Irfan Ullah
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Irfan Ullah has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Young-Koo Lee
Department of Artificial Intelligence, Kyung Hee University, Yongin-si, Republic of Korea
4 shared publications - Aftab Alam
College of Science and Engineering, Hamad bin Khalifa University, Doha, Qatar
2 shared publications Duong Thi Thu Van
Department of Computer Science and Engineering, Kyung Hee University, Global Campus, Yongin, South Korea
1 shared publication- Md Arif Rahman
Department of Computer Engineering, Kyung Hee University, Yongin-si, Republic of Korea
1 shared publication - Muhammad Numan Khan
Department of Computer Science and Engineering, Kyung-Hee University (Global Campus), Yongin-si, Gyeonggi-do, South Korea
1 shared publication - Syed Jalaluddin Hashmi
Department of Artificial Intelligence, Kyung Hee University, Yongin-si, Republic of Korea
1 shared publication
Latest Publications
- Open Access
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1538-1548
AdaptSteer: Workload-Adaptive Optimizer Steering via Execution-Derived Preference Supervision
Database query optimizers often select suboptimal execution plans, and effective steering requires identifying configurations that improve query and workload performance without modifying DBMS internals. Existing external steering systems can reason over multiple optimizer configurations, but many rely on online feedback, exploration, or engine-specific steering mechanisms. More recent LLM-based SQL steering approaches avoid deeper optimizer integration, but often depend on narrow supervision and general-purpose embeddings that are costly, not explicitly aligned with execution behavior, and difficult to adapt under workload shifts. We present AdaptSteer, a novel workload-adaptive optimizer-steering framework that learns from execution-derived preference supervision. AdaptSteer converts latencies measured across diverse query--configuration pairs into preference-based contrastive training signals for a compact SQL encoder, without invasive DBMS modifications or external APIs. The resulting representation supports two lightweight decision heads: a calibrated classifier for safety-first binary steering and a pairwise Condorcet ranker that decomposes K-way configuration selection into binary execution-preference comparisons. We further apply Reptile-based meta-learning to the trained encoder, enabling few-shot adaptation to unseen workloads. Evaluations on the JOB and CEB benchmarks show that AdaptSteer reduces total workload latency by 70.1% relative to the PostgreSQL default and achieves 4.4% lower total workload latency than LLMSteer under the same binary protocol, without relying on external API calls. Beyond binary steering, AdaptSteer improves multi-action selection through pairwise preference ranking across the evaluated candidate sets and benefits from multi-configuration supervision across broader action spaces. Under CEB?JOB adaptation, AdaptSteer's meta-learned variant improves AUROC by 4.2 percentage points over the contrastive-only baseline with only 20 target-workload adaptation queries. Together, AdaptSteer unifies execution-derived preference supervision, binary and multi-action steering, and few-shot adaptation in a compact, deployable optimizer-steering framework.
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1538-1548
Open AccessAdaptSteer: Workload-Adaptive Optimizer Steering via Execution-Derived Preference Supervision
Database query optimizers often select suboptimal execution plans, and effective steering requires identifying configurations that improve query and workload performance without modifying DBMS internals. Existing external steering systems can reason over multiple optimizer configurations, but many rely on online feedback, exploration, or engine-specific steering mechanisms. More recent LLM-based SQL steering approaches avoid deeper optimizer integration, but often depend on narrow supervision and general-purpose embeddings that are costly, not explicitly aligned with execution behavior, and difficult to adapt under workload shifts. We present AdaptSteer, a novel workload-adaptive optimizer-steering framework that learns from execution-derived preference supervision. AdaptSteer converts latencies measured across diverse query--configuration pairs into preference-based contrastive training signals for a compact SQL encoder, without invasive DBMS modifications or external APIs. The resulting representation supports two lightweight decision heads: a calibrated classifier for safety-first binary steering and a pairwise Condorcet ranker that decomposes K-way configuration selection into binary execution-preference comparisons. We further apply Reptile-based meta-learning to the trained encoder, enabling few-shot adaptation to unseen workloads. Evaluations on the JOB and CEB benchmarks show that AdaptSteer reduces total workload latency by 70.1% relative to the PostgreSQL default and achieves 4.4% lower total workload latency than LLMSteer under the same binary protocol, without relying on external API calls. Beyond binary steering, AdaptSteer improves multi-action selection through pairwise preference ranking across the evaluated candidate sets and benefits from multi-configuration supervision across broader action spaces. Under CEB?JOB adaptation, AdaptSteer's meta-learned variant improves AUROC by 4.2 percentage points over the contrastive-only baseline with only 20 target-workload adaptation queries. Together, AdaptSteer unifies execution-derived preference supervision, binary and multi-action steering, and few-shot adaptation in a compact, deployable optimizer-steering framework.
