LNLin Ni

Lin Ni

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Lin Ni has a total of 14 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    BiTTP: Bidirectional Consistency-Enhanced Framework for Low-Resource APT Tactic and Technique Attribution

    by Lin Ni, Gang Yang, Xiang Peng

    Attributing advanced persistent threat (APT) activity to tactics and techniques in the MITRE ATT&CK knowledge base supports threat hunting and incident response, but annotated attack data are scarce. We present BiTTP, a consistency-verified self-training framework for low-resource APT tactic and technique attribution. An annotator labels unlabeled audit-trace subgraphs, and a pseudo-label is retained only when it passes consensus, confidence, and cross-view consistency gates. The last gate compares the encoded input subgraph with the encoded ATT&CK definition of the predicted technique in a bidirectionally aligned space. Bidirectional data generation, dual-view alignment, retrieval-grounded decoding, and a learned attention aggregator support this loop. On a generated benchmark and a controlled-environment audit-trace benchmark derived from TREC, BiTTP reaches a technique F1 of 0.960 and a tactic micro-F1 of 0.939. Under cross-dataset transfer, it maintains a tactic micro-F1 of 0.912. The largest gains occur under severe label scarcity: with 100 seed examples per tactic, BiTTP exceeds the F1 of strong fine-tuning with 300 seeds. Cross-view consistency is also an effective open-set rejection signal for a withheld technique and its semantically adjacent neighbor.

  • Open Access

    Attack Chain Reconstruction Method Based on LLM Prior-Knowledge Guidance and Heterogeneous Graph Reasoning

    by Lin Ni, Jinchuan Pei, Huimei Wang, Haoyu Chen

    In the face of multi-stage and cross-device complex network attacks, reconstructing the complete attack link from massive heterogeneous security logs is the core problem of security operation. In the existing methods, the traditional graph model lacks a deep understanding of the semantics of alerts, and large language models (LLMs) have the ability of textual reasoning, but have difficulty effectively using the topology between entities, and have difficulty uniformly representing and globally relating multi-source heterogeneous data. Aiming at the above problems, this paper proposes a heterogeneous graph attack chain reconstruction method based on LLM prior-knowledge guidance. Firstly, ontology schemas of six types of core security entities and four types of relationship types are designed for multi-source security logs, and a unified heterogeneous graph representation is constructed by integrating temporal association, spatial association and semantic association. Events scattered in different devices and time windows are correlated into structured views. Secondly, the domain-adaptive fine-tuned LLM is used to encode the deep semantics of the node attribute text, and it is fused with the structural features to make up for the shortcomings of traditional graph neural networks that only rely on shallow statistical features. On this basis, a heterogeneous graph attention mechanism guided by LLM prior knowledge is designed, the attention weight is modulated by the LLM’s semantic score of edge rationality, and candidate attack links are generated by multi-step path reasoning along high-confidence edges. Finally, the logical verification and attack confirmation of the candidate link were carried out step by step with the help of the chain-of-thought ability of LLM, and the confidence was calibrated by temperature scaling. Experiments on DARPA TC and other datasets show that the integrity rate of attack link reconstruction is 86.4%, the coverage rate of attack phase is 82.1%, and the F1 value of attack confirmation is 90.6%, which is significantly better than the methods based on a probabilistic graph model, Heterogeneous Graph Transformer and pure LLM hint engineering. This verifies the effectiveness of the cooperation between semantic understanding and structural reasoning.

  • Open Access

    Beyond Pattern Matching: A Cognitive-Driven Framework for DGA Detection via Dual-Perspective Anomaly Perception

    by Xiang Peng, Jun He, Lin Ni, Gang Yang

    Domain Generation Algorithms (DGAs) pose a persistent threat by enabling malware to dynamically generate numerous command-and-control domains, evading traditional blocklists. While machine learning-based detectors have achieved high accuracy, they operate as statistical pattern matchers and lack the human-like anomaly perception that enables security experts to intuitively recognize unnatural domains. This paper introduces CogNormDGA, a cognitive-driven framework that models normal domain characteristics from a defender’s perspective while also anticipating how attackers might exploit cognitive blind spots. Inspired by dual-process theory, CogNormDGA combines intuitive, pattern-based screening (System 1) with analytical, rule-based evaluation of phonotactic, morphological, and semantic violations (System 2). The cognitive principles of System 1 and System 2 are computationally realized as two distinct pathways: an Attentional Salience Network and a Linguistic Constraint Evaluator, respectively. The framework produces interpretable outputs via attention saliency maps and cognitive violation reports. Extensive experiments on 400,000 domains spanning 33 DGA families demonstrate that CogNormDGA achieves competitive detection performance (F1-score 0.941) while establishing a cognitive-driven detection paradigm that produces human-aligned explanations—a property critical for practical security. It shows promising results on low-entropy and novel DGA families. Human subject studies confirm strong alignment between the model’s internal explanations and expert reasoning. Furthermore, CogNormDGA is particularly effective against low-entropy DGA families that exploit cognitive blind spots. By bridging cognitive science and cybersecurity, our work offers an interpretable and human-aligned approach to threat detection, with promising resilience that requires further validation.

