Camelia Lemnaru
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Camelia Lemnaru has a total of 36 co-authors. Here are the top co-authors by number of shared publications:
- Rodica Potolea
Department of Computer Science Technical University of Cluj-Napoca Cluj-Napoca Romania
17 shared publications - Mihaela Dînşoreanu
Department of Computer Science, Technical University of Cluj-Napoca, Cluj-Napoca, Romania
5 shared publications - Mihai Surdeanu
Department of Computer Science University of Arizona Tucson USA
3 shared publications - Vlad Andrei Negru
Department of Computer Science Technical University of Cluj-Napoca Cluj-Napoca Romania
3 shared publications - Robert Vacareanu
Department of Computer Science Technical University of Cluj-Napoca Cluj-Napoca Romania
2 shared publications Rodica Potolea
Technical University of Cluj-Napoca
2 shared publications
Latest Publications
- 3 August 2026
- Published by Springer Nature
- p. 603-617
Towards Concept-Based Explanations in Vision–Language Models
- 3 August 2026
- Published by Springer Nature
- p. 603-617
Towards Concept-Based Explanations in Vision–Language Models
- Open Access
Understanding Cultural Alignment in Multilingual LLMs via Natural Debate Statements
In this work we investigate the sociocultural values learned by large language models (LLMs). We introduce a novel open-access dataset, Sociocultural Statements, constructed from natural debate statements using a multi-step methodology. The dataset is synthetically labeled to enable the quantization of sociocultural norms and beliefs that LLMs exhibit in their responses to these statements, according to the Hofstede cultural dimensions. We verify the accuracy of synthetic labels using human quality control on a representative sample. We conduct a comparative analysis between two groups of LLMs developed in different countries (U.S. and China), and use as a comparative baseline patterns observed in human measurements. Using this new dataset and the analysis above, we found that culturally-distinct LLMs reflect the values and norms of the countries in which they were developed, highlighting their inability to adapt to the sociocultural backgrounds of their users.
Open AccessUnderstanding Cultural Alignment in Multilingual LLMs via Natural Debate Statements
In this work we investigate the sociocultural values learned by large language models (LLMs). We introduce a novel open-access dataset, Sociocultural Statements, constructed from natural debate statements using a multi-step methodology. The dataset is synthetically labeled to enable the quantization of sociocultural norms and beliefs that LLMs exhibit in their responses to these statements, according to the Hofstede cultural dimensions. We verify the accuracy of synthetic labels using human quality control on a representative sample. We conduct a comparative analysis between two groups of LLMs developed in different countries (U.S. and China), and use as a comparative baseline patterns observed in human measurements. Using this new dataset and the analysis above, we found that culturally-distinct LLMs reflect the values and norms of the countries in which they were developed, highlighting their inability to adapt to the sociocultural backgrounds of their users.
- Open Access
- 26 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 110528-110546
Unveiling Hidden Structures: Multi-Platform IoT Malware Analysis Using Graph Embeddings
This paper explores the use of control-flow information for detecting IoT malware across three different platforms. We propose a pipeline which extracts Control-Flow Graphs (CFGs) from program binaries, projects them onto an embedding space, then computes the similarity between the programs, in an attempt to explore whether this approach could help identify malware families across different platforms. We explore three different methods for generating graph embeddings – at node, edge and graph level – and find the latter to produce the best discrimination between malware families across different platforms. Moreover, we propose a computationally efficient method for computing graph similarities for graphs with multiple connected components, which achieves accuracy on par with the cubic Hungarian Algorithm matching approach. The evaluations performed on a dataset comprising of clean and malware samples belonging to 18 different families yielded that our CFG embedding similarity approach can easily identify most malware family samples belonging to the same family, for the same platform, obtaining AUC scores above 90%. The method shows strong potential for discriminating between malware families across different platforms. We are currently exploring cross-platform limitations and improvement points, and we share preliminary insights from our current work towards the end of the paper.
- 26 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 110528-110546
Open AccessUnveiling Hidden Structures: Multi-Platform IoT Malware Analysis Using Graph Embeddings
This paper explores the use of control-flow information for detecting IoT malware across three different platforms. We propose a pipeline which extracts Control-Flow Graphs (CFGs) from program binaries, projects them onto an embedding space, then computes the similarity between the programs, in an attempt to explore whether this approach could help identify malware families across different platforms. We explore three different methods for generating graph embeddings – at node, edge and graph level – and find the latter to produce the best discrimination between malware families across different platforms. Moreover, we propose a computationally efficient method for computing graph similarities for graphs with multiple connected components, which achieves accuracy on par with the cubic Hungarian Algorithm matching approach. The evaluations performed on a dataset comprising of clean and malware samples belonging to 18 different families yielded that our CFG embedding similarity approach can easily identify most malware family samples belonging to the same family, for the same platform, obtaining AUC scores above 90%. The method shows strong potential for discriminating between malware families across different platforms. We are currently exploring cross-platform limitations and improvement points, and we share preliminary insights from our current work towards the end of the paper.
