Hakem Beitollahi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Hakem Beitollahi has a total of 80 co-authors. Here are the top co-authors by number of shared publications:
- Mahdi Fazeli
School of Information Technology, Halmstad University, Halmstad, Sweden
5 shared publications - Amin Norollah
Iran University of Science and Technology, Iran
4 shared publications - Dyari Mohammed Sharif
Computer Science Department, Faculty of Science, Soran University
3 shared publications - Abdulhady Abas Abdullah
Artificial Intelligence and Innovation Centre, University of Kurdistan Hewler
2 shared publications - Ahmad Patooghy
North Carolina A&T State University
2 shared publications - Ali A. D. Farahani
School of Computer Engineering, Iran University of Science and Technology, Tehran, Iran
2 shared publications
Latest Publications
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (20) , 46734-46747
BDHybrid: A Scalable and Resilient Blockchain-DAG Hybrid for Industrial IoT
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (20) , 46734-46747
BDHybrid: A Scalable and Resilient Blockchain-DAG Hybrid for Industrial IoT
- 18 September 2026
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 82 (15) , 725
SMIT: Static mixed integer training quantization
- 18 September 2026
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 82 (15) , 725
SMIT: Static mixed integer training quantization
- Open Access
- 8 January 2026
- Published by Springer Nature in Research Square
LoRPIA: Low-power Reconfigurable Pallet-Integrated Accelerator for Depthwise Separable Convolutions
Convolutional Neural Networks (CNNs) have achieved remarkable success in tasks such as image classification and recognition, but their high computational and memory demands limit deployment on embedded devices. Depthwise Separable Convolutions (DSCs) address this challenge by reducing the number of parameters and operations while maintaining accuracy, making them an attractive choice for resource-constrained environments. Field-Programmable Gate Arrays (FPGAs) provide an energy-efficient alternative to traditional processors for accelerating CNNs, However, many state-of-the-art designs still suffer from inefficient resource usage and higher power consumption due to the way their hardware is implemented and optimized. In this work, we present a low-power and resource-efficient accelerator for depthwise separable convolutions, implemented entirely in SystemVerilog. The design effectively minimizes hardware resource usage, leading to a notable reduction in power consumption. Despite its compact footprint, the proposed accelerator maintains solid performance, achieving 8.54 FPS for MobileNetV1 and 13.05 FPS for MobileNetV2 on a Zynq XC7Z020 SoC while consuming only 1.25 watts of power.
- 8 January 2026
- Published by Springer Nature in Research Square
Open AccessLoRPIA: Low-power Reconfigurable Pallet-Integrated Accelerator for Depthwise Separable Convolutions
Convolutional Neural Networks (CNNs) have achieved remarkable success in tasks such as image classification and recognition, but their high computational and memory demands limit deployment on embedded devices. Depthwise Separable Convolutions (DSCs) address this challenge by reducing the number of parameters and operations while maintaining accuracy, making them an attractive choice for resource-constrained environments. Field-Programmable Gate Arrays (FPGAs) provide an energy-efficient alternative to traditional processors for accelerating CNNs, However, many state-of-the-art designs still suffer from inefficient resource usage and higher power consumption due to the way their hardware is implemented and optimized. In this work, we present a low-power and resource-efficient accelerator for depthwise separable convolutions, implemented entirely in SystemVerilog. The design effectively minimizes hardware resource usage, leading to a notable reduction in power consumption. Despite its compact footprint, the proposed accelerator maintains solid performance, achieving 8.54 FPS for MobileNetV1 and 13.05 FPS for MobileNetV2 on a Zynq XC7Z020 SoC while consuming only 1.25 watts of power.
