Muhammad Alfian Amrizal
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Muhammad Alfian Amrizal has a total of 35 co-authors. Here are the top co-authors by number of shared publications:
- Hiroyuki Takizawa
Tohoku University, Sendai, Japan
7 shared publications - Reza Pulungan
Universitas Gadjah Mada, Yogyakarta, Indonesia
6 shared publications - Santana Yuda Pradata
Universitas Gadjah Mada, Yogyakarta, Indonesia
5 shared publications - Kadek Gemilang Santiyuda
Institut Bisnis dan Teknologi Indonesia, Bali, Indonesia
3 shared publications Raka Satya Prasasta
Faculty of Industrial Technology Universitas Ahmad Dahlan 55191 Yogyakarta Indonesia
2 shared publications- Raka Satya Prasasta
Universitas Ahmad Dahlan, Yogyakarta, Indonesia
2 shared publications
Latest Publications
- Open Access
- 4 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1399-1414
Fully Homomorphic Encryption Inference of Neural Networks Using CKKS-TFHE Scheme Switching and Accelerated Linear Layers
Fully homomorphic encryption (FHE) allows neural network inference to be performed directly on encrypted data and models, preserving end-to-end privacy. One of the most widely used FHE schemes, CKKS, supports approximate arithmetic over real numbers and is well-suited for computing polynomial operations. However, CKKS does not natively support conditional branching, making it unsuitable for non-linear functions such as ReLU, which behaves differently depending on whether the input is negative or nonnegative. To address this, ReLU is typically approximated using polynomials. Unfortunately, this approach introduces two key drawbacks: (1) the approximation is only valid within a certain input range, and (2) identifying this range typically requires prior exposure of cleartext data to perform profiling, thus partially compromising privacy. This requirement is problematic in practice, as small profiling errors can lead to substantial accuracy degradation. On the other hand, the TFHE scheme supports binary gate-level computation, enabling precise implementation of conditional operations such as ReLU. Yet TFHE lacks the SIMD capabilities of CKKS, leading to significant computational latency. This paper explores the use of scheme switching for encrypted neural network inference: leveraging CKKS for polynomial-compatible layers and switching to TFHE for layers that require conditional logic, such as ReLU. We demonstrate that this hybrid approach substantially improves the numerical fidelity of encrypted neural network inference: while polynomial approximations suffer from numerical divergence, scheme switching more closely matches non-encrypted inference. This numerical-fidelity advantage is also reflected in classification accuracy: scheme switching matches or nearly matches the non-encrypted baseline, whereas the Chebyshev-based approach degrades accuracy by up to 11% on LeNet-5, due to the accumulation of approximation errors across ReLU layers. Even though scheme switching enables exact ReLU computation, it incurs additional overhead from the expensive TFHE operations. To address this, we propose a set of optimizations in the non-ReLU portions of the network to offset the computational cost and achieve a more balanced performance profile. Our proposed optimizations reduce linear-layer latency by up to 100 times in LeNet-5's pooling layers, effectively mitigating the latency cost of ReLU via scheme-switching.
