KyungHyun Han
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
Research Subjects
KyungHyun Han has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Seong Oun Hwang
Department of Computer Engineering, Gachon University, Seongnam, South Korea
12 shared publications - Wai-Kong Lee
Universiti Tunku Abdul Rahman
4 shared publications - Ali Raza
Honda Research Institute Europe, Europe, Germany
2 shared publications - Myung-Kyu Yi
Hanyang University
2 shared publications - Yasir Ali
Department of IT Convergence Engineering, Gachon University, Seongnam, South Korea
2 shared publications - Abdul Majeed
Secure Cyber Systems Research Group, WMG, University of Warwick, Coventry CV4 7AL, UK
1 shared publication
Latest Publications
- Open Access
- 1 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 60108-60121
An Optimal Two-Step Approach for Defense Against Poisoning Attacks in Federated Learning
Federated learning (FL) has gained widespread adoption for training artificial intelligence (AI) models while ensuring the confidentiality of client data. However, this privacy-preserving nature of FL also makes it vulnerable to poisoning attacks. To counter these attacks, several defense methods have been developed to identify and filter out poisoned local models before the aggregation process. Nevertheless, these defense methods demonstrate sub-optimal performance in retaining benign local models while discarding poisoned local models primarily due to inadequate filtering strategies. Consequently, these defense methods filter out large proportions of benign local models that are not poisoned, resulting in high false rejection rates or low detection accuracy, which leads to test accuracy degradation of the global model as well. In this paper, we propose the Two-step Defense Framework for Poisoning Attacks Detection (TDF-PAD), which first identifies the obvious-poisoned, obvious-benign, and ambiguous local models by utilizing the interquartile range method. The second step employs the Z-score method to classify ambiguous local models into benign or poisoned local models based on their performance history. Through extensive experimentation on two real-world benchmark datasets, we demonstrate that TDF-PAD outperforms state-of-the-art defense methods by achieving a 0% false positive rate on these benchmark datasets, showing it is generally applicable to any dataset.
- 1 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 60108-60121
Open AccessAn Optimal Two-Step Approach for Defense Against Poisoning Attacks in Federated Learning
Federated learning (FL) has gained widespread adoption for training artificial intelligence (AI) models while ensuring the confidentiality of client data. However, this privacy-preserving nature of FL also makes it vulnerable to poisoning attacks. To counter these attacks, several defense methods have been developed to identify and filter out poisoned local models before the aggregation process. Nevertheless, these defense methods demonstrate sub-optimal performance in retaining benign local models while discarding poisoned local models primarily due to inadequate filtering strategies. Consequently, these defense methods filter out large proportions of benign local models that are not poisoned, resulting in high false rejection rates or low detection accuracy, which leads to test accuracy degradation of the global model as well. In this paper, we propose the Two-step Defense Framework for Poisoning Attacks Detection (TDF-PAD), which first identifies the obvious-poisoned, obvious-benign, and ambiguous local models by utilizing the interquartile range method. The second step employs the Z-score method to classify ambiguous local models into benign or poisoned local models based on their performance history. Through extensive experimentation on two real-world benchmark datasets, we demonstrate that TDF-PAD outperforms state-of-the-art defense methods by achieving a 0% false positive rate on these benchmark datasets, showing it is generally applicable to any dataset.
- 14 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computing
- Vol. 13 (3) , 640-653
QuripfeNet: Quantum-Resistant IPFE-Based Neural Network
In order to protect the sensitive information in many applications involving neural networks, several privacy-preserving neural networks that operate on encrypted data have been developed. Unfortunately, existing encryption-based privacy-preserving neural networks are mainly built on classical cryptography primitives, which are not secure from the threat of quantum computing. In this paper, we propose the first quantum-resistant solution to protect neural network inferences based on an inner-product functional encryption scheme. The selected state-of-the-art functional encryption scheme based on lattice-based cryptography works with integer-type inputs, which is not directly compatible with neural network computations that operate in the floating point domain. We propose a polynomial-based secure convolution layer to allow a neural network to resolve this problem, along with a technique that reduces memory consumption. The proposed solution, named QuripfeNet, was applied in LeNet-5 and evaluated using the MNIST dataset. In a single-threaded implementation (CPU), QuripfeNet took 107.4 seconds for an inference to classify one image, achieving accuracy of 97.85%, which is very close to the unencrypted version. Additionally, the GPU-optimized QuripfeNet took 25.9 seconds to complete the same task, which is improved by 4.15 × compared to the CPU version.