- Open Access
- 7 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 27550-27569
MQ-GNN: A Multi-Queue Pipelined Architecture for Scalable and Efficient GNN Training
Graph Neural Networks (GNNs) are effective tools for graph data. Still, their scalability is limited by inefficient processing of training stages such as mini-batch generation, data transfer bottlenecks, inefficient inter-GPU gradient sharing, and model updates. Existing GNN training paradigms fail to effectively overlap these stages, leading to poor utilization of computing resources. Communication delays, data transfer, and computation stalls are common problems with large datasets, especially in multi-GPU setups where data transfer and model updates dominate the training time. In this paper, we propose MQ-GNN, a novel pipelined framework to address these challenges by interleaving GNN training stages and leveraging a multi-queueing mechanism to maximize resource utilization. It incorporates determining optimal queue size that balances computational efficiency and memory limitations to ensure GPU utilization. Furthermore, MQ-GNN also introduces Ready-to-Update Asynchronous Consistent Model (RaCoM) that asynchronously shares and accumulates gradients and updates models independently on each GPU, with periodic synchronization to minimize staleness and ensure model consistency. To further speed up the GPU data transfers, MQ-GNN utilizes global neighbor sampling by caching frequently accessed node features. Experiments on four standard datasets and 10 baseline models, demonstrate up to 4.6× speedups in training times and a 30% improvement in GPU utilization, with competitive evaluation metric. We also perform detailed analysis on node-wise and layer-wise approaches in a multi-GPU environment, comparing baseline and MQ-GNN models. By addressing these limitations, MQ-GNN offers a scalable and efficient solution for large-scale graph learning, advancing the development of high-performance GNN training architectures. Code is available at MQ-GNN.
- 7 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 27550-27569
Open AccessMQ-GNN: A Multi-Queue Pipelined Architecture for Scalable and Efficient GNN Training
Graph Neural Networks (GNNs) are effective tools for graph data. Still, their scalability is limited by inefficient processing of training stages such as mini-batch generation, data transfer bottlenecks, inefficient inter-GPU gradient sharing, and model updates. Existing GNN training paradigms fail to effectively overlap these stages, leading to poor utilization of computing resources. Communication delays, data transfer, and computation stalls are common problems with large datasets, especially in multi-GPU setups where data transfer and model updates dominate the training time. In this paper, we propose MQ-GNN, a novel pipelined framework to address these challenges by interleaving GNN training stages and leveraging a multi-queueing mechanism to maximize resource utilization. It incorporates determining optimal queue size that balances computational efficiency and memory limitations to ensure GPU utilization. Furthermore, MQ-GNN also introduces Ready-to-Update Asynchronous Consistent Model (RaCoM) that asynchronously shares and accumulates gradients and updates models independently on each GPU, with periodic synchronization to minimize staleness and ensure model consistency. To further speed up the GPU data transfers, MQ-GNN utilizes global neighbor sampling by caching frequently accessed node features. Experiments on four standard datasets and 10 baseline models, demonstrate up to 4.6× speedups in training times and a 30% improvement in GPU utilization, with competitive evaluation metric. We also perform detailed analysis on node-wise and layer-wise approaches in a multi-GPU environment, comparing baseline and MQ-GNN models. By addressing these limitations, MQ-GNN offers a scalable and efficient solution for large-scale graph learning, advancing the development of high-performance GNN training architectures. Code is available at MQ-GNN.
- Open Access
- 31 January 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 21684-21700
GDLL: A Scalable and Share Nothing Architecture Based Distributed Graph Neural Networks Framework
Deep learning has recently been shown to be effective in uncovering hidden patterns in non-Euclidean space, where data is represented as graphs with complex object relationships and interdependencies. Because of the implicit data dependence in the big graphs with millions of nodes and billions of edges, it is hard for industrial communities to exploit these methods to address real-world challenges at scale. The skewness property of big graphs, distributed file system performance penalty on small k-hop neighborhood subgraphs, and varying size of subgraph makes Graph Neural Networks (GNNs) training further challenging in a distributed environment using parameter servers. To address such issues, we propose a scalable, layered, fault-tolerance, and in-memory distributed computing-based graph neural network framework called Graph Distributed Learning Library (GDLL). The base layer utilizes an optimized distributed file system and a scalable graph data store to reduce the performance penalty. The second layer provides distributed graph processing using in-memory graph programming models while optimizing and hiding the underlying complexity of information complete subgraph computation. In the third layer, GNN modules are deployed on top of the first two layers for efficient distributed training using parameter servers. Finally, we evaluate and compare GDLL with the state-of-the-art solutions and outperform it significantly in terms of efficiency while maintaining similar GNN convergence.
