Kutay Tasci

PhD Student

Kutay Tasci is a Ph.D. student in Computer and Data Sciences at Case Western Reserve University, where he conducts research on high-performance computing systems for AI/ML workloads, with a focus on graph neural networks, geometric deep learning, and sparse/irregular computations.

Teaching Information

High-performance computing, parallel and distributed systems, machine learning systems, graph neural networks, and large-scale AI.

Education

Ph.D. in Computer and Data Sciences, Case Western Reserve University, 2024–2028 (expected)
M.S. in Computer Engineering, Bilkent University, 2021–2024
B.S. in Computer Engineering, Hacettepe University, 2017–2021

Awards and Honors

Swanger Graduate Fellowship, 2024
5G and Beyond Joint Graduate Support Program (ICTA), 2022
Bilkent University Comprehensive Scholarship, 2021
METU IEEE Hackathon – 1st Place, 2020

Kutay Tasci is a Ph.D. student in the Department of Computer and Data Sciences at Case Western Reserve University. His research focuses on the design of efficient parallel and distributed algorithms for emerging AI and machine learning workloads, particularly those involving sparse and irregular computation patterns. His current interests span graph neural networks (GNNs), geometric deep learning, sparse attention mechanisms, and high-performance computing (HPC) systems. Prior to his doctoral studies, he received his B.S. in Computer Engineering from Hacettepe University and his M.S. in Computer Engineering from Bilkent University. During his graduate research, he worked on large-scale sparse kernels, distributed graph neural network training, and communication-efficient algorithms for HPC systems. His previous work explored sparse matrix operations, hypergraph-based partitioning methods, and communication bottlenecks in distributed-memory architectures, contributing to projects supported by national and international research initiatives including TUBITAK and EuroHPC. More broadly, his work lies at the intersection of machine learning systems, scientific computing, and performance optimization. He is particularly interested in developing scalable methods that improve the efficiency of modern AI models on heterogeneous computing architectures, with the long-term goal of enabling larger, faster, and more practical machine learning systems for both industrial and scientific applications.

My research focuses on developing efficient parallel and distributed methods for sparse and irregular AI/ML workloads, with particular emphasis on graph neural networks, geometric deep learning, sparse attention, and high-performance computing systems.

Current Projects

Efficient Geometric Graph Neural Networks

My thesis work focuses on improving the efficiency and scalability of geometric graph neural networks (GNNs) for scientific machine learning applications. This research studies the computational and systems challenges of geometric message passing, with emphasis on designing methods that reduce redundant computation, improve memory efficiency, and better utilize modern GPU architectures. The broader goal is to enable faster and more scalable training and inference for geometric deep learning models used in domains such as molecular modeling, materials science, and physical simulation.

Related Publications:

Sparse Attention and Long-Context Transformer Optimization

This research focuses on improving the efficiency of block-sparse attention implementations for long-context transformer models. In particular, I study how token and block reordering techniques can transform irregular sparse attention patterns into hardware-friendly layouts that reduce the number of active blocks and improve GPU execution efficiency. The broader goal is to make sparse transformer inference more practical and scalable for large language models and other long-sequence learning workloads.

Alternative Decomposed Message Passing for Efficient Geometric GNNs

2026

Kutay Tasci , Sanmukh Kuppannagari

IEEE IPDPS 2026 Workshops (GrAPL)

This paper proposes an alternative decomposed message-passing framework for improving the efficiency of geometric graph neural networks. Instead of relying on concatenation-based message generation, the method decomposes node-, edge-, and angle-level transformations into reusable components, reducing redundant computation and memory overhead while preserving algebraic equivalence. The framework is designed for efficient GPU execution and serves as a drop-in replacement for representative architectures such as EGNN and CHGNet. Experimental results show up to 2x training speedup and 60% end-to-end memory reduction with no loss in accuracy across diverse geometric learning workloads.

Artificial Intelligence HPC

Transforming temporal-dynamic graphs into time-series data for solving event detection problems

2023

Kutay Tasci , Fuat Akal

Turkish Journal of Electrical Engineering and Computer Sciences

This paper proposes a workflow for detecting important events in temporal-dynamic graphs by transforming graph snapshots into multivariate time-series data. The method first generates graph-level embeddings for each time step using temporal graph representation learning, and then applies unsupervised time-series anomaly detection models to identify abnormal events. The approach was evaluated on multiple real-world social media datasets and showed competitive or improved performance compared to prior event detection methods. The work demonstrates that graph embeddings can serve as an effective bridge between dynamic graph analysis and time-series anomaly detection.

Artificial Intelligence

DOI