- 4 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1399-1414
Open AccessFully Homomorphic Encryption Inference of Neural Networks Using CKKS-TFHE Scheme Switching and Accelerated Linear Layers
Fully homomorphic encryption (FHE) allows neural network inference to be performed directly on encrypted data and models, preserving end-to-end privacy. One of the most widely used FHE schemes, CKKS, supports approximate arithmetic over real numbers and is well-suited for computing polynomial operations. However, CKKS does not natively support conditional branching, making it unsuitable for non-linear functions such as ReLU, which behaves differently depending on whether the input is negative or nonnegative. To address this, ReLU is typically approximated using polynomials. Unfortunately, this approach introduces two key drawbacks: (1) the approximation is only valid within a certain input range, and (2) identifying this range typically requires prior exposure of cleartext data to perform profiling, thus partially compromising privacy. This requirement is problematic in practice, as small profiling errors can lead to substantial accuracy degradation. On the other hand, the TFHE scheme supports binary gate-level computation, enabling precise implementation of conditional operations such as ReLU. Yet TFHE lacks the SIMD capabilities of CKKS, leading to significant computational latency. This paper explores the use of scheme switching for encrypted neural network inference: leveraging CKKS for polynomial-compatible layers and switching to TFHE for layers that require conditional logic, such as ReLU. We demonstrate that this hybrid approach substantially improves the numerical fidelity of encrypted neural network inference: while polynomial approximations suffer from numerical divergence, scheme switching more closely matches non-encrypted inference. This numerical-fidelity advantage is also reflected in classification accuracy: scheme switching matches or nearly matches the non-encrypted baseline, whereas the Chebyshev-based approach degrades accuracy by up to 11% on LeNet-5, due to the accumulation of approximation errors across ReLU layers. Even though scheme switching enables exact ReLU computation, it incurs additional overhead from the expensive TFHE operations. To address this, we propose a set of optimizations in the non-ReLU portions of the network to offset the computational cost and achieve a more balanced performance profile. Our proposed optimizations reduce linear-layer latency by up to 100 times in LeNet-5's pooling layers, effectively mitigating the latency cost of ReLU via scheme-switching.
- Open Access
SPARS: A reinforcement learning-enabled simulator for power management in HPC job scheduling
- Open Access
- 25 January 2026
- Published by Association for Computing Machinery (ACM)
- p. 13-21
Co-Design of a Power State-Aware Scheduler and an Intelligent Power Manager for Energy-Efficient HPC Systems
- 25 January 2026
- Published by Association for Computing Machinery (ACM)
- p. 13-21
Open AccessCo-Design of a Power State-Aware Scheduler and an Intelligent Power Manager for Energy-Efficient HPC Systems
- Open Access
- 25 January 2026
- Published by Association for Computing Machinery (ACM)
- p. 130-141
Deep Learning-Integrated Pairwise-Qubit Subsystems for Highly Efficient Quantum Circuit Simulation
- 25 January 2026
- Published by Association for Computing Machinery (ACM)
- p. 130-141
Open AccessDeep Learning-Integrated Pairwise-Qubit Subsystems for Highly Efficient Quantum Circuit Simulation
- Open Access
SPARS: A Reinforcement Learning-Enabled Simulator for Power Management in HPC Job Scheduling
High-performance computing (HPC) clusters consume enormous amounts of energy, with idle nodes as a major source of waste. Powering down unused nodes can mitigate this problem, but poorly timed transitions introduce long delays and reduce overall performance. To address this trade-off, we present SPARS, a reinforcement learning-enabled simulator for power management in HPC job scheduling. SPARS integrates job scheduling and node power state management within a discrete-event simulation framework. It supports traditional scheduling policies such as First Come First Served and EASY Backfilling, along with enhanced variants that employ reinforcement learning agents to dynamically decide when nodes should be powered on or off. Users can configure workloads and platforms in JSON format, specifying job arrivals, execution times, node power models, and transition delays. The simulator records comprehensive metrics-including energy usage, wasted power, job waiting times, and node utilization-and provides Gantt chart visualizations to analyze scheduling dynamics and power transitions. Unlike widely used Batsim-based frameworks that rely on heavy inter-process communication, SPARS provides lightweight event handling and consistent simulation results, making experiments easier to reproduce and extend. Its modular design allows new scheduling heuristics or learning algorithms to be integrated with minimal effort. By providing a flexible, reproducible, and extensible platform, SPARS enables researchers and practitioners to systematically evaluate power-aware scheduling strategies, explore the trade-offs between energy efficiency and performance, and accelerate the development of sustainable HPC operations.