- 14 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computing
- Vol. 13 (3) , 640-653
QuripfeNet: Quantum-Resistant IPFE-Based Neural Network
In order to protect the sensitive information in many applications involving neural networks, several privacy-preserving neural networks that operate on encrypted data have been developed. Unfortunately, existing encryption-based privacy-preserving neural networks are mainly built on classical cryptography primitives, which are not secure from the threat of quantum computing. In this paper, we propose the first quantum-resistant solution to protect neural network inferences based on an inner-product functional encryption scheme. The selected state-of-the-art functional encryption scheme based on lattice-based cryptography works with integer-type inputs, which is not directly compatible with neural network computations that operate in the floating point domain. We propose a polynomial-based secure convolution layer to allow a neural network to resolve this problem, along with a technique that reduces memory consumption. The proposed solution, named QuripfeNet, was applied in LeNet-5 and evaluated using the MNIST dataset. In a single-threaded implementation (CPU), QuripfeNet took 107.4 seconds for an inference to classify one image, achieving accuracy of 97.85%, which is very close to the unencrypted version. Additionally, the GPU-optimized QuripfeNet took 25.9 seconds to complete the same task, which is improved by 4.15 × compared to the CPU version.
- 25 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (11) , 18556-18567
Fall Detection of the Elderly Using Denoising LSTM-Based Convolutional Variant Autoencoder
As societies age, the issue of falls has become increasingly critical for the health and safety of the elderly. Fall detection in the elderly has traditionally relied on supervised learning methods, which require data on falls, which is difficult to obtain in real situations. Additionally, the complexity of integrating deep learning models into wearable devices for real-time fall detection has been challenging due to limited computational resources. In this paper, we propose a novel fall detection method using unsupervised learning based on a denoising long short term memory (LSTM)-based convolutional variational autoencoder (CVAE) model to solve the problem of lack of fall data. By utilizing the proposed data debugging and hierarchical data balancing techniques, the proposed method achieves an F1 score of 1.0 while reducing the parameter count by 25.6 times compared to the state-of-the-art unsupervised deep learning method. The resulting model occupies only 157.65 KB of memory, making it highly suitable for integration into wearable devices.
- 25 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (11) , 18556-18567
Fall Detection of the Elderly Using Denoising LSTM-Based Convolutional Variant Autoencoder
As societies age, the issue of falls has become increasingly critical for the health and safety of the elderly. Fall detection in the elderly has traditionally relied on supervised learning methods, which require data on falls, which is difficult to obtain in real situations. Additionally, the complexity of integrating deep learning models into wearable devices for real-time fall detection has been challenging due to limited computational resources. In this paper, we propose a novel fall detection method using unsupervised learning based on a denoising long short term memory (LSTM)-based convolutional variational autoencoder (CVAE) model to solve the problem of lack of fall data. By utilizing the proposed data debugging and hierarchical data balancing techniques, the proposed method achieves an F1 score of 1.0 while reducing the parameter count by 25.6 times compared to the state-of-the-art unsupervised deep learning method. The resulting model occupies only 157.65 KB of memory, making it highly suitable for integration into wearable devices.
