Zongxiong Chen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Zongxiong Chen has a total of 54 co-authors. Here are the top co-authors by number of shared publications:
- Sonja Schimmler
Fraunhofer FOKUS, Berlin, Germany
10 shared publications - Derui Zhu
Technical University of Munich, Munich, Germany
5 shared publications Jiahui Geng
Mohamed bin Zayed University of Artificial Intelligence
4 shared publications- Manfred Hauswirth
Fraunhofer FOKUS, Berlin, Germany
4 shared publications - Alexander Pretschner
Technical University of Munich, Munich, Germany
2 shared publications - Chunming Rong
University of Stavanger
2 shared publications
Latest Publications
- Open Access
- 7 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 461-473
Stop When It Matters: Detectability-Guided Microbenchmarking for Performance Regression Testing
- 7 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 461-473
Open AccessStop When It Matters: Detectability-Guided Microbenchmarking for Performance Regression Testing
- Open Access
- 28 May 2026
- Published by Association for Computing Machinery (ACM)
- p. 959-964
Mixture of Orthogonal Decomposition for Semantically Generalisable Multimodal Deepfake Detection with Deepfake Manifold Disentanglement
Multimodal deepfakes have caused significant social issues and raised substantial security concerns. Generalisably detecting these multimodal deepfakes can be factored into two intricately intertwined subspaces, i.e., the semantic subspace and the forgery subspace. However, contemporary deepfake detection methods predominantly focus on pursuing generalisability within the forgery subspace, usually neglecting the influences of the semantic subspace. Additionally, existing deepfake detection algorithms categorise all deepfakes generated by various deepfake generation techniques under a single ''Fake'' category. This methodology may fail to account for the unique characteristics and extensive aspects of various deepfake generation techniques. To address all these challenges, we propose an innovative sparsely-gated Mixture of ORthogonal Decomposition (MoORD) deepfake detection model. Whilst the orthogonal decomposition parameter-efficient fine-tuning technique guarantees the generalisation performance of deepfake detection, the MoORD structure further enhances our model's semantic generalisability. Our MoORD model further learns product manifold representations to disentangle the ''Fake'' category into separate sub-categories of various deepfake generation techniques. Comprehensive experiments in 6 real-world deepfake detection datasets with 2 evaluation protocols demonstrate leading-edge generalisation performance in deepfake detection of our MoORD model over known and unknown semantic categories.
- 28 May 2026
- Published by Association for Computing Machinery (ACM)
- p. 959-964
Open AccessMixture of Orthogonal Decomposition for Semantically Generalisable Multimodal Deepfake Detection with Deepfake Manifold Disentanglement
Multimodal deepfakes have caused significant social issues and raised substantial security concerns. Generalisably detecting these multimodal deepfakes can be factored into two intricately intertwined subspaces, i.e., the semantic subspace and the forgery subspace. However, contemporary deepfake detection methods predominantly focus on pursuing generalisability within the forgery subspace, usually neglecting the influences of the semantic subspace. Additionally, existing deepfake detection algorithms categorise all deepfakes generated by various deepfake generation techniques under a single ''Fake'' category. This methodology may fail to account for the unique characteristics and extensive aspects of various deepfake generation techniques. To address all these challenges, we propose an innovative sparsely-gated Mixture of ORthogonal Decomposition (MoORD) deepfake detection model. Whilst the orthogonal decomposition parameter-efficient fine-tuning technique guarantees the generalisation performance of deepfake detection, the MoORD structure further enhances our model's semantic generalisability. Our MoORD model further learns product manifold representations to disentangle the ''Fake'' category into separate sub-categories of various deepfake generation techniques. Comprehensive experiments in 6 real-world deepfake detection datasets with 2 evaluation protocols demonstrate leading-edge generalisation performance in deepfake detection of our MoORD model over known and unknown semantic categories.
- Open Access
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 2148-2160
LLM4JMH: Studying the Use of LLMs for Generating Java Performance Microbenchmarks
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 2148-2160
Open AccessLLM4JMH: Studying the Use of LLMs for Generating Java Performance Microbenchmarks
- Open Access
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 8661-8664
NERdME: A Named Entity Recognition Dataset for Indexing Research Artifacts in Code Repositories
Existing scholarly information extraction (SIE) datasets focus on scientific papers and overlook implementation-level details in code repositories. README files describe datasets, source code, and other implementation-level artifacts, however, their free-form Markdown offers little semantic structure, making automatic information extraction difficult. To address this gap, NERdME is introduced: 200 manually annotated README files with over νm10000 labeled spans and 10 entity types. Baseline results using large language models and fine-tuned transformers show clear differences between paper-level and implementation-level entities, indicating the value of extending SIE benchmarks with entity types available in README files. A downstream entity-linking experiment was conducted to demonstrate that entities derived from READMEs can support artifact discovery and metadata integration.
