Definition, governance and auditing of general-purpose AI systems, risks of generative AI, and AI used to audit AI.
Foundation models and other General-Purpose AI Systems (GPAIS) raise new scientific and societal questions. This line studies what makes a system general-purpose, how such systems can be evaluated, audited and governed, and which risks emerge when they are deployed at scale. I also serve on the Scientific Panel of independent experts that advises the European Commission on implementing the EU AI Act.
The line combines conceptual and technical work. On the conceptual side, with colleagues from the University of Granada and other institutions, I proposed a definition and taxonomy of GPAIS and discussed their societal implications and responsible governance, and later examined how evolutionary computation can help design and enrich them. On the technical side, it covers the risks of generative AI: the degradation caused by training on synthetic data (AI autophagy), toxic and harmful language, privacy leakage, and manipulated or fake content. It also studies how general-purpose models can be turned into tools for evaluating and auditing responsible AI systems. Guillermo Villate-Castillo’s ongoing doctoral thesis on toxicity detection and mitigation for AI alignment contributes to this line.
Main contributions
What GPAI is and how to govern it: a definition, taxonomy and responsible-governance framework for GPAIS (Triguero et al., 2024), a survey of how evolutionary computation can design and enrich them (Molina et al., 2026), and broader perspectives on trustworthy AI and on how society understands and perceives AI (Herrera et al., 2025; Lobo & Del Ser, 2026).
Risks of generative AI: analysis of AI autophagy, the degradation that follows when generative models are trained on AI-generated data (Nature Machine Intelligence) (Xing et al., 2025), and reviews of deepfake detection and of tampering detection in image data (Júnior et al., 2024; da Costa et al., 2020).
GPAI for evaluation and auditing: general-purpose AI as an auditor of responsible-AI requirements (Del Ser et al., 2027), a multimodal framework to compare specialized deep learning models with general-purpose LLMs (Nadeem et al., 2025), and an early study of ChatGPT for cross-lingual text classification (Zhao et al., 2023).
The rapid evolution of General-Purpose Artificial Intelligence Systems (GPAIS) has introduced new capabilities and significant governance challenges. As regulations impose stringent transparency, accountability and auditability requirements, evaluating responsible AI systems –namely, AI systems with built-in auditability and accountability mechanisms that support their lawful and ethical operation– becomes increasingly complex and resource-intensive. This paper examines GPAIS as enablers of scalable, evidence-based auditing of such systems, while recognising that their own risks and limitations constrain their suitability as auditors. Our primary contribution is GRAAF, a structured framework for integrating GPAIS into responsible AI evaluation pipelines, conceptualised as a multi-level fusion problem spanning data, models, agents, and audit judgements, and formalised through reliability-, disagreement- and uncertainty-aware fusion operators. GRAAF progresses from GPAIS as assistants to human auditors towards semi- and fully automated auditing agents. Through a taxonomy of use cases, operational metrics, and governance safeguards, we show how GPAIS can augment explanation generation, bias detection, robustness testing, and compliance monitoring. We also address risks of overreliance, adversarial manipulation, and audit capture, proposing mitigation strategies and a roadmap for safe adoption. We further instantiate GRAAF in a clinical triage drift-monitoring case study, in which a Large Language Model (LLM) auditor, meta-audited against deterministic reference rules, met all predefined success criteria while exhibiting limitations that preclude autonomous certification. Scalable AI auditing thus hinges on fusing heterogeneous evidence, multi-agent reasoning, and multimodal signals, with GPAIS as orchestrators; within layered oversight architectures, this fusion can make auditing more trustworthy and scalable without compromising human accountability.
