Emil Constantine Lupu
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
Emil Constantine Lupu has a total of 152 co-authors. Here are the top co-authors by number of shared publications:
- Morris Sloman
Imperial College London
31 shared publications Naranker Dulay
13 shared publications- Alberto Schaeffer-Filho
UFRGS, Porto Alegre, Brazil
11 shared publications Alessandra Russo
Imperial College London
11 shared publications- Sye Loong Keoh
University of Glasgow
11 shared publications - Erisa Karafili
University of Southampton, Southampton, United Kingdom
10 shared publications
Latest Publications
- 1 December 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (12) , 22966-22983
C ausal O nline L earning of Safe Regions in Cloud Radio Access Networks
- 1 December 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (12) , 22966-22983
C ausal O nline L earning of Safe Regions in Cloud Radio Access Networks
- 13 October 2026
- Published by Springer Nature
- p. 414-433
Multiagent Incident Response Planning with Code Models
- 13 October 2026
- Published by Springer Nature
- p. 414-433
Multiagent Incident Response Planning with Code Models
- Open Access
Causal Online Learning of Safe Regions in Cloud Radio Access Networks
Cloud radio access networks (RANs) enable cost-effective management of mobile networks by dynamically scaling their capacity on demand. However, deploying adaptive controllers to implement such dynamic scaling in operational networks is challenging due to the risk of breaching service agreements and operational constraints. To mitigate this challenge, we present a novel method for learning the safe operating region of the RAN, i.e., the set of resource allocations and network configurations for which its specification is fulfilled. The method, which we call (C)ausal (O)nline (L)earning, operates in two online phases: an inference phase and an intervention phase. In the first phase, we passively observe the RAN to infer an initial safe region via causal inference and Gaussian process regression. In the second phase, we gradually expand this region through interventional Bayesian learning. We prove that COL ensures that the learned region is safe with a specified probability and that it converges to the full safe region under standard conditions. We experimentally validate COL on a 5G testbed. The results show that COL quickly learns the safe region while incurring low operational cost and being up to 10x more sample-efficient than current state-of-the-art methods for safe learning.
Open AccessCausal Online Learning of Safe Regions in Cloud Radio Access Networks
Cloud radio access networks (RANs) enable cost-effective management of mobile networks by dynamically scaling their capacity on demand. However, deploying adaptive controllers to implement such dynamic scaling in operational networks is challenging due to the risk of breaching service agreements and operational constraints. To mitigate this challenge, we present a novel method for learning the safe operating region of the RAN, i.e., the set of resource allocations and network configurations for which its specification is fulfilled. The method, which we call (C)ausal (O)nline (L)earning, operates in two online phases: an inference phase and an intervention phase. In the first phase, we passively observe the RAN to infer an initial safe region via causal inference and Gaussian process regression. In the second phase, we gradually expand this region through interventional Bayesian learning. We prove that COL ensures that the learned region is safe with a specified probability and that it converges to the full safe region under standard conditions. We experimentally validate COL on a 5G testbed. The results show that COL quickly learns the safe region while incurring low operational cost and being up to 10x more sample-efficient than current state-of-the-art methods for safe learning.
- Open Access
Hallucination-Resistant Security Planning with a Large Language Model
Large language models (LLMs) are promising tools for supporting security management tasks, such as incident response planning. However, their unreliability and tendency to hallucinate remain significant challenges. In this paper, we address these challenges by introducing a principled framework for using an LLM as decision support in security management. Our framework integrates the LLM in an iterative loop where it generates candidate actions that are checked for consistency with system constraints and lookahead predictions. When consistency is low, we abstain from the generated actions and instead collect external feedback, e.g., by evaluating actions in a digital twin. This feedback is then used to refine the candidate actions through in-context learning (ICL). We prove that this design allows to control the hallucination risk by tuning the consistency threshold. Moreover, we establish a bound on the regret of ICL under certain assumptions. To evaluate our framework, we apply it to an incident response use case where the goal is to generate a response and recovery plan based on system logs. Experiments on four public datasets show that our framework reduces recovery times by up to 30% compared to frontier LLMs.
Open AccessHallucination-Resistant Security Planning with a Large Language Model
Large language models (LLMs) are promising tools for supporting security management tasks, such as incident response planning. However, their unreliability and tendency to hallucinate remain significant challenges. In this paper, we address these challenges by introducing a principled framework for using an LLM as decision support in security management. Our framework integrates the LLM in an iterative loop where it generates candidate actions that are checked for consistency with system constraints and lookahead predictions. When consistency is low, we abstain from the generated actions and instead collect external feedback, e.g., by evaluating actions in a digital twin. This feedback is then used to refine the candidate actions through in-context learning (ICL). We prove that this design allows to control the hallucination risk by tuning the consistency threshold. Moreover, we establish a bound on the regret of ICL under certain assumptions. To evaluate our framework, we apply it to an incident response use case where the goal is to generate a response and recovery plan based on system logs. Experiments on four public datasets show that our framework reduces recovery times by up to 30% compared to frontier LLMs.