- 16 December 2025
- Published by Elsevier in Integration
- Vol. 107
SARPAR: Systolic ARray Pallet-Integrated AcceleratoR for YOLO models on FPGA
- 16 December 2025
- Published by Elsevier in Integration
- Vol. 107
SARPAR: Systolic ARray Pallet-Integrated AcceleratoR for YOLO models on FPGA
- 27 November 2025
- Published by IOP Publishing in Biomedical Physics & Engineering Express
- Vol. 11 (6) , 065057
Metaheuristic-optimized generative adversarial network for enhanced sparse-view low-dose CT reconstruction
Sparse-view low-dose computed tomography (LDCT) imaging poses difficulties in preserving image quality while reducing radiation exposure. Recent research has focused extensively on artificial intelligence (AI) to reduce artifacts in LDCT. This paper presents a unique integration based on a conditional generative adversarial network (CGAN) with metaheuristic optimization to improve the reconstruction of sparse-view computed tomography (CT) images. A Pix2Pix CGAN-based model was integrated with Particle Swarm Optimization (PSO), Differential Evolution (DE), and Cuckoo Search (CS) to improve essential hyperparameters, such as learning rate and beta values. The LDCT-P and LUNA16 datasets were used, producing seven levels of sparse-view CT images (10, 16, 32, 64, 128, 256, and 512 views) for assessment. The findings indicated a substantial improvement in image quality with an increase in the number of view projections. Pix2Pix + PSO demonstrated superior performance, with the Structural Similarity Index metric (SSIM) rising from 0.900 (10 views) to 0.972 (512 views) for abdominal CT and from 0.801 to 0.971 for lung CT, respectively. The results underscore the capability of the Pix2Pix model integrated with metaheuristic optimization to enhance sparse-view CT reconstruction. This method adeptly reconciles computing economy with image integrity, enabling improved LDCT imaging applications in clinical settings.
- 27 November 2025
- Published by IOP Publishing in Biomedical Physics & Engineering Express
- Vol. 11 (6) , 065057
Metaheuristic-optimized generative adversarial network for enhanced sparse-view low-dose CT reconstruction
Sparse-view low-dose computed tomography (LDCT) imaging poses difficulties in preserving image quality while reducing radiation exposure. Recent research has focused extensively on artificial intelligence (AI) to reduce artifacts in LDCT. This paper presents a unique integration based on a conditional generative adversarial network (CGAN) with metaheuristic optimization to improve the reconstruction of sparse-view computed tomography (CT) images. A Pix2Pix CGAN-based model was integrated with Particle Swarm Optimization (PSO), Differential Evolution (DE), and Cuckoo Search (CS) to improve essential hyperparameters, such as learning rate and beta values. The LDCT-P and LUNA16 datasets were used, producing seven levels of sparse-view CT images (10, 16, 32, 64, 128, 256, and 512 views) for assessment. The findings indicated a substantial improvement in image quality with an increase in the number of view projections. Pix2Pix + PSO demonstrated superior performance, with the Structural Similarity Index metric (SSIM) rising from 0.900 (10 views) to 0.972 (512 views) for abdominal CT and from 0.801 to 0.971 for lung CT, respectively. The results underscore the capability of the Pix2Pix model integrated with metaheuristic optimization to enhance sparse-view CT reconstruction. This method adeptly reconciles computing economy with image integrity, enabling improved LDCT imaging applications in clinical settings.
- 11 September 2025
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 81 (14) , 1-45
Multi-objective feature selection of radiomics and deep learning features for breast cancer subtype detection
- 11 September 2025
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 81 (14) , 1-45
Multi-objective feature selection of radiomics and deep learning features for breast cancer subtype detection
- 9 September 2025
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 81 (14) , 1-25
TestLock: a testability logic locking method against machine learning-based oracle-less attacks
- 9 September 2025
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 81 (14) , 1-25
TestLock: a testability logic locking method against machine learning-based oracle-less attacks
- Open Access
- 29 August 2025
- Published by Springer Nature in Research Square
DLPack: A DSP-Based Low-Bitwidth Packing Architecture for Efficient 2-Bit CNN Inference on FPGA-based Edge Devices
Convolutional neural networks (CNNs) have become a fundamental component of modern deep learning, particularly in intelligent edge systems. However, deploying CNNs on such platforms presents challenges due to stringent constraints on power and computational resources. Field-programmable gate arrays (FPGAs), known for their reconfigurability and parallelism, offer a promising solution—yet are often inefficiently utilized for low-bitwidth inference. In this work, we present DLPack, a lightweight FPGA accelerator specifically designed for 2-bit CNN inference. DLPack introduces a structured packing technique that combines multiple low-precision multiply-accumulate (MAC) operations within a single DSP block, significantly enhancing processing density and resource efficiency. The architecture further incorporates a tile-wise dataflow strategy and a streamlined control mechanism to reduce latency and power consumption. Implemented on a Xilinx UltraScale+ FPGA, DLPack achieves up to 50% reduction in DSP usage, 83% lower power consumption, and around 99% improvement in inference latency compared to existing approaches. These results demonstrate the effectiveness of DLPack in enabling scalable, energy-efficient CNN inference on edge devices with limited computational budgets.
