DTDaniel Takabi

Daniel Takabi

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

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


Daniel Takabi has a total of 54 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • ReT-UAP: Rethinking multi-surrogate universal adversarial perturbations on vision-language pre-trained models

    by Ziyu Wang, Jian Li, Yiming Du, Rui Ning, Jiang Li, Daniel Takabi, Lusi Li

  • Open Access

    BioShield: A Context-Aware Firewall for Securing Bio-LLMs

    by Protiva Das, Sovon Chakraborty, Sidhant Narula, Lucas Nelson Potter, Xavier-Lewis Palmer, Pratip Rana, Daniel Takabi, Mohammad GhashemiGol

  • Open Access

    Toward Secure and Practical Machine Learning‐Based Access Control: A Framework With Real‐World Constraints and Adversarial Analysis

    by Olusesi Balogun, Mohammad GhasemiGol, Zhipeng Cai, Daniel Takabi

    Attribute‐Based Access Control (ABAC) frameworks coordinate access requests based on subject, object, and environment attributes, as well as policy rules, and are widely used in corporate security systems. Recently, machine learning has been applied to ABAC to address policy‐generation imbalances, misassigned privileges, and attribute leakages. However, existing MLBAC techniques do not consider the structural constraints and attribute interdependencies present in traditional ABAC systems. Moreover, these frameworks have not been extensively evaluated under black‐box attack scenarios. To address these gaps, we propose extensions to MLBAC that integrate structural constraints, attribute dynamism, and attribute weighting into the MLBAC objective function. Additionally, we study the behavior of these extended MLBAC models under black‐box adversarial attacks. We implemented the framework using five deep‐learning models: RNN, LSTM, Deep Belief Network (DBN), TabTransformer, and DeepFM, on both synthetic and real‐world datasets. Our findings show that the DBN model consistently achieves the best performance, while the black‐box adversarial attacks reveal general vulnerabilities across MLBAC systems. These outcomes highlight the need for more robust defense mechanisms in future research. Accordingly, we propose several potential mitigation strategies against black‐box attacks.

  • Not All, but the Right Ones: Energy-Guided Representation Learning for Incomplete Multiview Clustering

    by Ziyu Wang, Jian Li, Yiming Du, Rui Ning, Shuai Hao, Daniel Takabi, Lusi Li

    Incomplete multiview clustering (IMVC) aims to uncover shared cluster structures from data with partially missing views. Most existing approaches face a critical tradeoff: imputation-free methods struggle under high missingness, while imputation-based methods risk propagating errors from unreliable reconstructions. In contrast, we argue that not all missing views should be recovered-only the right, reliable ones matter. To this end, we propose energy-guided representation learning network (ERL-Net), a novel selective imputation framework that leverages energy-based modeling to adaptively guide feature imputation, fusion, and alignment. ERL-Net integrates four key components: 1) multiview feature extraction using view-specific autoencoders and a shared projection network; 2) energy-geometric graph encoding, which evaluates feature reliability using a learnable energy function and models interview dependencies via joint energy-geometric similarity, enabling instance-specific graph aggregation; 3) energy-gated imputation, which reconstructs missing views using global-to-view mappers, selectively retaining only low-energy (i.e., reliable) candidates; and 4) energy-weighted fusion and alignment, which integrates observed and imputed features into a unified representation, while enforcing semantic consistency across views at both the energy and representation levels. Extensive experiments demonstrate the superiority of ERL-Net, particularly under high missing ratios, achieving significant improvements over state-of-the-art methods across multiple benchmark datasets.

  • Open Access

    HarmNet: A Framework for Adaptive Multi-Turn Jailbreak Attacks on Large Language Models

    by Sidhant Narula, Javad Rafiei Asl, Mohammad Ghasemigol, Eduardo Blanco, Daniel Takabi

    Large Language Models (LLMs) remain vulnerable to multi-turn jailbreak attacks. We introduce HarmNet, a modular framework comprising ThoughtNet, a hierarchical semantic network; a feedback-driven Simulator for iterative query refinement; and a Network Traverser for real-time adaptive attack execution. HarmNet systematically explores and refines the adversarial space to uncover stealthy, high-success attack paths. Experiments across closed-source and open-source LLMs show that HarmNet outperforms state-of-the-art methods, achieving higher attack success rates. For example, on Mistral-7B, HarmNet achieves a 99.4% attack success rate, 13.9% higher than the best baseline. Index terms: jailbreak attacks; large language models; adversarial framework; query refinement.

