Chen Hajaj

@article{aharon2026http2dos,
  author = {Aharon, Udi and Dubin, Ran and Dvir, Amit and Hajaj, Chen},
  title = {{HTTP2DoS}: Detecting low-and-slow {HTTP/2} {DoS} attacks with complex transfer-entropy graph-temporal learning},
  journal = {Computers \& Security},
  volume = {172},
  pages = {105198},
  year = {2026},
  month = oct,
  doi = {10.1016/j.cose.2026.105198},
  url = {https://doi.org/10.1016/j.cose.2026.105198},
  month_numeric = {10}
}

Copy BibTeX

The dominance of encryption and the relentless growth of network traffic have rendered traditional deep packet inspection ineffective and made retraining-based machine learning classifiers operationally costly, as every new application or threat class requires a full model rebuild. In order to overcome this challenge, we present an ultra-fast encrypted traffic classification system based on Classification By Retrieval (CBR): each flow is represented by a fixed-length feature vector extracted from only the first few packets, indexed in a HNSW approximate nearest-neighbor structure, and classified by retrieving its nearest labeled neighbor. New classes are absorbed by inserting a handful of labeled vectors into the running index, with no retraining and no downtime, and out-of-distribution traffic is detected within the same retrieval step using a single global distance threshold. We evaluate handcrafted statistical features and learned CNN and MLP embeddings against a Random Forest baseline on the public CESNET-TLS22 dataset (184 application classes) and a proprietary cellular dataset (RADCOM). With CNN embeddings, CBR matches Random Forest accuracy (macro F1 of 0.923 versus 0.925 on CESNET) while training roughly ten times faster, adapts to new classes two to three orders of magnitude faster (seconds versus thousands of seconds), and detects out-of-distribution flows at up to 0.92 AUROC. The retrieval index is also far cheaper to keep resident than the baseline it replaces, holding 1.2 million CESNET flows in 171 MB against 1,350 MB for the serialized forest, because its size grows with the number of indexed flows rather than with the number of classes. The Random Forest remains the faster classifier per flow, so the trade-off is per-flow latency against adaptability and memory. These results show that retrieval-based classification is a practical and adaptable alternative for real-time encrypted traffic analysis in dynamic network environments.

@article{zion2026ultrafast,
  author = {Zion, Yehonatan and Dvir, Amit and Hajaj, Chen and Dubin, Ran and Filber, Neriya},
  title = {Ultra fast encrypted network traffic classification based on classification by retrieval powered by out-of-distribution},
  journal = {Computer Networks},
  year = {2026},
  doi = {10.1016/j.comnet.2026.112739},
  url = {https://doi.org/10.1016/j.comnet.2026.112739}
}

Copy BibTeX

Traditional CAPTCHAs are increasingly vulnerable to deep learning-based solvers that decode text and images with high accuracy. In this work, we propose methods to strengthen adversarial CAPTCHAs without compromising human usability. First, we introduce a Precise Gradient Method (PGM) that preserves gradient magnitude (rather than discarding it via a sign operator), producing adversarial perturbations with significantly lower perceptual noise. Second, we develop intelligent target class selection, using either dataset-level confusion structure (Class Relations Network) or image-specific softmax probabilities (Distance-Based Target), to steer adversarial perturbations more efficiently. Across multiple modern architectures (MobileNets, EfficientNets, ResNet, and Vision Transformer), our framework achieves faster convergence (fewer iterations), reduced visual distortion, and notably greater robustness under iterative adversarial retraining. Experiments show that our methods consistently reduce iteration counts and perceptual distortion while significantly increasing the difficulty for automated attacks. Our results offer a practical, scalable path toward the next generation of CAPTCHA systems and contribute new insights to the adversarial machine learning landscape focused on security and usability.