@article{delser2026from,title={{From Assistants to Auditors: Opportunities and Challenges of Using General-Purpose AI for Auditing Responsible AI Systems}},author={Del Ser, Javier and Moreno-Sánchez, Pedro A. and Gómez, Emília and Llorca, David Fernández},journal={Information Fusion},volume={139},pages={104861},year={2027},doi={10.1016/j.inffus.2026.104861},}
2026
IEEE Transactions on Evolutionary Computation
Evolutionary Computation for the Design and Enrichment of General-Purpose Artificial Intelligence Systems: Survey and Prospects
D. Molina, J. Poyatos, Javier Del Ser, S. García, H. Ishibuchi, I. Triguero, and 3 more authors
IEEE Transactions on Evolutionary Computation, 2026
In Artificial Intelligence, there is an increasing demand for adaptive models capable of dealing with a diverse spectrum of learning tasks, surpassing the limitations of systems devised to cope with a single task. The recent emergence of General-Purpose Artificial Intelligence Systems (GPAIS) poses model configuration and adaptability challenges at far greater complexity scales than the optimal design of traditional Machine Learning models. Evolutionary Computation (EC) has been a useful tool for both the design and optimization of Machine Learning models, endowing them with the capability to configure and/or adapt themselves to the task under consideration. Therefore, their application to GPAIS is a natural choice. This paper aims to analyze the role of EC in the field of GPAIS, exploring the use of EC for their design or enrichment. We also match GPAIS properties to Machine Learning areas in which EC has had a notable contribution, highlighting recent milestones of EC for GPAIS. Furthermore, we discuss the challenges of harnessing the benefits of EC for GPAIS, presenting different strategies to both design and improve GPAIS with EC, covering tangential areas, identifying research niches, and outlining potential research directions for EC and GPAIS.
@article{molina2025evolutionary,title={{Evolutionary Computation for the Design and Enrichment of General-Purpose Artificial Intelligence Systems: Survey and Prospects}},author={Molina, D. and Poyatos, J. and Del Ser, Javier and García, S. and Ishibuchi, H. and Triguero, I. and Xue, B. and Yao, X. and Herrera, F.},journal={IEEE Transactions on Evolutionary Computation},year={2026},pages={925--941},number={3},volume={30},doi={10.1109/tevc.2025.3530096},}
Can Transformative AI Shape a New Age for Our Civilization?
Understanding and Perceiving AI
Jesús L. Lobo and Javier Del Ser
In Can Transformative AI Shape a New Age for Our Civilization?, 2026
Throughout human history, certain technological advancements have marked profound turning points, pushing our civilization forward and reshaping the very fabric of society. These innovations are not just incremental improvements, but monumental leaps that have transformed how humans live, work, and interact with the world. The purpose of this section is to highlight some milestones in human history that are widely accepted, which have represented a real leap forward for our civilization (depicted in Figure 2.1 ). We start this tour with the aim of understanding what can be expected from modern AI, and whether it can be genuinely included in this compendium of transformative milestones.
@incollection{lobo2026understanding,title={{Understanding and Perceiving AI}},author={Lobo, Jesús L. and Del Ser, Javier},booktitle={Can Transformative AI Shape a New Age for Our Civilization?},pages={6--28},publisher={Crc Press},year={2026},doi={10.1201/9781003727071-2},}
Artificial Intelligence Review
A Systematic Review of Toxicity in Large Language Models: Definitions, Datasets, Detectors, Detoxification Methods and Challenges
Guillermo Villate-Castillo, Javier Del Ser, and Borja Sanz
The emergence of the transformer architecture has ushered in a new era of possibilities, showcasing remarkable capabilities in generative tasks exemplified by models like GPT4o, Claude 3, and Llama 3. However, these advancements come with a caveat: predominantly trained on data gleaned from social media platforms, these systems inadvertently perpetuate societal biases and toxicity. Recognizing the paramount importance of AI Safety and Alignment, our study embarks on a thorough exploration through a comprehensive literature review focused on toxic language. Delving into various definitions, detection methodologies, and mitigation strategies, we aim to shed light on the complexities of this issue. Our study primarily focuses on transformer-based architectures, with deep learning methods serving as a crucial baseline for comparison in our exploration of toxic language detection and mitigation. Through our investigation, we uncover a multitude of challenges inherent in toxicity mitigation and detection models. These challenges range from inherent biases and generalization issues to the necessity for standardized definitions of toxic language and the quality assurance of dataset annotations. Furthermore, we emphasize the significance of transparent annotation processes, resolution of annotation disagreements, and the enhancement of Large Language Models (LLMs) robustness. Additionally, we advocate for the creation of standardized benchmarks to gauge the effectiveness of toxicity mitigation and detection methods. Addressing these challenges is not just imperative, but pivotal in advancing the development of safer and more ethically aligned AI systems.
