Intent-Based Networking (IBN) enables operators to specify high-level outcomes while the system translates these intents into concrete policies and configurations. As IBN deployments grow in scale, heterogeneity and dynamicity, ensuring continuous alignment between network behavior and user objectives becomes both essential and increasingly difficult. This paper provides a technical survey of assurance and conflict detection techniques in IBN, with the goal of improving reliability, robustness, and policy compliance. We first position our survey with respect to existing work. We then review current assurance mechanisms, including the use of AI, machine learning, and real-time monitoring for validating intent fulfillment. We also examine conflict detection methods across the intent lifecycle, from capture to implementation. In addition, we outline relevant standardization efforts and open-source tools that support IBN adoption. Finally, we discuss key challenges, such as AI/ML integration, generalization, and scalability, and present a roadmap for future research aimed at strengthening robustness of IBN frameworks.

IntroductionDeep learning has achieved strong results in histopathological diagnosis, but the lack of transparency in deep neural network models limits their clinical acceptance. Explainable Artificial Intelligence (XAI) offers a principled route to address this gap by quantifying the reliability and morphological coherence of model predictions in support of clinical decision-making.MethodsWe present an integrated, explainability-augmented deep learning framework for binary histopathological classification on the BreaKHis dataset. The pipeline combines knowledge-distilled InceptionV3 → DenseNet-121 training with three complementary local explanation methods—Gradient-weighted Class Activation Mapping (Grad-CAM), SHapley Additive exPlanations (SHAP), and Local Interpretable Model-agnostic Explanations (LIME). Per-image outputs are aggregated at the dataset level into the ExpiScore composite interpretability index, which captures salience reliability, cross-method coherence, and morphological alignment. Additional regression and decision-tree modules build a trust meta-model that estimates the probability of prediction errors and supports automated triage of clinical cases. Two operational modes are evaluated and kept strictly separate throughout: a fully-automatic mode (CNN classification with no human intervention) and a human-in-the-loop mode (meta-model routing of high-risk cases to expert review).ResultsIn fully-automatic mode, the distilled DenseNet-121 student achieves accuracy 0.9916, F1-score 0.9939, Area Under the receiver-operating-characteristic Curve (AUC) 0.9989, sensitivity 0.9939, and specificity 0.9866 on the 1,187-image held-out test set, with five-fold cross-validation (BreaKHis official folds) further indicating generalization stability (mean accuracy 0.8962 ± 0.0177; mean AUC 0.9598 ± 0.0094). In human-in-the-loop mode, a meta-model trained on XAI-derived features and independently evaluated on the held-out test set routes the 20% of cases with the highest predicted error probability to expert review, raising post-routing accuracy from 0.9898 to 0.9983 (+0.85 pp) and reducing false negatives by 83.3% under the assumption of correct expert resolution of all routed cases.DiscussionThe two regimes are not directly comparable and are reported separately throughout the paper. The fully-automatic figures characterize purely algorithmic performance; the human-inthe-loop figures characterize a clinical-decision-support workflow in which the framework prioritizes cases for expert review. Together they suggest that ExpiScore-driven routing carries discriminative predictive signal for error detection and supports a principled trade-off between predictive accuracy and interpretative transparency, providing held-out test evidence for the potential clinical utility of the proposed routing mechanism within this evaluation setting. External multi-cohort validation and formal calibration of the triage thresholds are required before clinical deployment.

Students with dyslexia face persistent challenges in accessing and processing written content, which can compromise learning opportunities and motivation. While recent advances in generative AI have shown promise in education, few systems explicitly address the needs of dyslexic learners. This paper presents DyslexIA, an AI-powered learning assistant that combines text adaptation, readability-aware fine-tuning, and structured scaffolding techniques to support inclusive learning. The system introduces several innovations: customizable interaction modes, fine-tuned models optimized for readability and active voice, and contextual as well as cumulative concept and mind maps. Moreover, DyslexIA embeds reflective scaffolding inspired by the ICAP framework and Improved Vee Questions, guiding learners beyond comprehension toward constructive engagement. Widely used accessibility features such as text-to-speech, speech-to-text, and customizable fonts are also integrated in a unified environment. Preliminary evaluation with about 25 students and educators shows measurable improvements in readability metrics and highlights the usefulness of summarization, concept mapping, and reflective prompts. These findings, although exploratory, suggest that DyslexIA can serve as a communication bridge between students and teachers and as a step toward empowering dyslexic learners through AI-enabled accessible and reflective learning.