@inproceedings{litman2026_cleaner_adversarial_captchas,
  doi = {10.65109/afaw3962},
  author = {Litman, Meir and Hajaj, Chen},
  title = {Cleaner Adversarial CAPTCHAs: Intelligent Targets and Precise Noise for Usable Security},
  booktitle = {International Conference on Autonomous Agents and Multiagent Systems (AAMAS)},
  year = {2026}
}

Copy BibTeX

Application Programming Interface (API) Injection attacks refer to the unauthorized or malicious use of APIs, which are often exploited to gain access to sensitive data or manipulate online systems for illicit purposes. Identifying actors that deceitfully utilize an API poses a demanding problem. Although there have been notable advancements and contributions in the field of API security, there remains a significant challenge when dealing with attackers who use novel approaches that don’t match the well-known payloads commonly seen in attacks. Also, attackers may exploit standard functionalities unconventionally and with objectives surpassing their intended boundaries. Thus, API security needs to be more sophisticated and dynamic than ever, with advanced computational intelligence methods, such as machine learning models that can quickly identify and respond to abnormal behavior. In response to these challenges, we propose a novel unsupervised few-shot anomaly detection framework composed of two main parts: First, we train a dedicated generic language model for API based on FastText embedding. Next, we use Approximate Nearest Neighbor search in a classification-by-retrieval approach. Our framework allows for training a fast, lightweight classification model using only a few examples of normal API requests. We evaluated the performance of our framework using the CSIC 2010 and ATRDF 2023 datasets. The results demonstrate that our framework improves API attack detection accuracy compared to the state-of-the-art (SOTA) unsupervised anomaly detection baselines.

@article{aharon2025classification,
  title = {A Classification-by-Retrieval Framework for Few-Shot Anomaly Detection to Detect API Injection},
  author = {Aharon, Udi and Dubin, Ran and Dvir, Amit and Hajaj, Chen},
  journal = {Computers \& Security},
  volume = {150},
  pages = {104249},
  year = {2025},
  publisher = {Elsevier Advanced Technology},
  doi = {10.1016/j.cose.2024.104249}
}

Copy BibTeX

The accurate identification of internet traffic is crucial for network management. However, the use of encryption techniques and constant changes in network protocols make it difficult to extract useful features for traffic classification. Additionally, there may be limited data availability and a lack of diversity within the dataset, which poses further challenges. To address these issues, our research proposes a novel solution that uses an innovative data augmentation technique. This approach leverages the capabilities of LSTM networks to create synthetic data points that closely resemble real traffic data. By doing so, we can significantly enrich the dataset used for training and improve classification efficiency. We conducted thorough experiments to validate our approach and found that combining LSTM-generated data with actual traffic data leads to notable improvements in classification efficiency. We demonstrated the effectiveness of our methodology using academic and commercial datasets. Our classifier, trained on the generated data, showed a performance boost of 6%. Moreover, when classifying with only half of the time, thus utilizing half of the signal, our approach achieved a notable 4% improvement compared to the original classifier. The inclusion of augmented samples within the training set led to a noticeable improvement in both accuracy and F1-score. These findings compellingly demonstrate our data augmentation strategy’s practical utility and efficiency in earlier prediction with improved performance for encrypted traffic classification systems.

@article{hajaj2024art,
  title = {The Art of Time-Bending: Data Augmentation and Early Prediction for Efficient Traffic Classification},
  author = {Hajaj, Chen and Aharon, Porat and Dubin, Ran and Dvir, Amit},
  journal = {Expert Systems with Applications},
  volume = {252},
  pages = {124166},
  year = {2024},
  publisher = {Elsevier},
  doi = {10.1016/j.eswa.2024.124166}
}

Copy BibTeX

E-tourism websites offer users a vast array of travel destinations and opportunities, necessitating tools that enable destination comparison and intelligent search capabilities. One key requirement for such tools is the ability to measure the similarity between destinations. Over the years, various similarity measurement techniques have been proposed, including user-based and content-based approaches. However, many of these techniques require data preparation or prior domain knowledge from experts. In contrast, this study proposes an innovative approach that requires no prior domain knowledge of flight destinations or their relationships, and utilizes only readily available data. Our approach draws upon concepts from image recognition and natural language processing (NLP) to extract hidden aspects of destinations. Using data from a flight-search website as a testbed, we analyze similarity metrics based on state-of-the-art methods for image recognition, NLP, and product-network analysis. We then compare these metrics to those obtained by human subjects. Our findings suggest that no single method dominates in all aspects, leading us to propose a hybrid method that leverages the strengths of each. The proposed method can be readily applied to measure product similarity in other domains.