- 29 August 2025
- Published by Springer Nature in Research Square
Open AccessDLPack: A DSP-Based Low-Bitwidth Packing Architecture for Efficient 2-Bit CNN Inference on FPGA-based Edge Devices
Convolutional neural networks (CNNs) have become a fundamental component of modern deep learning, particularly in intelligent edge systems. However, deploying CNNs on such platforms presents challenges due to stringent constraints on power and computational resources. Field-programmable gate arrays (FPGAs), known for their reconfigurability and parallelism, offer a promising solution—yet are often inefficiently utilized for low-bitwidth inference. In this work, we present DLPack, a lightweight FPGA accelerator specifically designed for 2-bit CNN inference. DLPack introduces a structured packing technique that combines multiple low-precision multiply-accumulate (MAC) operations within a single DSP block, significantly enhancing processing density and resource efficiency. The architecture further incorporates a tile-wise dataflow strategy and a streamlined control mechanism to reduce latency and power consumption. Implemented on a Xilinx UltraScale+ FPGA, DLPack achieves up to 50% reduction in DSP usage, 83% lower power consumption, and around 99% improvement in inference latency compared to existing approaches. These results demonstrate the effectiveness of DLPack in enabling scalable, energy-efficient CNN inference on edge devices with limited computational budgets.
- 15 July 2025
- Published by Elsevier in Journal of Systems Architecture
- Vol. 167
MCED-H: An efficient scheduling policy for energy harvesting mixed-criticality real-time systems
- 15 July 2025
- Published by Elsevier in Journal of Systems Architecture
- Vol. 167
MCED-H: An efficient scheduling policy for energy harvesting mixed-criticality real-time systems
- Open Access
- 1 July 2025
- Published by Koya University in ARO-THE SCIENTIFIC JOURNAL OF KOYA UNIVERSITY
- Vol. 13 (2) , 349-358
An Ensemble-based Model for Sentiment Analysis of Kurdish Tweets
Thousands of comments are generated daily on social media in the Kurdistan Region. Sentiment analysis (SA) of these comments is valuable for organizations. The Kurdish language has three main dialects: Sorani (Central), Northern, and Southern. This study focuses on Sorani SA, where existing methods have limited accuracy. The proposed ensemble combines diverse models to improve sentiment classification. Preprocessing and word embedding using Roberta is the first phase of the method. The second phase consists of four proposed models, namely K-nearest neighbor, support vector machine, multilayer perceptron long short-term memory (LSTM), and bidirectional-LSTM (Bi-LSTM), which are used as classifiers. Finally, the ensemble weighted averaging technique is utilized to generate the final classification. To evaluate the performance of the proposed model, a dataset including 24211 unbalanced Soran tweets is first used, and after balancing, the dataset is used. The Bi-LSTM model attained an accuracy of 89.87% on the balanced dataset, and the proposed ensemble method increased the accuracy to 91.76%, which is better than the established state-of-the-art methods of Kurdish SA.
- 1 July 2025
- Published by Koya University in ARO-THE SCIENTIFIC JOURNAL OF KOYA UNIVERSITY
- Vol. 13 (2) , 349-358
Open AccessAn Ensemble-based Model for Sentiment Analysis of Kurdish Tweets
Thousands of comments are generated daily on social media in the Kurdistan Region. Sentiment analysis (SA) of these comments is valuable for organizations. The Kurdish language has three main dialects: Sorani (Central), Northern, and Southern. This study focuses on Sorani SA, where existing methods have limited accuracy. The proposed ensemble combines diverse models to improve sentiment classification. Preprocessing and word embedding using Roberta is the first phase of the method. The second phase consists of four proposed models, namely K-nearest neighbor, support vector machine, multilayer perceptron long short-term memory (LSTM), and bidirectional-LSTM (Bi-LSTM), which are used as classifiers. Finally, the ensemble weighted averaging technique is utilized to generate the final classification. To evaluate the performance of the proposed model, a dataset including 24211 unbalanced Soran tweets is first used, and after balancing, the dataset is used. The Bi-LSTM model attained an accuracy of 89.87% on the balanced dataset, and the proposed ensemble method increased the accuracy to 91.76%, which is better than the established state-of-the-art methods of Kurdish SA.