- 31 January 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 21684-21700
Open AccessGDLL: A Scalable and Share Nothing Architecture Based Distributed Graph Neural Networks Framework
Deep learning has recently been shown to be effective in uncovering hidden patterns in non-Euclidean space, where data is represented as graphs with complex object relationships and interdependencies. Because of the implicit data dependence in the big graphs with millions of nodes and billions of edges, it is hard for industrial communities to exploit these methods to address real-world challenges at scale. The skewness property of big graphs, distributed file system performance penalty on small k-hop neighborhood subgraphs, and varying size of subgraph makes Graph Neural Networks (GNNs) training further challenging in a distributed environment using parameter servers. To address such issues, we propose a scalable, layered, fault-tolerance, and in-memory distributed computing-based graph neural network framework called Graph Distributed Learning Library (GDLL). The base layer utilizes an optimized distributed file system and a scalable graph data store to reduce the performance penalty. The second layer provides distributed graph processing using in-memory graph programming models while optimizing and hiding the underlying complexity of information complete subgraph computation. In the third layer, GNN modules are deployed on top of the first two layers for efficient distributed training using parameter servers. Finally, we evaluate and compare GDLL with the state-of-the-art solutions and outperform it significantly in terms of efficiency while maintaining similar GNN convergence.
- Open Access
- 17 August 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 152377-152422
Video Big Data Analytics in the Cloud: A Reference Architecture, Survey, Opportunities, and Open Research Issues
The proliferation of multimedia devices over the Internet of Things (IoT) generates an unprecedented amount of data. Consequently, the world has stepped into the era of big data. Recently, on the rise of distributed computing technologies, video big data analytics in the cloud has attracted the attention of researchers and practitioners. The current technology and market trends demand an efficient framework for video big data analytics. However, the current work is too limited to provide a complete survey of recent research work on video big data analytics in the cloud, including the management and analysis of a large amount of video data, the challenges, opportunities, and promising research directions. To serve this purpose, we present this study, which conducts a broad overview of the state-of-the-art literature on video big data analytics in the cloud. It also aims to bridge the gap among large-scale video analytics challenges, big data solutions, and cloud computing. In this study, we clarify the basic nomenclatures that govern the video analytics domain and the characteristics of video big data while establishing its relationship with cloud computing.We propose a service-oriented layered reference architecture for intelligent video big data analytics in the cloud. Then, a comprehensive and keen review has been conducted to examine cutting-edge research trends in video big data analytics. Finally, we identify and articulate several open research issues and challenges, which have been raised by the deployment of big data technologies in the cloud for video big data analytics. To the best of our knowledge, this is the first study that presents the generalized view of the video big data analytics in the cloud. This paper provides the research studies and technologies advancing the video analyses in the era of big data and cloud computing.
- 17 August 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 152377-152422
Open AccessVideo Big Data Analytics in the Cloud: A Reference Architecture, Survey, Opportunities, and Open Research Issues
The proliferation of multimedia devices over the Internet of Things (IoT) generates an unprecedented amount of data. Consequently, the world has stepped into the era of big data. Recently, on the rise of distributed computing technologies, video big data analytics in the cloud has attracted the attention of researchers and practitioners. The current technology and market trends demand an efficient framework for video big data analytics. However, the current work is too limited to provide a complete survey of recent research work on video big data analytics in the cloud, including the management and analysis of a large amount of video data, the challenges, opportunities, and promising research directions. To serve this purpose, we present this study, which conducts a broad overview of the state-of-the-art literature on video big data analytics in the cloud. It also aims to bridge the gap among large-scale video analytics challenges, big data solutions, and cloud computing. In this study, we clarify the basic nomenclatures that govern the video analytics domain and the characteristics of video big data while establishing its relationship with cloud computing.We propose a service-oriented layered reference architecture for intelligent video big data analytics in the cloud. Then, a comprehensive and keen review has been conducted to examine cutting-edge research trends in video big data analytics. Finally, we identify and articulate several open research issues and challenges, which have been raised by the deployment of big data technologies in the cloud for video big data analytics. To the best of our knowledge, this is the first study that presents the generalized view of the video big data analytics in the cloud. This paper provides the research studies and technologies advancing the video analyses in the era of big data and cloud computing.