- 26 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Vehicular Technology Magazine
- Vol. 21 (1) , 27-37
Leveraging Adverse Weather for Enhanced LiDAR Spoofing in Autonomous Driving: Challenges and Opportunities
Lidar-based 3D perception systems are critical for autonomous vehicle (AV) navigation, yet they remain vulnerable to spoofing attacks that can create false detections (ghost objects) or hide real obstacles. Despite significant advances in object detection, existing methods remain highly susceptible to adversarial attacks. Furthermore, existing research has largely overlooked the impact of weather conditions on both attacks and defenses. No existing study provides a systematic analysis on how the rain effect affects spoofing and hiding attacks. Motivated by this critical gap, we propose a novel rain-aware threat model in this article, focusing on lidar spoofing attacks involving object insertion (ghost objects) and removal (hiding attacks). We review state-of-the-art attack implementations and emphasize how rain increases attack feasibility and stealth, enabling attackers to achieve effective spoofing with significantly fewer perturbed points. We call this a reduced attack budget (AB). Formally, we define the AB as the minimal number of spoofed points (and corresponding laser returns) needed to meet attack success (ghost insertion: high-confidence false positives; object hiding: low-confidence false negatives).
- 26 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Vehicular Technology Magazine
- Vol. 21 (1) , 27-37
Leveraging Adverse Weather for Enhanced LiDAR Spoofing in Autonomous Driving: Challenges and Opportunities
Lidar-based 3D perception systems are critical for autonomous vehicle (AV) navigation, yet they remain vulnerable to spoofing attacks that can create false detections (ghost objects) or hide real obstacles. Despite significant advances in object detection, existing methods remain highly susceptible to adversarial attacks. Furthermore, existing research has largely overlooked the impact of weather conditions on both attacks and defenses. No existing study provides a systematic analysis on how the rain effect affects spoofing and hiding attacks. Motivated by this critical gap, we propose a novel rain-aware threat model in this article, focusing on lidar spoofing attacks involving object insertion (ghost objects) and removal (hiding attacks). We review state-of-the-art attack implementations and emphasize how rain increases attack feasibility and stealth, enabling attackers to achieve effective spoofing with significantly fewer perturbed points. We call this a reduced attack budget (AB). Formally, we define the AB as the minimal number of spoofed points (and corresponding laser returns) needed to meet attack success (ghost insertion: high-confidence false positives; object hiding: low-confidence false negatives).
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21, 1
Adaptive Network Security Policies via Belief Aggregation and Rollout
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21, 1
Open AccessAdaptive Network Security Policies via Belief Aggregation and Rollout
- Open Access
Gen-Review: A Large-scale Dataset of AI-Generated (and Human-written) Peer Reviews
How does the progressive embracement of Large Language Models (LLMs) affect scientific peer reviewing? This multifaceted question is fundamental to the effectiveness -- as well as to the integrity -- of the scientific process. Recent evidence suggests that LLMs may have already been tacitly used in peer reviewing, e.g., at the 2024 International Conference of Learning Representations (ICLR). Furthermore, some efforts have been undertaken in an attempt to explicitly integrate LLMs in peer reviewing by various editorial boards (including that of ICLR'25). To fully understand the utility and the implications of LLMs' deployment for scientific reviewing, a comprehensive relevant dataset is strongly desirable. Despite some previous research on this topic, such dataset has been lacking so far. We fill in this gap by presenting GenReview, the hitherto largest dataset containing LLM-written reviews. Our dataset includes 81K reviews generated for all submissions to the 2018--2025 editions of the ICLR by providing the LLM with three independent prompts: a negative, a positive, and a neutral one. GenReview is also linked to the respective papers and their original reviews, thereby enabling a broad range of investigations. To illustrate the value of GenReview, we explore a sample of intriguing research questions, namely: if LLMs exhibit bias in reviewing (they do); if LLM-written reviews can be automatically detected (so far, they can); if LLMs can rigorously follow reviewing instructions (not always) and whether LLM-provided ratings align with decisions on paper acceptance or rejection (holds true only for accepted papers). GenReview can be accessed at the following link: https://anonymous.4open.science/r/gen_review.