@article{villatecastillo2026systematic,title={{A Systematic Review of Toxicity in Large Language Models: Definitions, Datasets, Detectors, Detoxification Methods and Challenges}},author={Villate-Castillo, Guillermo and Del Ser, Javier and Sanz, Borja},journal={Artificial Intelligence Review},volume={59},number={11},year={2026},pages={240},doi={10.1007/s10462-026-11548-z},}
2025
Annual ADIA Lab Transactions in Data Science and Finance
Trustworthy Artificial Intelligence: Nature, Requirements, Regulation, and Emerging Discussions
Francisco Herrera, Andres Herrera, Javier Del Ser, Enrique Herrera‐Viedma, and Marcos López Prado
In Annual ADIA Lab Transactions in Data Science and Finance, 2025
Trustworthy artificial intelligence (TAI) is a framework and critical goal for developing and practically deploying reliable and ethical AI systems, properties essential for user confidence, societal acceptance, and responsible use. In this chapter, we explore TAI from three perspectives: (1) the importance of TAI and why it is essential; (2) a brief analysis of TAI requirements and characteristics as proposed by the high-level expert group on artificial intelligence commissioned by the European Commission and the American National Institute of Standards and Technology, preceded by a short discussion on trust in AI; and (3) an analysis of TAI’s role within emerging discussions on AI regulation and governance. We also review contributions from the literature, offering reflections that complement this chapter’s holistic vision of TAI, from theory to practice. TAI is expected to play a crucial role in the development of responsible AI systems, ensuring auditability and accountability throughout the design, development, and utilization phases, thereby creating ethical, reliable, and safe AI systems from a technological perspective.
@incollection{herrera2025trustworthy,title={{Trustworthy Artificial Intelligence: Nature, Requirements, Regulation, and Emerging Discussions}},author={Herrera, Francisco and Herrera, Andres and Del Ser, Javier and Herrera‐Viedma, Enrique and de Prado, Marcos López},booktitle={Annual ADIA Lab Transactions in Data Science and Finance},pages={317--342},publisher={World Scientific},year={2025},doi={10.1142/9789819813049_0012},}
Nature Machine Intelligence
On the Caveats of AI Autophagy
X. Xing, F. Shi, J. Huang, Y. Wu, Y. Nan, S. Zhang, and 5 more authors
Generative artificial intelligence (AI) technologies and large models are producing realistic outputs across various domains, such as images, text, speech and music. Creating these advanced generative models requires significant resources, particularly large and high-quality datasets. To minimize training expenses, many algorithm developers use data created by the models themselves as a cost-effective training solution. However, not all synthetic data effectively improve model performance, necessitating a strategic balance in the use of real versus synthetic data to optimize outcomes. Currently, the previously well-controlled integration of real and synthetic data is becoming uncontrollable. The widespread and unregulated dissemination of synthetic data online leads to the contamination of datasets traditionally compiled through web scraping, now mixed with unlabelled synthetic data. This trend, known as the AI autophagy phenomenon, suggests a future where generative AI systems may increasingly consume their own outputs without discernment, raising concerns about model performance, reliability and ethical implications. What will happen if generative AI continuously consumes itself without discernment? What measures can we take to mitigate the potential adverse effects? To address these research questions, this Perspective examines the existing literature, delving into the consequences of AI autophagy, analysing the associated risks and exploring strategies to mitigate its impact. Our aim is to provide a comprehensive perspective on this phenomenon advocating for a balanced approach that promotes the sustainable development of generative AI technologies in the era of large models. With widespread generation and availability of synthetic data, AI systems are increasingly trained on their own outputs, leading to various technical and ethical challenges. The authors analyse this development and discuss measures to mitigate the potential adverse effects of ‘AI eating itself’.
@article{xing2025caveats,title={{On the Caveats of AI Autophagy}},author={Xing, X. and Shi, F. and Huang, J. and Wu, Y. and Nan, Y. and Zhang, S. and Fang, Y. and Roberts, M. and Schönlieb, C.-B. and Del Ser, Javier and Yang, G.},journal={Nature Machine Intelligence},volume={7},pages={172--180},year={2025},number={2},doi={10.1038/s42256-025-00984-1},}
IEEE Transactions on Big Data
A Multi-Modal Assessment Framework for Comparison of Specialized Deep Learning and General-Purpose Large Language Models
Mohammad Nadeem, Shahab Saquib Sohail, Dag Øivind Madsen, Ahmed Ibrahim Alzahrani, Javier Del Ser, and Khan Muhammad
Recent years have witnessed tremendous advancements in Al tools (e.g., ChatGPT, GPT-4, and Bard), driven by the growing power, reasoning, and efficiency of Large Language Models (LLMs). LLMs have been shown to excel in tasks ranging from poem writing and coding to essay generation and puzzle solving. Despite their proficiency in general queries, specialized tasks such as metaphor understanding and fake news detection often require finely tuned models, posing a comparison challenge with specialized Deep Learning (DL). We propose an assessment framework to compare task-specific intelligence with general-purpose LLMs on suicide and depression tendency identification. For this purpose, we trained two DL models on a suicide and depression detection dataset, followed by testing their performance on a test set. Afterward, the same test dataset is used to evaluate the performance of four LLMs (GPT-3.5, GPT-4, Google Bard, and MS Bing) using four classification metrics. The BERT-based DL model performed the best among all, with a testing accuracy of 94.61%, while GPT-4 was the runner-up with accuracy 92.5%. Results demonstrate that LLMs do not outperform the specialized DL models but are able to achieve comparable performance, making them a decent option for downstream tasks without specialized training. However, LLMs outperformed specialized models on the reduced dataset.
