Xue Jiang, Ge Li, Jiaru Qian, Xianjie Shi, Chenjie Li, Hao Zhu, Ziyu Wang, Jielun Zhang, Zeyu Zhao, Kechi Zhang, Jia Li, Wenpin Jiao, Zhi Jin, Yihong Dong. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Automatic Modulation Recognition (AMR) is crucial for optimizing communication systems, facilitating effective spectrum management and robust signal processing. Traditional AMR techniques, leveraging Machine Learning and Deep Learning algorithms, perform well under controlled conditions but struggle with domain variability, particularly changes in Signal-to-Noise Ratio (SNR). SNR variability, common in real-world environments due to mobility and interference, degrades classification accuracy and system reliability. In this regard, we propose a novel AMR framework leveraging Domain-Adversarial Neural Networks to address SNR variability. Our approach employs domain adversarial learning techniques to align feature distributions across different SNR levels, mitigating domain shifts and enhancing modulation recognition robustness. Extensive experiments demonstrate significant improvements in classification accuracy compared to existing techniques, highlighting the potential of domain adversarial methods in overcoming domain discrepancies in AMR.

Efficient and reliable transmission of CSI is crucial for maintaining optimal performance in wireless communication networks. However, the feedback overhead remains a significant challenge. In this paper, we propose a novel feedback compression scheme leveraging the Sobel operator to enhance the efficiency of CSI transmission as yet another novel computer vision based solution. Specifically, our proposed scheme differentiates the dynamic and quasi-static components in given CSI matrices and builds on top of the CsiNet to learn and compress the dynamic components in CSI matrices, thereby significantly reducing feedback overhead while maintaining high reconstruction quality. We perform extensive simulations to demonstrate the effectiveness of our proposed scheme. Evaluation results indicate that our algorithm achieves comparable compression performance to baseline algorithms while significantly reducing computational overhead.

Automated Internet of Things (IoT) devices generate a considerable amount of data continuously. However, an IoT network can be vulnerable to botnet attacks, where a group of IoT devices can be infected by malware and form a botnet. Recently, Artificial Intelligence (AI) algorithms have been introduced to detect and resist such botnet attacks in IoT networks. However, most of the existing Deep Learning-based algorithms are designed and implemented in a centralized manner. Therefore, these approaches can be sub-optimal in detecting zero-day botnet attacks against a group of IoT devices. Besides, a centralized AI approach requires sharing of data traces from the IoT devices for training purposes, which jeopardizes user privacy. To tackle these issues in this paper, we propose a federated learning based framework for a zero-day botnet attack detection model, where a new aggregation algorithm for the IoT devices is developed so that a better model aggregation can be achieved without compromising user privacy. Evaluations are conducted on an open dataset, i.e., the N-BaIoT. The evaluation results demonstrate that the proposed learning framework with the new aggregation algorithm outperforms the existing baseline aggregation algorithms in federated learning for zero-day botnet attack detection in IoT networks.

Network traffic classification plays an essential role in network measurement and management. Emerging Artificial Intelligence (AI) algorithms have become a viable solution to encrypted network traffic classification. Nonetheless, the classification performance of existing AI-based traffic classifiers is restricted to a limited number of network applications depending on the coverage of the knowledge database. Such AI-based traffic classifiers cannot maintain high performance to provide accurate traffic classification when dealing with updated or new network applications. To tackle the issues, we present an autonomous model update mechanism to sustain the high performance of AI-based traffic classifiers. Specifically, an instability check algorithm is derived to evaluate if the current classifier requires an update. A filtering algorithm is proposed to extract unknown traffic and build a new knowledge database based on a new metric, i.e., familiarity, defined based on the prediction confidence and instability. Extensive experiment results demonstrate that our proposed updating mechanism can provide prompt model updates and establish a proper new knowledge base to maintain high accuracy in various experimental scenarios. Moreover, the comparison is conducted and the results show the proposed familiarity-based filtering algorithm can filter about 7 and 3 times more true positive packets in the two considered scenarios, respectively.

