Graph Convolutional Networks (GCNs) and Graph Attention Networks (GATs), adapting its selection policy to their specific aggregation mechanisms. Experiments show that BLISS maintains or exceeds the accuracy of full-batch training.

cs.LG↗

arXiv2024

Cooperative Minibatching in Graph Neural Networks

Training large scale Graph Neural Networks (GNNs) requires significant computational resources, and the process is highly data-intensive. One of the most effective ways to reduce resource requirements is minibatch training coupled with graph sampling. GNNs have the unique property that items in a minibatch have overlapping data. However, the commonly implemented Independent Minibatching approach assigns each Processing Element (PE, i.e., cores and/or GPUs) its own minibatch to process, leading to duplicated computations and input data access across PEs. This amplifies the Neighborhood Explosion Phenomenon (NEP), which is the main bottleneck limiting scaling. To reduce the effects of NEP in the multi-PE setting, we propose a new approach called Cooperative Minibatching. Our approach capitalizes on the fact that the size of the sampled subgraph is a concave function of the batch size, leading to significant reductions in the amount of work as batch sizes increase. Hence, it is favorable for processors equipped with a fast interconnect to work on a large minibatch together as a single larger processor, instead of working on separate smaller minibatches, even though global batch size is identical. We also show how to take advantage of the same phenomenon in serial execution by generating dependent consecutive minibatches. Our experimental evaluations show up to 4x bandwidth savings for fetching vertex embeddings, by simply increasing this dependency without harming model convergence. Combining our proposed approaches, we achieve up to 64% speedup over Independent Minibatching on single-node multi-GPU systems, using same resources.

cs.LG↗

arXiv2024

A Scalable and Effective Alternative to Graph Transformers

Graph Neural Networks (GNNs) have shown impressive performance in graph representation learning, but they face challenges in capturing long-range dependencies due to their limited expressive power. To address this, Graph Transformers (GTs) were introduced, utilizing self-attention mechanism to effectively model pairwise node relationships. Despite their advantages, GTs suffer from quadratic complexity w.r.t. the number of nodes in the graph, hindering their applicability to large graphs. In this work, we present Graph-Enhanced Contextual Operator (GECO), a scalable and effective alternative to GTs that leverages neighborhood propagation and global convolutions to effectively capture local and global dependencies in quasilinear time. Our study on synthetic datasets reveals that GECO reaches 169x speedup on a graph with 2M nodes w.r.t. optimized attention. Further evaluations on diverse range of benchmarks showcase that GECO scales to large graphs where traditional GTs often face memory and time limitations. Notably, GECO consistently achieves comparable or superior quality compared to baselines, improving the SOTA up to 4.5%, and offering a scalable and effective solution for large-scale graph learning.

cs.LG↗

arXiv2023

SGORP: A Subgradient-based Method for d-Dimensional Rectilinear Partitioning

Partitioning for load balancing is a crucial first step to parallelize any type of computation. In this work, we propose SGORP, a new spatial partitioning method based on Subgradient Optimization, to solve the $d$-dimensional Rectilinear Partitioning Problem (RPP). Our proposed method allows the use of customizable objective functions as well as some user-specific constraints, such as symmetric partitioning on selected dimensions. Extensive experimental evaluation using over 600 test matrices shows that our algorithm achieves favorable performance against the state-of-the-art RPP and Symmetric RPP algorithms. Additionally, we show the effectiveness of our algorithm to do application-specific load balancing using two applications as motivation: Triangle Counting and Sparse Matrix Multiplication (SpGEMM), where we model their load-balancing problems as $3$-dimensional RPPs.

cs.DS↗

arXiv2020

On Symmetric Rectilinear Matrix Partitioning

Even distribution of irregular workload to processing units is crucial for efficient parallelization in many applications. In this work, we are concerned with a spatial partitioning called rectilinear partitioning (also known as generalized block distribution) of sparse matrices. More specifically, in this work, we address the problem of symmetric rectilinear partitioning of a square matrix. By symmetric, we mean the rows and columns of the matrix are identically partitioned yielding a tiling where the diagonal tiles (blocks) will be squares. We first show that the optimal solution to this problem is NP-hard, and we propose four heuristics to solve two different variants of this problem. We present a thorough analysis of the computational complexities of those proposed heuristics. To make the proposed techniques more applicable in real life application scenarios, we further reduce their computational complexities by utilizing effective sparsification strategies together with an efficient sparse prefix-sum data structure. We experimentally show the proposed algorithms are efficient and effective on more than six hundred test matrices. With sparsification, our methods take less than 3 seconds in the Twitter graph on a modern 24 core system and output a solution whose load imbalance is no worse than 1%.

cs.DS↗