@article{nadeem2025multi,title={{A Multi-Modal Assessment Framework for Comparison of Specialized Deep Learning and General-Purpose Large Language Models}},author={Nadeem, Mohammad and Sohail, Shahab Saquib and Madsen, Dag Øivind and Alzahrani, Ahmed Ibrahim and Del Ser, Javier and Muhammad, Khan},journal={IEEE Transactions on Big Data},volume={11},number={3},pages={1001--1012},year={2025},doi={10.1109/tbdata.2025.3536937},}
Neurocomputing
A Collaborative Content Moderation Framework for Toxicity Detection Based on Multitask Neural Networks and Conformal Estimates of Annotation Disagreement
Guillermo Villate-Castillo, Javier Del Ser, and Borja Sanz
Content moderation typically combines the efforts of human moderators and machine learning models. However, these systems often rely on data where significant disagreement occurs during moderation, reflecting the subjective nature of toxicity perception. Rather than dismissing this disagreement as noise, we interpret it as a valuable signal that highlights the inherent ambiguity of the content,an insight missed when only the majority label is considered. In this work, we introduce a novel content moderation framework that emphasizes the importance of capturing annotation disagreement. Our approach uses multitask learning, where toxicity classification serves as the primary task and annotation disagreement is addressed as an auxiliary task. Additionally, we leverage uncertainty estimation techniques, specifically Conformal Prediction, to account for both the ambiguity in comment annotations and the model’s inherent uncertainty in predicting toxicity and disagreement.The framework also allows moderators to adjust thresholds for annotation disagreement, offering flexibility in determining when ambiguity should trigger a review. We demonstrate that our joint approach enhances model performance, calibration, and uncertainty estimation, while offering greater parameter efficiency and improving the review process in comparison to single-task methods.
@article{villatecastillo2025collaborative,title={{A Collaborative Content Moderation Framework for Toxicity Detection Based on Multitask Neural Networks and Conformal Estimates of Annotation Disagreement}},author={Villate-Castillo, Guillermo and Del Ser, Javier and Sanz, Borja},journal={Neurocomputing},volume={647},pages={130542},year={2025},doi={10.1016/j.neucom.2025.130542},}
IEEE/ACM International Conference on Big Data Computing, Applications and Technologies (BDCAT)
Reflect, Reason, Rephrase (R³-Detox): An in-Context Learning Approach to Text Detoxification
Guillermo Villate-Castillo, Javier Del Ser, and Borja Sanz
In IEEE/ACM International Conference on Big Data Computing, Applications and Technologies (BDCAT), 2025
Traditional content moderation, while effective in reducing toxicity through content removal or censoring, can discourage user participation by making them feel restricted or unfairly targeted, especially in nuanced discussions. Text detoxification offers a more constructive alternative by rephrasing offensive language into respectful forms. We propose R3-Detox, a Reflect-Reason-Rephrase framework that structures detoxification into three steps within a single prompt. The model identifies potentially toxic elements guided by Shapley values to reduce fabricated predictions, evaluates overall toxicity, and then revises the text to eliminate toxicity while retaining meaning. We augment three offensive text paraphrasing datasets (ParaDetox, Parallel Detoxification, APPDIA) with explicit detoxification reasoning. Evaluated with in-context learning, R3-Detox outperforms state-of-the-art methods, including instruction following models.
