Author name cluster

jinfu chen

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

3 papers

1 author row

EAAI Journal 2026 Journal Article

A novel android malware classification approach based on multi-scale feature fusion for encrypted traffic

Jiahui Lu
jinfu chen
Saihua Cai
Yisong Liu
Shengran Wang

With the widespread adoption of Android system, the volume of malware targeting Android devices continues to increase, posing significant threats to security and privacy data. To safeguard data transmission, most applications have adopted the Transport Layer Security (TLS) protocol. Although TLS improves communication security, it also provides a convenient means for attackers to conceal malicious payloads, thereby challenging traditional malware classification methods in encrypted traffic scenarios. However, most existing methods rely on a single type of network traffic feature, limiting their ability to capture complex malicious behaviors and leading to the issues such as limited feature diversity, weak generalization and sensitivity to encryption. To address these challenges, this paper proposes a novel Android Malware Classification method based on Multi-scale Feature Fusion (AMC-MFF) for encrypted network traffic. This method treats session flows as the basic unit and extracts gray-scale images, statistical features and TLS protocol features from raw byte streams, using Residual Network 18 (ResNet18), Bidirectional Gated Recurrent Unit (Bi-GRU), and Multi-Layer Perceptron (MLP) models, respectively, to achieve multi-scale feature modeling. Then, a two-layer cross-attention mechanism is introduced to capture the semantic associations among features from different scales, and a Transformer-based fusion module is employed to integrate these features effectively. Finally, the residual connections and a fully connected layer are used for accurate malware classification. Experimental results on two public datasets demonstrate that AMC-MFF outperforms existing baselines, achieving a classification accuracy of up to 99. 87%, which highlights its strong classification performance.

Details DOI

EAAI Journal 2026 Journal Article

An efficient framework for malicious network traffic detection using optimized deep learning techniques

Mukhtar Ahmed
jinfu chen
Ernest Akpaku
Ajmal Latif

The increasing sophistication of network attacks and the limitations of traditional machine learning-based detection methods pose significant challenges to modern cybersecurity. Existing approaches often depend heavily on labeled data and exhibit poor generalization across heterogeneous environments, limiting their effectiveness against emerging threats. To address these challenges, this paper introduces a novel framework for Malicious Network Traffic Detection (MNTD), designed to improve robustness and adaptability through deep learning methods. The model integrates Convolutional Neural Networks (CNNs), Bidirectional Long Short-Term Memory networks (BiLSTM), and Multi-Head Attention (MHA) mechanisms to capture both spatial and temporal dependencies in traffic. It further employs Adaptive Weighted Delay Velocity (AWDV) for hyperparameter optimization and contrastive learning to enhance feature discrimination, supported by an adaptive loss function and a regularized feature representation strategy to mitigate overfitting. The MNTD framework addresses a binary classification task, distinguishing between benign and malicious traffic. Experimental results demonstrate consistent state-of-the-art performance across four benchmark datasets. On the Canadian Institute for Cybersecurity Intrusion Detection System 2017 (CICIDS2017) dataset, it achieves 98. 52% accuracy and a 98. 98% F1-score. On the Canadian Institute for Cybersecurity Domain Name System (DNS) protocol Browser 2020 (CIRA-CIC-DoHBrw2020) dataset, it reaches 98. 82% accuracy and 98. 66% F1-score. For the Botnet Internet of Things (BoT-IoT) dataset, it obtains 98. 65% accuracy and 98. 40% F1-score, while on the University of New South Wales Network Benchmark 2015 (UNSW-NB15) dataset, it maintains 97. 91% accuracy and 97. 61% F1-score. This study demonstrates how artificial intelligence techniques can be effectively applied to cybersecurity applications, specifically malicious network traffic detection.

Details DOI

NeurIPS Conference 2025 Conference Paper

More Than Just Functional: LLM-as-a-Critique for Efficient Code Generation

Derui Zhu
Dingfan Chen
jinfu chen
Jens Grossklags
Alexander Pretschner
Weiyi Shang

Large language models (LLMs) have demonstrated remarkable progress in generating functional code, leading to numerous AI-based coding program tools. However, their reliance on the perplexity objective during both training and inference primarily emphasizes functionality, often at the expense of efficiency—an essential consideration for real-world coding tasks. Perhaps interestingly, we observed that well-trained LLMs inherently possess knowledge about code efficiency, but this potential remains underutilized with standard decoding approaches. To address this, we design strategic prompts to activate the model’s embedded efficiency understanding, effectively using LLMs as \textit{efficiency critiques} to guide code generation toward higher efficiency without sacrificing—and sometimes even improving—functionality, all without the need for costly real code execution. Extensive experiments on benchmark datasets (EffiBench, HumanEval+) across multiple representative code models demonstrate up to a 70. 6\% reduction in average execution time and a 13. 6\% decrease in maximum memory usage, highlighting the computational efficiency and practicality of our approach compared to existing alternatives.

PDF Details