Open-ended questions offer deep insights into students’ reasoning, but present persistent challenges in terms of grading consistency and bias. In this paper, we introduce Criterium, a teacher-assistant system designed to support fair, explainable, and scalable evaluation of open-ended student responses, particularly in the humanities. Unlike traditional automated grading solutions, Criterium places the teacher at the center of the evaluation loop, offering structured rubrics, prompt-based LLM scoring, and optional content retrieval to anchor judgments in curricular material. We present the system architecture, the underlying scoring model – which distinguishes minimum from advanced criteria – and results from a real-world pilot involving multiple classes and assignments. Criterium aims not to replace teacher judgment, but to enhance its objectivity and traceability in both formative and summative contexts.

The rapid evolution of AI has reshaped many educational practices, raising new questions about its role in supporting students with learning disorders. This work presents a systematic and interdisciplinary review on the use of artificial intelligence (AI) and generative AI (GAI) to support students with dyslexia in educational settings. Dyslexia is the most common disorder among people and, despite the rapid development of AI technologies in recent years, the connections between AI, dyslexia, and education remain underexplored, and comprehensive, up-to-date analyses are lacking. To address this gap, an extensive literature, published between 2018 and 2024, was analyzed, ultimately identifying n = 72 relevant studies, guided by the PRISMA methodology. These studies were classified according to a set of semantic categories carefully defined by the authors, allowing to map current research trends and identify emerging interdisciplinary research directions. This analysis shows that AI has been used to enhance detection, assistive support, and personalized learning for students with dyslexia, yet the evidence remains fragmented and experimental validation is limited. While GAI offers promising opportunities for content generation and interactive support, its presence in the current literature is still minimal. The review highlights existing achievements, identifies key challenges, and outlines future research avenues to advance inclusive, AI-supported learning for individuals with dyslexia.

The rapid advancement of generative artificial intelligence (GAI) has opened new pathways for enhancing educational technologies, particularly through the integration of learning analytics (LA). This paper explores the dynamic relationship between LA and GAI, assessing how their convergence could potentially transform educational outcomes. LA, which involves the measurement, collection, analysis, and reporting of data about learners and their contexts, serves as a crucial tool for understanding and improving learning processes. Combined with GAI, capable of creating content, simulating educational interactions, and generating personalized learning resources, the potential for tailored educational experiences is significantly expanded. This synergy not only enhances adaptive learning environments but also supports real-time decision-making, feedback, and intervention strategies. By examining various case studies and experimental implementations, the study highlights the efficacy of GAI in interpreting complex datasets provided by LA, thereby promoting an adaptive and responsive learning ecosystem. Additionally, ethical considerations, challenges, and future directions of this interdisciplinary approach are discussed to guide current and future applications. This paper proposes an exploratory study on the synergy between LA and GAI, using the AI4Design system, a web-based platform based on GAI for Design learning, as the framework where GAI complements LA. A case study is proposed using ChatGPT as the GAI engine. The first results show the power of the synergy GAI-LA.

Recent advances in image processing and artificial intelligence have significantly improved the analysis of histopathological images for breast cancer. Deep learning models applied to high-resolution histopathological images (Whole Slide Images, WSIs) are capable of capturing complex, multiscale morphological patterns, achieving high performance in classification, segmentation and risk stratification tasks. However, the increasing architectural complexity of these models has introduced major interpretability issues, limiting their reliability and adoption in sensitive medical image processing applications. Explainable Artificial Intelligence (XAI) has emerged as a key research area within signal and image processing, aiming to make the internal representations and decision-making mechanisms of complex models more transparent and understandable. In the histopathological context, signal- and feature-level interpretability is essential to verify that model predictions are driven by meaningful morphological and textural patterns rather than by spurious correlations or acquisition artefacts. This review work analyses and synthesizes the main XAI techniques applied to the analysis of breast cancer histopathological images, including saliency-based methods, feature attribution approaches, concept-based techniques and intrinsically interpretable architectures. The strengths and limitations of each approach are discussed from an image processing perspective, with particular emphasis on conceptual aspects of spatial localization, multiscale coherence and signal fidelity. Finally, an emerging trend towards multidimensional and composite approaches to explainability is highlighted, which may support the development of standardized evaluation strategies and the design of explainable-by-design image analysis systems.