- 26 May 2025
- Published by Springer Nature
- p. 75-99
QdGF: A Framework Which Leverages Training-Time Information for Enhanced Multilingual-Multitask Classification
- 26 May 2025
- Published by Springer Nature
- p. 75-99
QdGF: A Framework Which Leverages Training-Time Information for Enhanced Multilingual-Multitask Classification
- Open Access
MorphNLI: A Stepwise Approach to Natural Language Inference Using Text Morphing
We introduce MorphNLI, a modular step-by-step approach to natural language inference (NLI). When classifying the premise-hypothesis pairs into {entailment, contradiction, neutral}, we use a language model to generate the necessary edits to incrementally transform (i.e., morph) the premise into the hypothesis. Then, using an off-the-shelf NLI model we track how the entailment progresses with these atomic changes, aggregating these intermediate labels into a final output. We demonstrate the advantages of our proposed method particularly in realistic cross-domain settings, where our method always outperforms strong baselines with improvements up to 12.6% (relative). Further, our proposed approach is explainable as the atomic edits can be used to understand the overall NLI label.
Open AccessMorphNLI: A Stepwise Approach to Natural Language Inference Using Text Morphing
We introduce MorphNLI, a modular step-by-step approach to natural language inference (NLI). When classifying the premise-hypothesis pairs into {entailment, contradiction, neutral}, we use a language model to generate the necessary edits to incrementally transform (i.e., morph) the premise into the hypothesis. Then, using an off-the-shelf NLI model we track how the entailment progresses with these atomic changes, aggregating these intermediate labels into a final output. We demonstrate the advantages of our proposed method particularly in realistic cross-domain settings, where our method always outperforms strong baselines with improvements up to 12.6% (relative). Further, our proposed approach is explainable as the atomic edits can be used to understand the overall NLI label.
- Open Access
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 6938-6953
MorphNLI: A Stepwise Approach to Natural Language Inference Using Text Morphing
Vlad Andrei Negru, Robert Vacareanu, Camelia Lemnaru, Mihai Surdeanu, Rodica Potolea. Findings of the Association for Computational Linguistics: NAACL 2025. 2025.
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 6938-6953
Open AccessMorphNLI: A Stepwise Approach to Natural Language Inference Using Text Morphing
Vlad Andrei Negru, Robert Vacareanu, Camelia Lemnaru, Mihai Surdeanu, Rodica Potolea. Findings of the Association for Computational Linguistics: NAACL 2025. 2025.
- Open Access
- 1 January 2024
- Published by Association for Computational Linguistics (ACL)
- p. 52-62
Prompting Fairness: Learning Prompts for Debiasing Large Language Models
Andrei-Victor Chisca, Andrei-Cristian Rad, Camelia Lemnaru. Proceedings of the Fourth Workshop on Language Technology for Equality, Diversity, Inclusion. 2024.
- 1 January 2024
- Published by Association for Computational Linguistics (ACL)
- p. 52-62
Open AccessPrompting Fairness: Learning Prompts for Debiasing Large Language Models
Andrei-Victor Chisca, Andrei-Cristian Rad, Camelia Lemnaru. Proceedings of the Fourth Workshop on Language Technology for Equality, Diversity, Inclusion. 2024.
- 22 November 2023
- Published by Springer Nature
- p. 147-161
Unsupervised Clustering and Explainable AI for Unveiling Behavioral Variations Across Time in Home-Appliance Generated Data
- 22 November 2023
- Published by Springer Nature
- p. 147-161
Unsupervised Clustering and Explainable AI for Unveiling Behavioral Variations Across Time in Home-Appliance Generated Data
- 22 November 2023
- Published by Springer Nature
- p. 547-552
Multitask, Cross-Lingual Recipe Classification Using Joint Fine-Tuning Mechanisms
- 22 November 2023
- Published by Springer Nature
- p. 547-552
Multitask, Cross-Lingual Recipe Classification Using Joint Fine-Tuning Mechanisms
- Open Access
A Comparative Analysis between Efficient Attention Mechanisms for Traffic Forecasting without Structural Priors
Dot-product attention is a powerful mechanism for capturing contextual information. Models that build on top of it have acclaimed state-of-the-art performance in various domains, ranging from sequence modelling to visual tasks. However, the main bottleneck is the construction of the attention map, which is quadratic with respect to the number of tokens in the sequence. Consequently, efficient alternatives have been developed in parallel, but it was only recently that their performances were compared and contrasted. This study performs a comparative analysis between some efficient attention mechanisms in the context of a purely attention-based spatio-temporal forecasting model used for traffic prediction. Experiments show that these methods can reduce the training times by up to 28% and the inference times by up to 31%, while the performance remains on par with the baseline.
Open AccessA Comparative Analysis between Efficient Attention Mechanisms for Traffic Forecasting without Structural Priors
Dot-product attention is a powerful mechanism for capturing contextual information. Models that build on top of it have acclaimed state-of-the-art performance in various domains, ranging from sequence modelling to visual tasks. However, the main bottleneck is the construction of the attention map, which is quadratic with respect to the number of tokens in the sequence. Consequently, efficient alternatives have been developed in parallel, but it was only recently that their performances were compared and contrasted. This study performs a comparative analysis between some efficient attention mechanisms in the context of a purely attention-based spatio-temporal forecasting model used for traffic prediction. Experiments show that these methods can reduce the training times by up to 28% and the inference times by up to 31%, while the performance remains on par with the baseline.