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 8661-8664
Open AccessNERdME: A Named Entity Recognition Dataset for Indexing Research Artifacts in Code Repositories
Existing scholarly information extraction (SIE) datasets focus on scientific papers and overlook implementation-level details in code repositories. README files describe datasets, source code, and other implementation-level artifacts, however, their free-form Markdown offers little semantic structure, making automatic information extraction difficult. To address this gap, NERdME is introduced: 200 manually annotated README files with over νm10000 labeled spans and 10 entity types. Baseline results using large language models and fine-tuned transformers show clear differences between paper-level and implementation-level entities, indicating the value of extending SIE benchmarks with entity types available in README files. A downstream entity-linking experiment was conducted to demonstrate that entities derived from READMEs can support artifact discovery and metadata integration.
- Open Access
NFDI4DS Shared Tasks for Scholarly Document Processing
Shared tasks are powerful tools for advancing research through community-based standardised evaluation. As such, they play a key role in promoting findable, accessible, interoperable, and reusable (FAIR), as well as transparent and reproducible research practices. This paper presents an updated overview of twelve shared tasks developed and hosted under the German National Research Data Infrastructure for Data Science and Artificial Intelligence (NFDI4DS) consortium, covering a diverse set of challenges in scholarly document processing. Hosted at leading venues, the tasks foster methodological innovations and contribute open-access datasets, models, and tools for the broader research community, which are integrated into the consortium's research data infrastructure.
Open AccessNFDI4DS Shared Tasks for Scholarly Document Processing
Shared tasks are powerful tools for advancing research through community-based standardised evaluation. As such, they play a key role in promoting findable, accessible, interoperable, and reusable (FAIR), as well as transparent and reproducible research practices. This paper presents an updated overview of twelve shared tasks developed and hosted under the German National Research Data Infrastructure for Data Science and Artificial Intelligence (NFDI4DS) consortium, covering a diverse set of challenges in scholarly document processing. Hosted at leading venues, the tasks foster methodological innovations and contribute open-access datasets, models, and tools for the broader research community, which are integrated into the consortium's research data infrastructure.
- 15 June 2025
- Published by Springer Nature
- p. 233-245
GDCK: Efficient Large-Scale Graph Distillation Utilizing a Model-Free Kernelized Approach
- 15 June 2025
- Published by Springer Nature
- p. 233-245
GDCK: Efficient Large-Scale Graph Distillation Utilizing a Model-Free Kernelized Approach
- Open Access
- 11 April 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 39 (26) , 27428-27436
Internal Activation Revision: Safeguarding Vision Language Models Without Parameter Update
Warning: This paper contains offensive content that may disturb some readers. Vision-language models (VLMs) demonstrate strong multimodal capabilities but have been found to be more susceptible to generating harmful content compared to their backbone large language models (LLMs). Our investigation reveals that the integration of images significantly shifts the model's internal activations during the forward pass, diverging from those triggered by textual input. Moreover, the safety alignments of LLMs embedded within VLMs are not sufficiently robust to handle the activations discrepancies, making the models vulnerable to even the simplest jailbreaking attacks. To address this issue, we propose an internal activation revision approach that efficiently revises activations during generation, steering the model toward safer outputs. Our framework incorporates revisions at both the layer and head levels, offering control over the model's generation at varying levels of granularity. In addition, we explore three strategies for constructing positive and negative samples and two approaches for extracting revision vectors, resulting in different variants of our method. Comprehensive experiments demonstrate that the internal activation revision method significantly improves the safety of widely used VLMs, reducing attack success rates by an average of 48.94%, 34.34%, 43.92%, and 52.98% on SafeBench, Safe-Unsafe, Unsafe, and MM-SafetyBench, respectively, while minimally impacting model helpfulness.
- 11 April 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 39 (26) , 27428-27436
Open AccessInternal Activation Revision: Safeguarding Vision Language Models Without Parameter Update
Warning: This paper contains offensive content that may disturb some readers. Vision-language models (VLMs) demonstrate strong multimodal capabilities but have been found to be more susceptible to generating harmful content compared to their backbone large language models (LLMs). Our investigation reveals that the integration of images significantly shifts the model's internal activations during the forward pass, diverging from those triggered by textual input. Moreover, the safety alignments of LLMs embedded within VLMs are not sufficiently robust to handle the activations discrepancies, making the models vulnerable to even the simplest jailbreaking attacks. To address this issue, we propose an internal activation revision approach that efficiently revises activations during generation, steering the model toward safer outputs. Our framework incorporates revisions at both the layer and head levels, offering control over the model's generation at varying levels of granularity. In addition, we explore three strategies for constructing positive and negative samples and two approaches for extracting revision vectors, resulting in different variants of our method. Comprehensive experiments demonstrate that the internal activation revision method significantly improves the safety of widely used VLMs, reducing attack success rates by an average of 48.94%, 34.34%, 43.92%, and 52.98% on SafeBench, Safe-Unsafe, Unsafe, and MM-SafetyBench, respectively, while minimally impacting model helpfulness.