Chatbots serve as valuable instruments for enhancing students' educational experience and aiding them in their day-to-day academic tasks. Advances in Generative AI (GAI) have ushered in increasingly sophisticated and adaptive chatbots, with ChatGPT and DALL⋅E being prime examples. ChatGPT excels at generating text-based answers across diverse areas of knowledge, while DALL⋅E is adept at converting text-based concepts into visual imagery. These technologies are increasingly used by students across various levels of education. In this study, we introduce AI4Design, a web-based system designed to assist design students with their course projects by acting as an intelligent chatbot. The field of design is propitious for such work because of the increasing use of technology and the necessity of introducing its critical use during study. Comprising two integrated modules, the system is based on a two-step workflow. The first step is anchored on ChatGPT, enabling students to prompt questions and receive answers. The second step allows for the generation of one or more images based on the system's answer to the initial question. Our research assesses whether our system can offer valuable insights and inspiration to students in their design work. We conducted an exploratory study in the Design domain involving 31 students from the Lebanese American University. Over a two- to three-day period, participants used the AI4Design system to enhance their projects. A subsequent evaluation of their work indicated improvements in conceptual clarity and visual outputs that highlighted a measurable increase in creativity, supporting the efficacy of both the system and its foundational learning model, which will be confirmed in the future through a large-scale experimental study. Meanwhile, our study suggests that in the iterative design process, GAI can assist students in making better decisions by giving them just-in-time access to a broader palette of possibilities.

Abstract reasoning is a key ability for students’ cognitive development, yet traditional methods often fail to provide an accurate and objective assessment. We propose a novel machine learning approach that utilises EEG data from low-cost headsets to predict response correctness in abstract reasoning tasks. This paper presents an adaptive LSTM model incorporating multi-head attention for analysing EEG data acquired during both the reasoning (pre-response) and feedback (post-response) phases of abstract reasoning questions. Results demonstrate the model’s ability to predict answer correctness with promising performance across multiple evaluation metrics. Additionally, we investigate the influence of self-reported confidence on EEG-based correctness prediction, yielding valuable insights. Our model demonstrates scalability and robustness when applied to larger datasets. This research paves a way for prospective applications of innovative tools in personalised feedback and adaptive learning systems, and advances methods for assessing and supporting cognitive growth.

Intent-Based Networking (IBN) has simplified network management and orchestration at a high level, but configuring User Equipment (UE), like 5G modems, is still a complex and demanding task due to dynamic requirements and intricate device-specific settings, and scalability challenges, especially when dealing with distributed, edge-based devices. This paper explores the potential of using Intelligent Chatbots powered by Generative Artificial Intelligence and Large Language Models (LLMs) operating as co-pilots to automate and optimize modem configurations. We propose a scalable chatbot system that translates user intents into actionable configurations, enhancing security, performance, and adaptability. To this end, we introduce a middleware that bridges LLMs with 5G Modem interfaces, eliminating retraining needs while ensuring engaging, real-time interaction with users. Additionally, we analyze key challenges in integrating LLM-based chatbots with UE and discuss the benefits of query caching in optimizing response times. Our findings highlight the potential of Intelligent Chatbots in extending IBN principles to UE, enabling a more automated and user-friendly approach to network configuration.

Intent-based Networking (IBN) has emerged as an innovative approach to automate the provisioning of network services while simplifying the interaction between the users and the network, allowing users (e.g., administrators) to define high-level desired outcomes (i.e., intents), and translating expressed intents into automated network configurations. One of the main challenge in IBN is the correct acquisition of the user intents and subsequently the accurate translation into actionable configurations to enforce into the network. Despite some efforts in improving user-to-IBN system interaction, a gap still remains in ensuring satisfactory user experiences and contextually appropriate and coherent responses or translation results. To this purpose we consider using recent advancements in Generative AI, and in particular in Large Language Models, a promising approach to enhance IBN in the scope of intent acquisition and translation. Accordingly, this work investigates the integration of IBN systems with LLM-based Conversational Agents (i.e., intelligent chatbots), on the one hand to enhance the user experience while injecting intents and, on the other hand, to assure an accurate understanding of user intents and their translation into a coherent set of network configurations, which are generated automatically. The chatbot operation according to the proposed approach is illustrated in a DHCP configuration use case.