@inproceedings{villatecastillo2025reflect,title={{Reflect, Reason, Rephrase (R³-Detox): An in-Context Learning Approach to Text Detoxification}},author={Villate-Castillo, Guillermo and Del Ser, Javier and Sanz, Borja},booktitle={IEEE/ACM International Conference on Big Data Computing, Applications and Technologies (BDCAT)},pages={1--10},year={2025},doi={10.1145/3773276.3774282},}
Machine Learning and Knowledge Extraction
Membership Inference Attacks Fueled by Few-Shot Learning to Detect Privacy Leakage and Address Data Integrity
Daniel Jiménez-López, Nuria Rodríguez-Barroso, M. Victoria Luzón, Javier Del Ser, and Francisco Herrera
Deep learning models have an intrinsic privacy issue as they memorize parts of their training data, creating a privacy leakage. Membership inference attacks (MIAs) exploit this to obtain confidential information about the data used for training, aiming to steal information. They can be repurposed as a measurement of data integrity by inferring whether the data were used to train a machine learning model. While state-of-the-art attacks achieve significant privacy leakage, their requirements render them infeasible, hindering their use as practical tools to assess the magnitude of the privacy risk. Moreover, the most appropriate evaluation metric of MIA, the true positive rate at a low false positive rate, lacks interpretability. We claim that the incorporation of few-shot learning techniques into the MIA field and a suitable qualitative and quantitative privacy evaluation measure should resolve these issues. In this context, our proposal is twofold. We propose a few-shot learning-based MIA, termed the FeS-MIA model, which eases the evaluation of the privacy breach of a deep learning model by significantly reducing the number of resources required for this purpose. Furthermore, we propose an interpretable quantitative and qualitative measure of privacy, referred to as the Log-MIA measure. Jointly, these proposals provide new tools to assess privacy leakages and to ease the evaluation of the training data integrity of deep learning models, i.e., to analyze the privacy breach of a deep learning model. Experiments carried out with MIA over image classification and language modeling tasks, and a comparison to the state of the art, show that our proposals excel in identifying privacy leakages in a deep learning model with little extra information.
@article{jimenezlopez2025membership,title={{Membership Inference Attacks Fueled by Few-Shot Learning to Detect Privacy Leakage and Address Data Integrity}},author={Jiménez-López, Daniel and Rodríguez-Barroso, Nuria and Luzón, M. Victoria and Del Ser, Javier and Herrera, Francisco},journal={Machine Learning and Knowledge Extraction},volume={7},number={2},pages={43},year={2025},doi={10.3390/make7020043},}
2024
Information Fusion
General Purpose Artificial Intelligence Systems (GPAIS): Properties, Definition, Taxonomy, Societal Implications and Responsible Governance
I. Triguero, D. Molina, J. Poyatos, Javier Del Ser, and F. Herrera
@article{triguero2024general,title={{General Purpose Artificial Intelligence Systems (GPAIS): Properties, Definition, Taxonomy, Societal Implications and Responsible Governance}},author={Triguero, I. and Molina, D. and Poyatos, J. and Del Ser, Javier and Herrera, F.},journal={Information Fusion},volume={103},pages={102135},year={2024},doi={10.1016/j.inffus.2023.102135},}
Expert Systems
A Review of Deep Learning-Based Approaches for Deepfake Content Detection
Leandro Aparecido Passos Júnior, Danilo Samuel Jodas, Kelton Augusto Pontara Costa, Luis Antonio Souza Jr, Douglas Rodrigues, Javier Del Ser, and 2 more authors
Recent advancements in deep learning generative models have raised concerns as they can create highly convincing counterfeit images and videos. This poses a threat to people’s integrity and can lead to social instability. To address this issue, there is a pressing need to develop new computational models that can efficiently detect forged content and alert users to potential image and video manipulations. This paper presents a comprehensive review of recent studies for deepfake content detection using deep learning‐based approaches. We aim to broaden the state‐of‐the‐art research by systematically reviewing the different categories of fake content detection. Furthermore, we report the advantages and drawbacks of the examined works, and prescribe several future directions towards the issues and shortcomings still unsolved on deepfake detection.