  • Open Access

    NeSySwarm-IDS: End-to-End Differentiable Neuro-Symbolic Logic for Privacy-Preserving Intrusion Detection in UAV Swarms

    by Gang Yang, Lin Ni, Tao Xia, Qinfang Shi, Jiajian Li

    Unmanned Aerial Vehicle (UAV) swarms operating in contested environments face a critical “semantic gap” between raw, high-velocity network traffic and high-level mission security constraints, compounded by the risk of privacy leakage during collaborative learning. Existing deep learning (DL)-based Network Intrusion Detection Systems (NIDSs) suffer from opacity, prohibitive resource consumption, and vulnerability to gradient leakage attacks in federated settings, while traditional rule-based systems fail to handle encrypted payloads and evolving attack patterns. To bridge this gap, we present NeSySwarm-IDS (Neuro-Symbolic Swarm Intrusion Detection System), an end-to-end differentiable neuro-symbolic framework that simultaneously achieves high accuracy, strong privacy guarantees, and built-in interpretability under resource constraints. NeSySwarm-IDS integrates an extremely lightweight 1D convolutional neural network with a differentiable Łukasiewicz fuzzy logic reasoner incorporating attack-specific rules. By aggregating only low-dimensional logic rule weights with calibrated differential privacy noise, we drastically reduce communication overhead while providing (ϵ,δ)-DP guarantees with negligible utility loss. Extensive experiments on the UAV-NIDD dataset and our self-collected dataset demonstrate that NeSySwarm-IDS achieves near-perfect detection accuracy, significantly outperforming traditional machine learning baselines despite using limited training data. A detailed case study on GPS spoofing confirms the interpretability of our approach, providing axiomatic explanations suitable for autonomous mission verification. These results establish that end-to-end neuro-symbolic learning can effectively bridge the semantic gap in UAV swarm security while ensuring privacy and interpretability, offering a practical pathway for deploying trustworthy AI in contested environments.


  • Top Cited Papers
    Open Access

    FedLTN-CubeSat: Neuro-Symbolic Federated Learning for Intrusion Detection in LEO CubeSat Constellations

    by Gang Yang, Lin Ni, Junfeng Geng, Xiang Peng

    Low Earth Orbit (LEO) mega-constellations are becoming the backbone of global communications, yet their cybersecurity remains critically under-addressed. Intrusion detection systems (IDSs) for such constellations face a unique trilemma of accuracy, efficiency, and interpretability under extreme SWaP-C (size, weight, power, and cost) constraints. We present FedLTN-CubeSat (FedLTN refers to Federated Logic Tensor Networks), a neuro-symbolic federated learning framework for intrusion detection in LEO CubeSat constellations. The framework first employs a lightweight spatio-temporal separable perception encoder to efficiently extract features from telemetry and IQ data, designed to operate within the computational budgets of resource-constrained on-board processors. These features feed into a differentiable first-order logic layer based on Logic Tensor Networks, which incorporates domain knowledge as logical axioms to guide learning and enhance interpretability. To enable collaborative learning across a constellation, FedLTN-CubeSat introduces an intra-orbit symbolic federated learning mechanism that aggregates only the logic-layer parameters via inter-satellite links, drastically reducing communication overhead while preserving data privacy. Furthermore, an orbit-adaptive predicate migration module transfers learned rules across different orbital configurations with minimal supervision, facilitating rapid deployment. We evaluate on two benchmarks: the CuCD-ID dataset (NASA NOS3 telemetry) and the STIN dataset (satellite-terrestrial integrated networks). FedLTN-CubeSat achieves 0.98 F1-score on CuCD-ID and 0.96 accuracy on STIN—significantly outperforming prior federated learning baselines (7% improvement) while incurring a minimal daily communication load per satellite. The framework also outputs interpretable decision traces grounded in logical axioms, enabling operators to understand and validate detections. Logical constraints improve detection of unseen attack variants by 25% over pure neural baselines.