In the last two years, thanks to the advent of generative AI, new avenues are opening up in its use in the educational field, e.g. through the use of ChatGPT. But, while generative AI is opening up new avenues for both teaching and learning, there are no proposals for its use in the context of students with Dyslexia and in general for students with Specific Learning Disorders (SLD), that afflict between 10 and 15 percent of people worldwide. In this paper we present AI4LA, a web intelligent chatbot based on generative AI, with the aim to help dyslexic students during their learning process. This system is capable of analyzing conversational data to model students' understandings and misconceptions, thereby offering personalized educational support. Significantly, we highlight the chatbot's ability to create Concept Maps from conversations, visually representing the student's knowledge structure and progression. An evaluation of the effectiveness of the system is presented, in a realworld educational settings, based on feedback from a sample of dyslexic students. The findings suggest that our system holds great promise in revolutionizing the way dyslexic students engage with educational content, making learning more accessible, interactive, and effective.

In order to examine the reliability of peer assessment settings for the evaluation of programming skills using peer assessment and the item response theory (IRT) technique in a small classes, the optimal conditions such as the numbers of peers and the number of tasks are investigated using parameters extracted from the surveyed data. The survey data consisted of 31 students whose partial participation consisted of joining three peer assessments out of the 5 sessions during which these took place. Peer rating conditions such as the number of peer raters or tasks are examined using mean expected standard errors. Also, the relationship between instructor's ratings and estimated ability is examined using the IRT model to look at variations in the number of peer raters and tasks. The results provide evidence that a set of guidelines could better organise peer assessment of the evaluation of programming skills in actual course settings.

The recent developments in Deep Learning Generative methods provided significant new and developing assets in Technology Enhanced Learning, with the production of conversational agents able to sustain a dialogue with the learner, and answer direct questions about the topic to learn. The possibility to help students at any time, about the topic of a study course, is important at any level of education, and it is almost crucial when it comes to students in Massive Open Online Courses, where the possibility to have direct interaction with the teacher, could be extremely limited. In this paper we present Eduport, a system designed to answer questions posted by learners in natural language. The system is based on a multi-tiered virtual interaction, ranging from using FAQs to generating responses grounded in the course material. One of its main feature is the use of OpenAI API, specifically utilizing the GPT 3.5-turbo model, for more refined assistance on the question at hand. This paper delineates the Eduport system and shares preliminary findings, encouraging outcomes in two primary dimensions: 1) the significant adherence of Eduport’s responses to the established learning materials, and 2) the proficient harnessing of the OpenAI GPT 3.5-turbo model, to yield answers closely aligned with the course’s didactic content.

This paper presents a comparative analysis of novel LLM-based architectures designed specifically for system configuration purposes. Generative Artificial Intelligence (Gen AI) has rapidly evolved, offering transformative capabilities in content generation across various domains. Large Language Models (LLMs) stand at the forefront of this evolution, revolutionizing natural language understanding and enabling sophisticated conversational systems. Leveraging the potential of LLMs, our study introduces a novel system architecture centered around an intelligent chatbot tailored to assist learners in complex network configurations. By integrating Generative Pre-trained Transformer-based models with Retrieval Augmented Generation (RAG) and Function Calling features, our architecture aims to provide a co-pilot-like experience, guiding users through understanding requirements and generating configuration scripts. Through a comparative analysis of three LLM architectures, each tailored to handle system network configuration, we evaluate their effectiveness, strengths, and limitations. Our findings offer valuable insights into the potential applications of Generative AI in network operations and highlight avenues for future research and development.