@article{junior2024review,title={{A Review of Deep Learning-Based Approaches for Deepfake Content Detection}},author={Júnior, Leandro Aparecido Passos and Jodas, Danilo Samuel and da Costa, Kelton Augusto Pontara and de Souza Jr, Luis Antonio and Rodrigues, Douglas and Del Ser, Javier and Camacho, David and Papa, João Paulo},journal={Expert Systems},volume={41},number={8},year={2024},pages={e13570},doi={10.1111/exsy.13570},}
IEEE/CAA Journal of Automatica Sinica
A Tutorial on Federated Learning from Theory to Practice: Foundations, Software Frameworks, Exemplary Use Cases, and Selected Trends
M. Victoria Luzón, Nuria Rodríguez-Barroso, Alberto Argente-Garrido, Daniel Jiménez-López, Jose M. Moyano, Javier Del Ser, and 2 more authors
When data privacy is imposed as a necessity, Federated learning (FL) emerges as a relevant artificial intelligence field for developing machine learning (ML) models in a distributed and decentralized environment. FL allows ML models to be trained on local devices without any need for centralized data transfer, thereby reducing both the exposure of sensitive data and the possibility of data interception by malicious third parties. This paradigm has gained momentum in the last few years, spurred by the plethora of real-world applications that have leveraged its ability to improve the efficiency of distributed learning and to accommodate numerous participants with their data sources. By virtue of FL, models can be learned from all such distributed data sources while preserving data privacy. The aim of this paper is to provide a practical tutorial on FL, including a short methodology and a systematic analysis of existing software frameworks. Furthermore, our tutorial provides exemplary cases of study from three complementary perspectives: i) Foundations of FL, describing the main components of FL, from key elements to FL categories; ii) Implementation guidelines and exemplary cases of study, by systematically examining the functionalities provided by existing software frameworks for FL deployment, devising a methodology to design a FL scenario, and providing exemplary cases of study with source code for different ML approaches; and iii) Trends, shortly reviewing a non-exhaustive list of research directions that are under active investigation in the current FL landscape. The ultimate purpose of this work is to establish itself as a referential work for researchers, developers, and data scientists willing to explore the capabilities of FL in practical applications.
@article{luzon2024tutorial,title={{A Tutorial on Federated Learning from Theory to Practice: Foundations, Software Frameworks, Exemplary Use Cases, and Selected Trends}},author={Luzón, M. Victoria and Rodríguez-Barroso, Nuria and Argente-Garrido, Alberto and Jiménez-López, Daniel and Moyano, Jose M. and Del Ser, Javier and Ding, Weiping and Herrera, Francisco},journal={IEEE/CAA Journal of Automatica Sinica},volume={11},number={4},pages={824--850},year={2024},doi={10.1109/jas.2024.124215},}
IEEE International Conference on Fuzzy Systems (FUZZ-IEEE)
Defense Strategy Against Byzantine Attacks in Federated Machine Learning: Developments Towards Explainability
Nuria Rodríguez-Barroso, Javier Del Ser, M. Victoria Luzón, and Francisco Herrera
In IEEE International Conference on Fuzzy Systems (FUZZ-IEEE), 2024
The rise of high-risk AI systems has led to escalating concerns, prompting regulatory efforts such as the recently approved EU AI Act. In this context, the development of responsible AI systems is crucial. To this end, trustworthy AI techniques aim at requirements (including transparency, privacy awareness and fairness) that contribute to the development of responsible, robust and safe AI systems. Among them, Federated Learning (FL) has emerged as a key approach to safeguarding data privacy while enabling the collaborative training of AI models. However, FL is prone to adversarial attacks, particularly byzantine attacks, which aim to modify the behavior of the model. This work addresses this issue by proposing an eXplainable and Impartial Dynamic Defense against Byzantine Attacks (XI-DDaBA). This defense mechanism relies on robust aggregation operators and filtering techniques to mitigate the effects of adversarial attacks in FL, while providing explanations for its decisions and ensuring that clients with poor data quality are not discriminated. Experimental simulations are discussed to assess the performance of XI-DDaBA against other baselines from the literature, and to showcase its provided explanations. Overall, XI-DDaBA aligns with the need for responsible AI systems in high-risk collaborative learning scenarios through the explainable and impartial provision of robustness against attacks.