Open AccessGen-Review: A Large-scale Dataset of AI-Generated (and Human-written) Peer Reviews
How does the progressive embracement of Large Language Models (LLMs) affect scientific peer reviewing? This multifaceted question is fundamental to the effectiveness -- as well as to the integrity -- of the scientific process. Recent evidence suggests that LLMs may have already been tacitly used in peer reviewing, e.g., at the 2024 International Conference of Learning Representations (ICLR). Furthermore, some efforts have been undertaken in an attempt to explicitly integrate LLMs in peer reviewing by various editorial boards (including that of ICLR'25). To fully understand the utility and the implications of LLMs' deployment for scientific reviewing, a comprehensive relevant dataset is strongly desirable. Despite some previous research on this topic, such dataset has been lacking so far. We fill in this gap by presenting GenReview, the hitherto largest dataset containing LLM-written reviews. Our dataset includes 81K reviews generated for all submissions to the 2018--2025 editions of the ICLR by providing the LLM with three independent prompts: a negative, a positive, and a neutral one. GenReview is also linked to the respective papers and their original reviews, thereby enabling a broad range of investigations. To illustrate the value of GenReview, we explore a sample of intriguing research questions, namely: if LLMs exhibit bias in reviewing (they do); if LLM-written reviews can be automatically detected (so far, they can); if LLMs can rigorously follow reviewing instructions (not always) and whether LLM-provided ratings align with decisions on paper acceptance or rejection (holds true only for accepted papers). GenReview can be accessed at the following link: https://anonymous.4open.science/r/gen_review.
- Open Access
Incident Response Planning Using a Lightweight Large Language Model with Reduced Hallucination
Timely and effective incident response is key to managing the growing frequency of cyberattacks. However, identifying the right response actions for complex systems is a major technical challenge. A promising approach to mitigate this challenge is to use the security knowledge embedded in large language models (LLMs) to assist security operators during incident handling. Recent research has demonstrated the potential of this approach, but current methods are mainly based on prompt engineering of frontier LLMs, which is costly and prone to hallucinations. We address these limitations by presenting a novel way to use an LLM for incident response planning with reduced hallucination. Our method includes three steps: fine-tuning, information retrieval, and lookahead planning. We prove that our method generates response plans with a bounded probability of hallucination and that this probability can be made arbitrarily small at the expense of increased planning time under certain assumptions. Moreover, we show that our method is lightweight and can run on commodity hardware. We evaluate our method on logs from incidents reported in the literature. The experimental results show that our method a) achieves up to 22% shorter recovery times than frontier LLMs and b) generalizes to a broad range of incident types and response actions.
Open AccessIncident Response Planning Using a Lightweight Large Language Model with Reduced Hallucination
Timely and effective incident response is key to managing the growing frequency of cyberattacks. However, identifying the right response actions for complex systems is a major technical challenge. A promising approach to mitigate this challenge is to use the security knowledge embedded in large language models (LLMs) to assist security operators during incident handling. Recent research has demonstrated the potential of this approach, but current methods are mainly based on prompt engineering of frontier LLMs, which is costly and prone to hallucinations. We address these limitations by presenting a novel way to use an LLM for incident response planning with reduced hallucination. Our method includes three steps: fine-tuning, information retrieval, and lookahead planning. We prove that our method generates response plans with a bounded probability of hallucination and that this probability can be made arbitrarily small at the expense of increased planning time under certain assumptions. Moreover, we show that our method is lightweight and can run on commodity hardware. We evaluate our method on logs from incidents reported in the literature. The experimental results show that our method a) achieves up to 22% shorter recovery times than frontier LLMs and b) generalizes to a broad range of incident types and response actions.
- Open Access
- 5 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 84057-84080
Exploring Research and Tools in AI Security: A Systematic Mapping Study
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.
- 5 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 84057-84080
Open AccessExploring Research and Tools in AI Security: A Systematic Mapping Study
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
- 11 February 2025
- Published by AIP Publishing in APL Machine Learning
- Vol. 3 (1)
Nonideality-aware training makes memristive networks more robust to adversarial attacks
Neural networks are now deployed in a wide number of areas, from object classification to natural language systems. Implementations using analog devices such as memristors promise better power efficiency, potentially bringing these applications to a greater number of environments. However, such systems suffer from more frequent device faults, and overall, their exposure to adversarial attacks has not been studied extensively. In this work, we investigate how nonideality-aware training—a common technique to deal with physical nonidealities—affects adversarial robustness. We find that adversarial robustness is significantly improved, even with limited knowledge of what nonidealities will be encountered during test time.
- 11 February 2025
- Published by AIP Publishing in APL Machine Learning
- Vol. 3 (1)
Open AccessNonideality-aware training makes memristive networks more robust to adversarial attacks
Neural networks are now deployed in a wide number of areas, from object classification to natural language systems. Implementations using analog devices such as memristors promise better power efficiency, potentially bringing these applications to a greater number of environments. However, such systems suffer from more frequent device faults, and overall, their exposure to adversarial attacks has not been studied extensively. In this work, we investigate how nonideality-aware training—a common technique to deal with physical nonidealities—affects adversarial robustness. We find that adversarial robustness is significantly improved, even with limited knowledge of what nonidealities will be encountered during test time.