In the new global reality, distance learning has grown exponentially, and students with Specific Learning Disorders (SLD), i.e., those with dyslexia, dysorthographia, or dyscalculia, struggle to learn. Because they have no direct contact with the teacher, they resort to using suitable tools and instruments, together with personalised learning plans. In addition, students with SLDs who take in-person courses must also follow an adapted didactic strategy. This problem affects millions of people around the world, and Technology Enhanced Learning currently proposes instruments and tools to help learners during their learning process. In this paper we propose a brief review of the literature about the use of Artificial Intelligence from 2018 to 2023 in the educational field specifically for dyslexic students in the post-transformer era. We aim to baseline the contribution of the use of generative and classic AI to help learners, highlighting possible strengths and weaknesses. The results of our inquiry show that in the specified years, the AI community has offered little to this field.

Intent-based networking has attracted interest in the academic research for enhancing network management operations with user-oriented features. One of the main challenge in this field is the acquisition of the user intents and subsequently the relative translation into policies for the automatic management of the network. Concerning this task, the primary technique employed is relying on Graphical User Interfaces (GUI)s. In addition, the use of Natural Language Processing techniques has been extensively adopted for improved user experience. Recently, some preliminary studies have shown that using Large Language Models (LLMs) for this purpose leads to achieve interesting results. However, based on a comprehensive analysis of the state of the art, it has emerged that the works utilizing the LLMs do not fully exploit all the capabilities these tools could potentially offer. For this reason, the doctoral work aims to address the following challenges: enhancing user experience through the utilization of intelligent chatbots, improving the correct understanding of user intents and ensuring the translation of user intentions into a coherent set of network configurations, which are generated automatically.

Concept mapping is a valuable method to represent a domain of knowledge, also with the aim of supporting educational needs. Students are called upon to construct their own knowledge through a meaningful learning process, linking new concepts to concepts they have already learned, i.e., connecting new knowledge to knowledge they already possess. Moreover, the particular graphic form of a concept map makes it easy for the teacher to construct and interpret both. Consequently, for an educator, the ability to assess concept maps offered by students, facilitated by an automated system, can prove invaluable. This becomes even more apparent in educational settings where there is a large number of students, such as in Massive Open Online Courses. Here, we propose two new measures devised to evaluate the similarity between concept maps based on two deep-learning embedding models: InferSent and Universal Sentence Encoder. An experimental evaluation with a sample of teachers confirms the validity of one such deep-learning model as the baseline of the new similarity measure. Subsequently, we present a proof-of-concept dashboard where the measures are used to encode a concept map in a 2D space point, with the aim of helping teachers monitor students’ concept-mapping activity.

We present TutorChat, an intelligent chatbot conceived to be able support search and synthesis of information during a learning task accomplishment, in particular for dyslexic students. TutorChat is based on ChatGPT; it is able to support question/answer inter-activity of learners, and to generate concept maps on the topics at hand, with the possibility, beside analysis, to have such maps extended with additional sub-maps starting from a selected concept. We let TutorChat be used by a sample of dyslexic learners, coming from different educational levels. Then we collected encouraging sample’s feedback, through a questionnaire, about appreciation of the system’s services, and perception of the usefulness coming from its use.

This paper introduces a novel approach to enhancing educational outcomes by integrating electroencephalography (EEG) and supervised machine learning. Our methodological framework leverages real-time EEG data analysis, focusing on alpha, beta, gamma, delta, and theta wave patterns to develop cognitive indices such as Focus, Engagement, Relaxation, Fatigue, Involvement, and Stress. These indices are pivotal for delineating the Flow state among learners, a mental state conducive to optimal learning. We detail the process of EEG data collection where students are equipped with a non-intrusive EEG headset that monitors their brainwave patterns in real time. This setup involves creating a baseline of each student's cognitive patterns during an initial calibration phase, which is refined over time to enhance system accuracy. Using this data, we employ feature extraction techniques to develop predictive models capable of assessing and predicting the learners' cognitive states. Our research advances the personalization of learning environments by providing real-time feedback to students about their mental states. This feedback allows students to adjust their engagement strategies dynamically, aiming to maintain or achieve the mental states that are most conducive to learning. Initial findings suggest that our approach can significantly improve educational practices by adapting to and fostering students' cognitive states. The implications of this study extend beyond simple academic performance enhancement, promoting a deeper integration of cognitive neuroscience within educational systems. By developing tools that adapt to students' cognitive needs, we aim to foster an educational environment that values and enhances individual learning capacities.