@inproceedings{rodriguezbarroso2024defense,title={{Defense Strategy Against Byzantine Attacks in Federated Machine Learning: Developments Towards Explainability}},author={Rodríguez-Barroso, Nuria and Del Ser, Javier and Luzón, M. Victoria and Herrera, Francisco},booktitle={IEEE International Conference on Fuzzy Systems (FUZZ-IEEE)},pages={1--8},year={2024},doi={10.1109/fuzz-ieee60900.2024.10611769},}
2023
Neurocomputing
ChatAgri: Exploring Potentials of ChatGPT on Cross-Linguistic Agricultural Text Classification
Biao Zhao, Weiqiang Jin, Javier Del Ser, and Guang Yang
In the era of sustainable smart agriculture, a vast amount of agricultural news text is posted online, accumulating significant agricultural knowledge. To efficiently access this knowledge, effective text classification techniques are urgently needed. Deep learning approaches, such as fine-tuning strategies on pre-trained language models (PLMs), have shown remarkable performance gains. Nonetheless, these methods face several complex challenges, including limited agricultural training data, poor domain transferability (especially across languages), and complex and expensive deployment of large models. Inspired by the success of recent ChatGPT models (e.g., GPT-3.5, GPT-4), this work explores the potential of applying ChatGPT in the field of agricultural informatization. Various crucial factors, such as prompt construction, answer parsing, and different ChatGPT variants, are thoroughly investigated to maximize its capabilities. A preliminary comparative study is conducted, comparing ChatGPT with PLMs-based fine-tuning methods and PLMs-based prompt-tuning methods. Empirical results demonstrate that ChatGPT effectively addresses the mentioned research challenges and bottlenecks, making it an ideal solution for agricultural text classification. Moreover, ChatGPT achieves comparable performance to existing PLM-based fine-tuning methods, even without fine-tuning on agricultural data samples. We hope this preliminary study could inspire the emergence of a general-purpose AI paradigm for agricultural text processing.
@article{zhao2023chatagri,title={{ChatAgri: Exploring Potentials of ChatGPT on Cross-Linguistic Agricultural Text Classification}},author={Zhao, Biao and Jin, Weiqiang and Del Ser, Javier and Yang, Guang},journal={Neurocomputing},volume={557},pages={126708},year={2023},doi={10.1016/j.neucom.2023.126708},}
Journal of Ambient Intelligence and Humanized Computing
A Survey on Extremism Analysis Using Natural Language Processing: Definitions, Literature Review, Trends, and Challenges
Javier Torregrosa, Gema Bello-Orgaz, Eugenio Martinez-Cámara, Javier Del Ser, and David Camacho
Journal of Ambient Intelligence and Humanized Computing, 2023
Extremism has grown as a global problem for society in recent years, especially after the apparition of movements such as jihadism. This and other extremist groups have taken advantage of different approaches, such as the use of Social Media, to spread their ideology, promote their acts and recruit followers. The extremist discourse, therefore, is reflected on the language used by these groups. Natural language processing (NLP) provides a way of detecting this type of content, and several authors make use of it to describe and discriminate the discourse held by these groups, with the final objective of detecting and preventing its spread. Following this approach, this survey aims to review the contributions of NLP to the field of extremism research, providing the reader with a comprehensive picture of the state of the art of this research area. The content includes a first conceptualization of the term extremism, the elements that compose an extremist discourse and the differences with other terms. After that, a review description and comparison of the frequently used NLP techniques is presented, including how they were applied, the insights they provided, the most frequently used NLP software tools, descriptive and classification applications, and the availability of datasets and data sources for research. Finally, research questions are approached and answered with highlights from the review, while future trends, challenges and directions derived from these highlights are suggested towards stimulating further research in this exciting research area.
@article{torregrosa2022survey,title={{A Survey on Extremism Analysis Using Natural Language Processing: Definitions, Literature Review, Trends, and Challenges}},author={Torregrosa, Javier and Bello-Orgaz, Gema and Martinez-Cámara, Eugenio and Del Ser, Javier and Camacho, David},journal={Journal of Ambient Intelligence and Humanized Computing},pages={9869--9905},publisher={Springer Berlin Heidelberg},year={2023},number={8},volume={14},doi={10.1007/s12652-021-03658-z},}
IEEE Conference on Artificial Intelligence (CAI)
The Right to Be Forgotten in Artificial Intelligence: Issues, Approaches, Limitations and Challenges
Jesús L. Lobo, Sergio Gil-López, and Javier Del Ser
In IEEE Conference on Artificial Intelligence (CAI), 2023
The Right To Be Forgotten is widely conceived as a fundamental principle of the human being. It has become a subject of capital importance in domains where sensitive information is collected from individuals, requiring the provision of monitoring, governance and audit tools to control where such information is used. Artificial Intelligence models are not an exception to this statement: since they are learned from data, this fundamental right should allow individuals to have their personal information erased from AI-based systems. However, the application of this right is not straightforward: what does erasing mean in the context of a model learned from data? Is it just a matter of removing the concerned data and retraining the models? This manuscript provides a brief overview of these and more issues, proposing a desiderata for technical advances noted in this direction, and outlining research directions for prospective studies.