@article{goldstein2024measuring,
  title = {Measuring Flight-Destination Similarity: A Multidimensional Approach},
  author = {Goldstein, Anat and Hajaj, Chen},
  journal = {Expert Systems with Applications},
  volume = {238},
  pages = {121802},
  year = {2024},
  publisher = {Elsevier},
  doi = {10.1016/j.eswa.2023.121802}
}

Copy BibTeX

Speaker Change Detection (SCD) is the task of segmenting an input audio-recording according to speaker interchanges. Nowadays, many applications, such as Speaker Diarization (SD) or automatic vocal transcription, depend on this segmentation task. In this paper, we focus on the essential task of the SD problem, the audio segmenting process, and suggest a solution for the SCD problem, as well as the assignment of clustered speaker labels for the extracted segments, and applying the solution over two datasets: a commercial dataset in Hebrew and the ICSI Meeting Corpus. As such, we propose a hybrid framework for the SCD problem that is learned by textual information and speech signals and the meta-data features that can be extracted from them. Moreover, we demonstrate the negative correlation between an increase in the number of speakers in the training dataset and the influence on the overall diarization system’s performance, which is improved using our efficient SCD component. Finally, we show how our proposed hybrid framework remains robust compared to the ICSI Meeting Corpus, as the experimental evaluation’s training and testing is based on two languages.

@article{anidjar2021hybrid,
  title = {Hybrid Speech and Text Analysis Methods for Speaker Change Detection},
  author = {Anidjar, Or Haim and Lapidot, Itshak and Hajaj, Chen and Dvir, Amit and Gilad, Issachar},
  journal = {IEEE/ACM Transactions on Audio, Speech, and Language Processing},
  volume = {29},
  pages = {2324--2338},
  year = {2021},
  publisher = {IEEE},
  doi = {10.1109/taslp.2021.3093817}
}

Copy BibTeX

Machine learning (ML) techniques are increasingly common in security applications, such as malware and intrusion detection. However, ML models are often susceptible to evasion attacks, in which an adversary makes changes to the input (such as malware) in order to avoid being detected. A conventional approach to evaluate ML robustness to such attacks, as well as to design robust ML, is by considering simplified feature-space models of attacks, where the attacker changes ML features directly to effect evasion, while minimizing or constraining the magnitude of this change. We investigate the effectiveness of this approach to designing robust ML in the face of attacks that can be realized in actual malware (realizable attacks). We demonstrate that in the context of structure-based PDF malware detection, such techniques appear to have limited effectiveness, but they are effective with content-based detectors. In either case, we show that augmenting the feature space models with conserved features (those that cannot be unilaterally modified without compromising malicious functionality) significantly improves performance. Finally, we show that feature space models enable generalized robustness when faced with a variety of realizable attacks, as compared to classifiers which are tuned to be robust to a specific realizable attack.

@inproceedings{tong2019improving,
  url = {https://www.usenix.org/conference/usenixsecurity19/presentation/tong},
  title = {Improving Robustness of ML Classifiers Against Realizable Evasion Attacks Using Conserved Features},
  author = {Tong, Liang and Li, Bo and Hajaj, Chen and Xiao, Chaowei and Zhang, Ning and Vorobeychik, Yevgeniy},
  booktitle = {28th USENIX Security Symposium (USENIX Security 19)},
  pages = {285--302},
  year = {2019}
}

Copy BibTeX