- Open Access
- 6 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 124139-124149
GPU-Accelerated Deep Learning-Based Correlation Attack on Tor Networks
The Tor network, renowned for its provision of online privacy and anonymity, faces the constant threat of correlation attacks that aim to compromise user identities. For almost two decades, these correlation attacks were based on statistical methods. However, in recent years, deep learning-based correlation attacks have been introduced to make them more accurate. Nevertheless, in addition to being accurate, fast correlation attacks on Tor are crucial for assessing the real-world viability of such attacks because reduced correlation time aids in estimating its practical implications. Moreover, a reduction in correlation time also helps improve efficiency and ensures practical relevance of the attack. The existing state-of-the-art implementation of a correlation attack on Tor suffers from slow performance and large memory requirements. For instance, training the model required 133 GB of memory, and correlating 10,000 flows takes about 976 seconds. In this paper, we present a novel GPU-based correlation strategy and a fast traffic flow loading technique to reduce time complexity by 7.12× compared to existing methods. Our computational approach, reliant on PyCUDA, facilitates the parallelization of operations used in the attack, thereby enabling efficient execution through the utilization of GPU architecture. Leveraging these two approaches, we introduced an improved correlation attack, which shows high accuracy and fast performance compared to state-of-the-art methods. Moreover, we address resource limitation issues by reducing memory consumption by 47.37% during the training phase, which allows the model to be trained with much fewer resources.
- 6 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 124139-124149
Open AccessGPU-Accelerated Deep Learning-Based Correlation Attack on Tor Networks
The Tor network, renowned for its provision of online privacy and anonymity, faces the constant threat of correlation attacks that aim to compromise user identities. For almost two decades, these correlation attacks were based on statistical methods. However, in recent years, deep learning-based correlation attacks have been introduced to make them more accurate. Nevertheless, in addition to being accurate, fast correlation attacks on Tor are crucial for assessing the real-world viability of such attacks because reduced correlation time aids in estimating its practical implications. Moreover, a reduction in correlation time also helps improve efficiency and ensures practical relevance of the attack. The existing state-of-the-art implementation of a correlation attack on Tor suffers from slow performance and large memory requirements. For instance, training the model required 133 GB of memory, and correlating 10,000 flows takes about 976 seconds. In this paper, we present a novel GPU-based correlation strategy and a fast traffic flow loading technique to reduce time complexity by 7.12× compared to existing methods. Our computational approach, reliant on PyCUDA, facilitates the parallelization of operations used in the attack, thereby enabling efficient execution through the utilization of GPU architecture. Leveraging these two approaches, we introduced an improved correlation attack, which shows high accuracy and fast performance compared to state-of-the-art methods. Moreover, we address resource limitation issues by reducing memory consumption by 47.37% during the training phase, which allows the model to be trained with much fewer resources.
- 30 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computing
- Vol. 12 (1) , 328-343
cuFE: High Performance Privacy Preserving Support Vector Machine With Inner-Product Functional Encryption
Privacy preservation is a sensitive and important issue in this ever-growing and highly-connected digital era. Functional encryption is a computation on encrypted data paradigm that allows users to retrieve the evaluation of a function on encrypted data without revealing the data, effectively protecting user's privacy. However, existing functional encryption implementations are still very time-consuming for practical deployment, especially when applied to machine learning applications that involve huge amount of data. In this paper, we present a high-performance implementation of inner-product functional encryption (IPFE) based on ring-learning with errors on graphics processing units. We execute a systematic investigation to select the best strategy for implementing number theoretic transform for different security levels, which is the most time-consuming operations in the IPFE scheme. We further propose novel techniques to parallelize the Gaussian sampling. Compared to the existing AVX2 implementation, our implementation on a RTX 2060 GPU achieves $34.24\times$ , $40.02\times$ , $156.30\times$ and $18.76\times$ speed-up for Setup , Encrypt , KeyGen and Decrypt respectively. Finally, we propose a fast privacy-preserving SVM to classify data securely using our GPU-accelerated IPFE scheme. On average, our implementation can classify one input with 591 support vectors in 688 ms ( $< 1$ second), which is $33.12\times$ faster than the AVX2 version.