@inproceedings{lobo2023right,title={{The Right to Be Forgotten in Artificial Intelligence: Issues, Approaches, Limitations and Challenges}},author={Lobo, Jesús L. and Gil-López, Sergio and Del Ser, Javier},booktitle={IEEE Conference on Artificial Intelligence (CAI)},pages={179--180},year={2023},doi={10.1109/cai54212.2023.00085},}
2022
2022 International Joint Conference on Neural Networks (IJCNN)
Efficient Fake News Detection Using Bagging Ensembles of Bidirectional Echo State Networks
Javier Del Ser, Miren Nekane Bilbao, Ibai Laña, Khan Muhammad, and David Camacho
In 2022 International Joint Conference on Neural Networks (IJCNN), 2022
The dissemination of fake news is one of the most concerning issues in current digital media platforms, originating from the quick and easy spread of unverified information therethrough. Consequently, intense research efforts have been invested towards automating the process of identifying fake news from textual data by means of Artificial Intelligence methods. Among the manifold approaches proposed for this purpose to date, a large fraction of studies have examined the performance of modern deep neural network architectures, mostly relying on pretrained word embeddings and neural processing modules of diverse kind. Unfortunately, such sophisticated Deep Learning methods often require intense computational efforts for training. In this work we explore a novel approach based on randomization-based recurrent neural networks. Specifically, our proposal consists of a weighted ensemble of bidirectional Echo State Networks learned from word sequences processed through pretrained embeddings. Experiments over two fake news detection datasets reveal that competitive detection statistics are obtained by our proposed approach when compared to shallow learning and avant-garde Deep Learning models, but at a dramatically less computational complexity in their training phase.
@inproceedings{delser2022efficient,title={{Efficient Fake News Detection Using Bagging Ensembles of Bidirectional Echo State Networks}},author={Del Ser, Javier and Bilbao, Miren Nekane and Laña, Ibai and Muhammad, Khan and Camacho, David},booktitle={2022 International Joint Conference on Neural Networks (IJCNN)},year={2022},pages={1--7},doi={10.1109/ijcnn55064.2022.9892331},}
2020
Applied Soft Computing
A Critical Literature Survey and Prospects on Tampering and Anomaly Detection in Image Data
Kelton A. P. Costa, Joao P. Papa, Leandro A. Passos, Danilo Colombo, Javier Del Ser, Khan Muhammad, and 1 more author
Concernings related to image security have increased in the last years. One of the main reasons relies on the replacement of conventional photography to digital images, once the development of new technologies for image processing, as much as it has helped in the evolution of many new techniques in forensic studies, it also provided tools for image tampering. In this context, many companies and researchers devoted many efforts towards methods for detecting such tampered images, mostly aided by autonomous intelligent systems. Therefore, this work focuses on introducing a rigorous survey contemplating the state-of-the-art literature on computer-aided tampered image detection using machine learning techniques, as well as evolutionary computation, neural networks, fuzzy logic, Bayesian reasoning, among others. Besides, it also contemplates anomaly detection methods in the context of images due to the intrinsic relation between anomalies and tampering. Moreover, it aims at recent and in-depth researches relevant to the context of image tampering detection, performing a survey over more than 100 works related to the subject, spanning across different themes related to image tampering detection. Finally, a critical analysis is performed over this comprehensive compilation of literature, yielding some research opportunities and discussing some challenges in an attempt to align future efforts of the community with the niches and gaps remarked in this exciting field.
@article{dacosta2020critical,title={{A Critical Literature Survey and Prospects on Tampering and Anomaly Detection in Image Data}},author={da Costa, Kelton A. P. and Papa, Joao P. and Passos, Leandro A. and Colombo, Danilo and Del Ser, Javier and Muhammad, Khan and de Albuquerque, Victor Hugo C.},journal={Applied Soft Computing},pages={106727},year={2020},volume={97},doi={10.1016/j.asoc.2020.106727},}
2018
International Symposium on Intelligent and Distributed Computing
On the Design and Tuning of Machine Learning Models for Language Toxicity Classification in Online Platforms
Macej Rybinski, William Miller, Javier Del Ser, Miren Nekane Bilbao, and Jose F. Aldana-Montes
In International Symposium on Intelligent and Distributed Computing, 2018
@inproceedings{rybinski2018design,title={{On the Design and Tuning of Machine Learning Models for Language Toxicity Classification in Online Platforms}},author={Rybinski, Macej and Miller, William and Del Ser, Javier and Bilbao, Miren Nekane and Aldana-Montes, Jose F.},booktitle={International Symposium on Intelligent and Distributed Computing},series={Studies in Computational Intelligence},pages={329--343},year={2018},doi={10.1007/978-3-319-99626-4_29},}