- 28 January 2025
- Published by Springer Nature
- p. 202-206
Converter: Enhancing Interoperability in Research Data Management
- 28 January 2025
- Published by Springer Nature
- p. 202-206
Converter: Enhancing Interoperability in Research Data Management
- Open Access
Internal Activation Revision: Safeguarding Vision Language Models Without Parameter Update
Vision-language models (VLMs) demonstrate strong multimodal capabilities but have been found to be more susceptible to generating harmful content compared to their backbone large language models (LLMs). Our investigation reveals that the integration of images significantly shifts the model's internal activations during the forward pass, diverging from those triggered by textual input. Moreover, the safety alignments of LLMs embedded within VLMs are not sufficiently robust to handle the activations discrepancies, making the models vulnerable to even the simplest jailbreaking attacks. To address this issue, we propose an \textbf{internal activation revision} approach that efficiently revises activations during generation, steering the model toward safer outputs. Our framework incorporates revisions at both the layer and head levels, offering control over the model's generation at varying levels of granularity. In addition, we explore three strategies for constructing positive and negative samples and two approaches for extracting revision vectors, resulting in different variants of our method. Comprehensive experiments demonstrate that the internal activation revision method significantly improves the safety of widely used VLMs, reducing attack success rates by an average of 48.94\%, 34.34\%, 43.92\%, and 52.98\% on SafeBench, Safe-Unsafe, Unsafe, and MM-SafetyBench, respectively, while minimally impacting model helpfulness.
Open AccessInternal Activation Revision: Safeguarding Vision Language Models Without Parameter Update
Vision-language models (VLMs) demonstrate strong multimodal capabilities but have been found to be more susceptible to generating harmful content compared to their backbone large language models (LLMs). Our investigation reveals that the integration of images significantly shifts the model's internal activations during the forward pass, diverging from those triggered by textual input. Moreover, the safety alignments of LLMs embedded within VLMs are not sufficiently robust to handle the activations discrepancies, making the models vulnerable to even the simplest jailbreaking attacks. To address this issue, we propose an \textbf{internal activation revision} approach that efficiently revises activations during generation, steering the model toward safer outputs. Our framework incorporates revisions at both the layer and head levels, offering control over the model's generation at varying levels of granularity. In addition, we explore three strategies for constructing positive and negative samples and two approaches for extracting revision vectors, resulting in different variants of our method. Comprehensive experiments demonstrate that the internal activation revision method significantly improves the safety of widely used VLMs, reducing attack success rates by an average of 48.94\%, 34.34\%, 43.92\%, and 52.98\% on SafeBench, Safe-Unsafe, Unsafe, and MM-SafetyBench, respectively, while minimally impacting model helpfulness.
- Open Access
NFDI4DSO: Towards a BFO Compliant Ontology for Data Science
The NFDI4DataScience (NFDI4DS) project aims to enhance the accessibility and interoperability of research data within Data Science (DS) and Artificial Intelligence (AI) by connecting digital artifacts and ensuring they adhere to FAIR (Findable, Accessible, Interoperable, and Reusable) principles. To this end, this poster introduces the NFDI4DS Ontology, which describes resources in DS and AI and models the structure of the NFDI4DS consortium. Built upon the NFDICore ontology and mapped to the Basic Formal Ontology (BFO), this ontology serves as the foundation for the NFDI4DS knowledge graph currently under development.
Open AccessNFDI4DSO: Towards a BFO Compliant Ontology for Data Science
The NFDI4DataScience (NFDI4DS) project aims to enhance the accessibility and interoperability of research data within Data Science (DS) and Artificial Intelligence (AI) by connecting digital artifacts and ensuring they adhere to FAIR (Findable, Accessible, Interoperable, and Reusable) principles. To this end, this poster introduces the NFDI4DS Ontology, which describes resources in DS and AI and models the structure of the NFDI4DS consortium. Built upon the NFDICore ontology and mapped to the Basic Formal Ontology (BFO), this ontology serves as the foundation for the NFDI4DS knowledge graph currently under development.