- 30 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computing
- Vol. 12 (1) , 328-343
cuFE: High Performance Privacy Preserving Support Vector Machine With Inner-Product Functional Encryption
Privacy preservation is a sensitive and important issue in this ever-growing and highly-connected digital era. Functional encryption is a computation on encrypted data paradigm that allows users to retrieve the evaluation of a function on encrypted data without revealing the data, effectively protecting user's privacy. However, existing functional encryption implementations are still very time-consuming for practical deployment, especially when applied to machine learning applications that involve huge amount of data. In this paper, we present a high-performance implementation of inner-product functional encryption (IPFE) based on ring-learning with errors on graphics processing units. We execute a systematic investigation to select the best strategy for implementing number theoretic transform for different security levels, which is the most time-consuming operations in the IPFE scheme. We further propose novel techniques to parallelize the Gaussian sampling. Compared to the existing AVX2 implementation, our implementation on a RTX 2060 GPU achieves $34.24\times$ , $40.02\times$ , $156.30\times$ and $18.76\times$ speed-up for Setup , Encrypt , KeyGen and Decrypt respectively. Finally, we propose a fast privacy-preserving SVM to classify data securely using our GPU-accelerated IPFE scheme. On average, our implementation can classify one input with 591 support vectors in 688 ms ( $< 1$ second), which is $33.12\times$ faster than the AVX2 version.
- Open Access
- 29 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 32051-32060
Non-Zero Grid for Accurate 2-Bit Additive Power-of-Two CNN Quantization
Quantization is an effective technique to reduce the memory and computational complexity of CNNs. Recent advances utilize additive powers-of-two to perform non-uniform quantization, which resembles a normal distribution and shows better performance than uniform quantization. With powers-of-two quantization, the computational complexity is also largely reduced because the slow multiplication operations are replaced with lightweight shift operations. However, there are serious problems in the previously proposed grid formulation for 2-bit quantization. In particular, these powers-of-two schemes produce zero values, generating significant training error and causing low accuracy. In addition, due to improper grid formulation, they also fallback to uniform quantization when the quantization level reaches 2-bit. Due to these reasons, on large CNN like ResNet-110, these powers-of-two schemes may not even train properly. To resolve these issues, we propose a new non-zero grid formulation that enables 2-bit non-uniform quantization and allow the CNN to be trained successfully in every attempt, even for a large network. The proposed technique quantizes weight as power-of-two values and projects it close to the mean area through a simple constant product on the exponential part. This allows our quantization scheme to closely resemble a non-uniform quantization at 2-bit, enabling successful training at 2-bit quantization, which is not found in the previous work. The proposed technique achieves 70.57% accuracy on the CIFAR-100 dataset trained with ResNet-110. This result is 6.24% higher than the additive powers-of-two scheme which only achieves 64.33% accuracy. Beside achieving higher accuracy, our work also maintains the same memory and computational efficiency with the original additive powers-of-two scheme.
- 29 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 32051-32060
Open AccessNon-Zero Grid for Accurate 2-Bit Additive Power-of-Two CNN Quantization
Quantization is an effective technique to reduce the memory and computational complexity of CNNs. Recent advances utilize additive powers-of-two to perform non-uniform quantization, which resembles a normal distribution and shows better performance than uniform quantization. With powers-of-two quantization, the computational complexity is also largely reduced because the slow multiplication operations are replaced with lightweight shift operations. However, there are serious problems in the previously proposed grid formulation for 2-bit quantization. In particular, these powers-of-two schemes produce zero values, generating significant training error and causing low accuracy. In addition, due to improper grid formulation, they also fallback to uniform quantization when the quantization level reaches 2-bit. Due to these reasons, on large CNN like ResNet-110, these powers-of-two schemes may not even train properly. To resolve these issues, we propose a new non-zero grid formulation that enables 2-bit non-uniform quantization and allow the CNN to be trained successfully in every attempt, even for a large network. The proposed technique quantizes weight as power-of-two values and projects it close to the mean area through a simple constant product on the exponential part. This allows our quantization scheme to closely resemble a non-uniform quantization at 2-bit, enabling successful training at 2-bit quantization, which is not found in the previous work. The proposed technique achieves 70.57% accuracy on the CIFAR-100 dataset trained with ResNet-110. This result is 6.24% higher than the additive powers-of-two scheme which only achieves 64.33% accuracy. Beside achieving higher accuracy, our work also maintains the same memory and computational efficiency with the original additive powers-of-two scheme.