  • Open Access

    Heterogeneous Graph Backdoor Attack

    by Jiawei Chen, Lusi Li, Daniel Takabi, Masha Sosonkina, Rui Ning

    Heterogeneous Graph Neural Networks (HGNNs) excel in modeling complex, multi-typed relationships across diverse domains, yet their vulnerability to backdoor attacks remains unexplored. To address this gap, we conduct the first investigation into the susceptibility of HGNNs to existing graph backdoor attacks, revealing three critical issues: (1) high attack budget required for effective backdoor injection, (2) inefficient and unreliable backdoor activation, and (3) inaccurate attack effectiveness evaluation. To tackle these issues, we propose the Heterogeneous Graph Backdoor Attack (HGBA), the first backdoor attack specifically designed for HGNNs, introducing a novel relation-based trigger mechanism that establishes specific connections between a strategically selected trigger node and poisoned nodes via the backdoor metapath. HGBA achieves efficient and stealthy backdoor injection with minimal structural modifications and supports easy backdoor activation through two flexible strategies: Self-Node Attack and Indiscriminate Attack. Additionally, we improve the ASR measurement protocol, enabling a more accurate assessment of attack effectiveness. Extensive experiments demonstrate that HGBA far surpasses multiple state-of-the-art graph backdoor attacks in black-box settings, efficiently attacking HGNNs with low attack budgets. Ablation studies show that the strength of HBGA benefits from our trigger node selection method and backdoor metapath selection strategy. In addition, HGBA shows superior robustness against node feature perturbations and multiple types of existing graph backdoor defense mechanisms. Finally, extension experiments demonstrate that the relation-based trigger mechanism can effectively extend to tasks in homogeneous graph scenarios, thereby posing severe threats to broader security-critical domains.

  • Open Access

    Exploring Research and Tools in AI Security: A Systematic Mapping Study

    by Sidhant Narula, Mohammad Ghasemigol, Javier Carnerero-Cano, Amanda Minnich, Emil Lupu, Daniel Takabi

    With the pervasive integration of artificial intelligence (AI) in various facets of modern technology, the importance of AI security has been thrust into the spotlight. The field is rapidly evolving, with new challenges and solutions emerging at a swift pace. However, the breadth and depth of AI security research have not been comprehensively mapped in recent times, presenting a crucial need for an extensive review and synthesis of existing literature. Given the increasing reliance on AI in critical domains such as healthcare, finance, and national security, ensuring the resilience and trustworthiness of these systems is imperative. This survey fulfills the pressing need for a structured and comprehensive overview of the current research landscape, enabling researchers to address emerging threats and vulnerabilities effectively. This paper presents a systematic mapping study (SMS), aimed at identifying and classifying the prevailing research topics, tools, and frameworks in the field of AI security. A total of 123 studies were meticulously selected and analyzed, leading to the identification of key metrics, tools, standards, and research themes that are currently shaping the landscape of AI security research. This effort not only aids in distilling the collective wisdom of the research community but also sets a firm foundation for future work in this critical area. The findings from this SMS will serve as an invaluable guide for researchers and practitioners alike, enabling them to navigate the complexities of AI security and fostering the development of innovative, robust security solutions. This study also highlights significant gaps in the current literature, thereby outlining potential directions for new research initiatives.


  • Open Access

    Problem-Centered Post-Secondary Computer Science Education: A Study of the Private Artificial Intelligence Curriculum

    by Golnoush Haddadian, Prajwal Panzade, Daniel Takabi, Min Kyu Kim

    In response to the demand for Artificial Intelligence (AI) experts, this study introduced a curriculum development initiative. The aim was to design and implement a Private AI curriculum to understand the computer science (CS) students’ evaluations of the curricular activities and their levels of interest and motivation. Twenty-five students, a mix of undergraduates and graduates, were recruited and a scaled-down version of the curriculum was implemented. A parallel mixed-methods approach was employed. The results reinforced the significance of problem-centered curricula in CS context. Students rated the curricular activities highly and demonstrated strong motivation; however, graduates expressed more favorable view of pairwise collaboration and reported higher self-efficacy. Analysis of coding problem-solving behaviors suggested less competent students often relied on trial-and-error, whereas more competent students employed systematic, forward problem-solving strategies. This study contributes to the field of CS by emphasizing the importance of problem-centered learning to prepare students for real-world AI challenges.

  • Utilizing Federated Learning and SHAP for Predictive Analysis in Smart Grid Security

    by Sushmitha Halli Sudhakara, Lida Haghnegahdar, Mohammad GhasemiGol, Daniel Takabi


  • BlindTuner: On Enhancement of Privacy-Preserving Fine-Tuning of Transformers Based on Homomorphic Encryption

    by Prajwal Panzade, Javad Rafiei Asl, Daniel Takabi, Zhipeng Cai

    Fine-tuning pretrained models has emerged as a pivotal technique in machine learning, especially in situations with limited, task-aligned training data. However, challenges surface when data sharing encounters obstacles due to stringent privacy regulations or user apprehension regarding personal data disclosure. Earlier works based on secure multiparty computation (SMC) and fully homomorphic encryption (FHE) for privacy-preserving machine learning (PPML) focused more on privacy-preserving inference than privacy-preserving training. While privacy-preserving transfer learning has made progress, efficiency and accuracy tradeoffs remain a key limitation. In response to this gap, we introduce BlindTuner, a system that enables privacy-preserving fine-tuning by training data efficient image transformers (DEiT) directly on homomorphically encrypted data. Additionally, BlindTuner performs privacy-preserving inference, extending its usability beyond training. Experimental results show that BlindTuner achieves accuracy comparable to unencrypted models while delivering training speedups between $1.16{\times }$ and $600{\times }$ over existing approaches. Furthermore, we explore its application in privacy-preserving federated fine-tuning, enabling solution for collaborative learning without compromising privacy.