Recently, many Artificial Intelligence (AI) based schemes have been proposed to support network measurement and management, such as network traffic classification, intrusion detection, traffic prediction, etc. These AI schemes have demonstrated promising performance in supporting networking. However, the development of these AI schemes requires a massive amount of fresh databases. The scarcity and futility of public datasets are straining the development of the networking AI models. Not to mention that most available datasets are not up-to-date. Collecting new datasets can be time-consuming and restricted by networking capabilities. To address the issues, we have introduced a real-application enabled network traffic generator. In this work, we further enhance the network traffic generator with more functionalities. In particular, a traffic flow segment scheme is proposed for the quick establishment of traffic flow databases. An intelligent generator is implemented to simulate point-to-point communications, including network multiplexing, and network duplexing. The evaluation results demonstrate that the improved intelligent traffic generator can generate a large amount of diverse network traffic with practical settings more efficiently than collecting data in real life. Moreover, a case study is given to demonstrate the quality of the generated traffic data.

Network measurement and management are more challenging in the next generation network systems due to the increasing demand for communications and complex network infrastructure. Recently, artificial intelligence (AI) algorithms have attracted much attention in networking systems, such as AI-based network traffic classification, traffic prediction, intrusion detection systems, etc. The development and maintenance of networking AI models usually require a large amount of traffic data samples from real applications. However, the publicly available datasets for network development are limited and rarely updated. In this paper, we develop a real application enabled traffic generator for AI model development in networking. In particular, a data loader is provided to establish two databases. One is a payload database that consists of packets from real applications. The other one is a traffic database that consists of network traffic flow statistics. The traffic generator allows a user to simulate data traffic flows that mimic one or more real applications. Moreover, two networking AI models are implemented to validate the simulated traffic flows. Evaluation results demonstrate that the developed traffic generator can help with networking AI model development.

Network intrusion detection is the fundamental of the Cybersecurity which plays an important role in preventing the systems away from malicious network traffic. Recent Artificial Intelligence (AI) based intrusion detection systems provide simple and accurate intrusion detection compared with the conventional intrusion detection schemes, however, the detection performance may not be reliable because the models in the AI algorithms must output a prediction result for each incoming instance even when the models are not confident. To tackle the issue, we propose to adopt Bayesian Deep Learning, specifically, Bayesian Convolutional Neural Network, to build intrusion detection models. Moreover, an ensemble-based detection scheme is further proposed to enhance the detection performance. Two open datasets (i.e., NSL-KDD and UNSW-NB15) are used to evaluate the proposed schemes. In comparison, Convolutional Neural Network and Support Vector Machine are implemented as baseline IDS (i.e., CNN-IDS and SVM-IDS). The evaluation results demonstrate that the proposed BCNN-IDS can significantly boost the detection accuracy and reduce the false alarm rate by adopting the proposed T-ensemble detection scheme.

Network traffic classification has been widely studied to fundamentally advance network measurement and management. Machine Learning is one of the effective approaches for network traffic classification. Specifically, Deep Learning (DL) has attracted much attention from the researchers due to its effectiveness even in encrypted network traffic without compromising neither user privacy nor network security. However, most of the existing models are created from closed-world datasets, thus they can only classify those existing classes previously sampled and labeled. In this case, unknown classes cannot be correctly classified. To tackle this issue, an autonomous learning framework is proposed to effectively update DL-based traffic classification models during active operations. The core of the proposed framework consists of a DL-based classifier, a self-learned discriminator, and an autonomous self-labeling model. The discriminator and self-labeling process can generate new dataset during active operations to support classifier update. Evaluation of the proposed framework is performed on an open dataset, i.e., ISCX VPN-nonVPN, and independently collected data packets. The results demonstrate that the proposed autonomous learning framework can filter packets from unknown classes and provide accurate labels. Thus, corresponding DL-based classification models can be updated successfully with the autonomously generated dataset.

Network traffic classification is essential in access network for end-to-end network management and measurement such as network intrusion detection, network resource allocation. State-of- the-art Deep Learning based classifiers have high accuracy even when processing encrypted data packets. Such classifiers would need to be updated when a new application is in the network traffic. However, it is challenging to build and label a dataset of the unknown application from active network traffic. In this paper, we propose an autonomous model update scheme to (i) filter the data packets of a new application from active network traffic and build a corresponding training dataset; and (ii) update the current network traffic classifier with transfer learning. In particular, the core of the proposed scheme is a discriminator that consists of a statistical filter and a convolutional neural network based binary classifier to filter and build a dataset of new application packets from active network traffic. Evaluation is conducted based on an open dataset (i.e., ISCX VPN-nonVPN dataset). The results demonstrated that our proposed autonomous classifier update scheme can successfully filter packets of a new application from network traffic and build a corresponding training dataset. Moreover, the packet classifier can be effectively updated through transfer learning. The proposed update scheme can contribute significantly in the access network for further end-to-end network measurement and management.