- 1 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IT Professional
- Vol. 23 (6) , 46-51
CAPTCHA-Based Secret-Key Sharing Using Quantum Communication
Digital communication has revolutionized the modern era, where two parties can communicate apart easily; but this also opens up opportunities for adversarial attacks. To guarantee a secure communication over such channels, there have been several approaches proposed based on secret-key distribution including the BB84 quantum secret-key sharing protocol. Although it is almost resistant to eavesdropping attacks, the BB84 protocol presents certain challenges. Specifically, it is vulnerable to impersonation attacks. Moreover, it is slow owing to the hit-and-trail method for secret-key distribution. Hence, in this article, we propose a secure and efficient protocol for secret-key sharing by combining image-based CAPTCHA and the classical BB84 protocol. We demonstrate that the proposed protocol is secure against impersonation attacks, including DDoS attacks, and it is at least twice as efficient as BB84 in terms of communicated qubits.
- 1 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IT Professional
- Vol. 23 (6) , 46-51
CAPTCHA-Based Secret-Key Sharing Using Quantum Communication
Digital communication has revolutionized the modern era, where two parties can communicate apart easily; but this also opens up opportunities for adversarial attacks. To guarantee a secure communication over such channels, there have been several approaches proposed based on secret-key distribution including the BB84 quantum secret-key sharing protocol. Although it is almost resistant to eavesdropping attacks, the BB84 protocol presents certain challenges. Specifically, it is vulnerable to impersonation attacks. Moreover, it is slow owing to the hit-and-trail method for secret-key distribution. Hence, in this article, we propose a secure and efficient protocol for secret-key sharing by combining image-based CAPTCHA and the classical BB84 protocol. We demonstrate that the proposed protocol is secure against impersonation attacks, including DDoS attacks, and it is at least twice as efficient as BB84 in terms of communicated qubits.
- 21 September 2021
- Published by Springer Nature in Cluster Computing
- Vol. 25 (1) , 433-450
cuGimli: optimized implementation of the Gimli authenticated encryption and hash function on GPU for IoT applications
- 21 September 2021
- Published by Springer Nature in Cluster Computing
- Vol. 25 (1) , 433-450
cuGimli: optimized implementation of the Gimli authenticated encryption and hash function on GPU for IoT applications
- Open Access
- 3 September 2020
- Published by MDPI AG in Applied Sciences
- Vol. 10 (17) , 6116
Lightweight Detection Method of Obfuscated Landing Sites Based on the AST Structure and Tokens
Attackers use a variety of techniques to insert redirection JavaScript that leads a user to a malicious webpage, where a drive-by-download attack is executed. In particular, the redirection JavaScript in the landing site is obfuscated to avoid detection systems. In this paper, we propose a lightweight detection system based on static analysis to classify the obfuscation type and to promptly detect the obfuscated redirection JavaScript. The proposed model detects the obfuscated redirection JavaScript by converting the JavaScript into an abstract syntax tree (AST). Then, the structure and token information are extracted. Specifically, we propose a lightweight AST to identify the obfuscation type and the revised term frequency-inverse document frequency to efficiently detect the malicious redirection JavaScript. This approach enables rapid identification of the obfuscated redirection JavaScript and proactive blocking of the webpages that are used in drive-by-download attacks.