Open AccessSPARS: A Reinforcement Learning-Enabled Simulator for Power Management in HPC Job Scheduling
High-performance computing (HPC) clusters consume enormous amounts of energy, with idle nodes as a major source of waste. Powering down unused nodes can mitigate this problem, but poorly timed transitions introduce long delays and reduce overall performance. To address this trade-off, we present SPARS, a reinforcement learning-enabled simulator for power management in HPC job scheduling. SPARS integrates job scheduling and node power state management within a discrete-event simulation framework. It supports traditional scheduling policies such as First Come First Served and EASY Backfilling, along with enhanced variants that employ reinforcement learning agents to dynamically decide when nodes should be powered on or off. Users can configure workloads and platforms in JSON format, specifying job arrivals, execution times, node power models, and transition delays. The simulator records comprehensive metrics-including energy usage, wasted power, job waiting times, and node utilization-and provides Gantt chart visualizations to analyze scheduling dynamics and power transitions. Unlike widely used Batsim-based frameworks that rely on heavy inter-process communication, SPARS provides lightweight event handling and consistent simulation results, making experiments easier to reproduce and extend. Its modular design allows new scheduling heuristics or learning algorithms to be integrated with minimal effort. By providing a flexible, reproducible, and extensible platform, SPARS enables researchers and practitioners to systematically evaluate power-aware scheduling strategies, explore the trade-offs between energy efficiency and performance, and accelerate the development of sustainable HPC operations.
- Open Access
- 10 March 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-13
Improving the Efficiency of a Deep Reinforcement Learning-Based Power Management System for HPC Clusters Using Curriculum Learning
- 10 March 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-13
Open AccessImproving the Efficiency of a Deep Reinforcement Learning-Based Power Management System for HPC Clusters Using Curriculum Learning
- Open Access
- 13 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Letters
- Vol. 8 (7) , 1-4
Investigating the Impact of Optimal Data Transfer Intervals on Failure-Prone Wireless Sensor Networks
Wireless sensor networks (WSNs) typically consist of failure-prone sensor nodes and more reliable sink nodes. To prevent data loss, sensor nodes must regularly transfer their data to sink nodes. Therefore, setting an appropriate data transfer interval between them is crucial. This paper presents a method to optimize the data transfer interval using a Markov model. Although the Markov model directly implies the assumption of exponentially distributed failures' inter-arrival times, using extensive simulations, we showed that the optimal data transfer interval derived by this model still performs exceptionally well under other failure distributions with less than 7.3% difference in terms of period of data loss compared to the true optimal data transfer interval. We also discuss how to integrate the proposed method into the widely used communication protocol in WSN, TDMA, and show that it can improve data collection time and energy efficiency.
- 13 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Letters
- Vol. 8 (7) , 1-4
Open AccessInvestigating the Impact of Optimal Data Transfer Intervals on Failure-Prone Wireless Sensor Networks
Wireless sensor networks (WSNs) typically consist of failure-prone sensor nodes and more reliable sink nodes. To prevent data loss, sensor nodes must regularly transfer their data to sink nodes. Therefore, setting an appropriate data transfer interval between them is crucial. This paper presents a method to optimize the data transfer interval using a Markov model. Although the Markov model directly implies the assumption of exponentially distributed failures' inter-arrival times, using extensive simulations, we showed that the optimal data transfer interval derived by this model still performs exceptionally well under other failure distributions with less than 7.3% difference in terms of period of data loss compared to the true optimal data transfer interval. We also discuss how to integrate the proposed method into the widely used communication protocol in WSN, TDMA, and show that it can improve data collection time and energy efficiency.
- Open Access
- 1 February 2024
- Published by Elsevier in Data in Brief
- Vol. 53, 110131
A pothole video dataset for semantic segmentation
- 1 February 2024
- Published by Elsevier in Data in Brief
- Vol. 53, 110131
Open AccessA pothole video dataset for semantic segmentation
- 8 April 2023
- Published by Springer Nature
- p. 94-107
An Advantage Actor-Critic Deep Reinforcement Learning Method for Power Management in HPC Systems
- 8 April 2023
- Published by Springer Nature
- p. 94-107
An Advantage Actor-Critic Deep Reinforcement Learning Method for Power Management in HPC Systems