- 3 September 2020
- Published by MDPI AG in Applied Sciences
- Vol. 10 (17) , 6116
Open AccessLightweight Detection Method of Obfuscated Landing Sites Based on the AST Structure and Tokens
Attackers use a variety of techniques to insert redirection JavaScript that leads a user to a malicious webpage, where a drive-by-download attack is executed. In particular, the redirection JavaScript in the landing site is obfuscated to avoid detection systems. In this paper, we propose a lightweight detection system based on static analysis to classify the obfuscation type and to promptly detect the obfuscated redirection JavaScript. The proposed model detects the obfuscated redirection JavaScript by converting the JavaScript into an abstract syntax tree (AST). Then, the structure and token information are extracted. Specifically, we propose a lightweight AST to identify the obfuscation type and the revised term frequency-inverse document frequency to efficiently detect the malicious redirection JavaScript. This approach enables rapid identification of the obfuscated redirection JavaScript and proactive blocking of the webpages that are used in drive-by-download attacks.
- Open Access
- 30 April 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 83425-83443
PhishHaven—An Efficient Real-Time AI Phishing URLs Detection System
Different machine learning and deep learning-based approaches have been proposed for designing defensive mechanisms against various phishing attacks. Recently, researchers showed that phishing attacks can be performed by employing a deep neural network-based phishing URL generating system called DeepPhish. To prevent this kind of attack, we design an ensemble machine learning-based detection system called PhishHaven to identify AI-generated as well as human-crafted phishing URLs. To the best of our knowledge, this is the first study to consider detecting phishing attacks by both AI and human attackers. PhishHaven employs lexical analysis for feature extraction. To further enhance lexical analysis, we introduce URL HTML Encoding to classify URL on-the-fly and proactively compare with some of the existing methods. We also introduce a URL Hit approach to deal with tiny URLs, which is an open problem yet to be solved. Moreover, the final classification of URLs is made on an unbiased voting mechanism in PhishHaven, which aims to avoid misclassification when the number of votes is equal. To speed up the ensemble-based machine learning models, PhishHaven employs a multi-threading approach to execute the classification in parallel, leading to real-time detection. Theoretical analysis of our solution shows that (1) it can always detect tiny URLs, and (2) it can detect future AI-generated Phishing URLs based on our selected lexical features with 100% accuracy. Through experiments, we analyze our solution with a benchmark dataset of 100,000 phishing and normal URLs. The results show that PhishHaven can achieve 98.00% accuracy, outperforming the existing lexical-based human-crafted phishing URLs detection systems.
- 30 April 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 83425-83443
Open AccessPhishHaven—An Efficient Real-Time AI Phishing URLs Detection System
Different machine learning and deep learning-based approaches have been proposed for designing defensive mechanisms against various phishing attacks. Recently, researchers showed that phishing attacks can be performed by employing a deep neural network-based phishing URL generating system called DeepPhish. To prevent this kind of attack, we design an ensemble machine learning-based detection system called PhishHaven to identify AI-generated as well as human-crafted phishing URLs. To the best of our knowledge, this is the first study to consider detecting phishing attacks by both AI and human attackers. PhishHaven employs lexical analysis for feature extraction. To further enhance lexical analysis, we introduce URL HTML Encoding to classify URL on-the-fly and proactively compare with some of the existing methods. We also introduce a URL Hit approach to deal with tiny URLs, which is an open problem yet to be solved. Moreover, the final classification of URLs is made on an unbiased voting mechanism in PhishHaven, which aims to avoid misclassification when the number of votes is equal. To speed up the ensemble-based machine learning models, PhishHaven employs a multi-threading approach to execute the classification in parallel, leading to real-time detection. Theoretical analysis of our solution shows that (1) it can always detect tiny URLs, and (2) it can detect future AI-generated Phishing URLs based on our selected lexical features with 100% accuracy. Through experiments, we analyze our solution with a benchmark dataset of 100,000 phishing and normal URLs. The results show that PhishHaven can achieve 98.00% accuracy, outperforming the existing lexical-based human-crafted phishing URLs detection systems.