Abstract This paper addresses the emerging threat of indirect prompt injection, a technique in which malicious agents embed prompts into seemingly innocuous text to manipulate the behaviour and output of Generative Large Language Models (LLMs). As LLMs become more popular and commonly used in everyday activities, this type of attack poses serious concerns about their responses. Without knowledge and protection, processes that depend on them may not be reliable. We present and analyse real-world high-risk cases, most notably in the scenario of a comparative analysis of Curriculum Vitae documents. In this scenario, prompt injection is used to mislead the human resources manager who uses LLMs to support personnel selection. This risk is also becoming increasingly relevant in educational contexts where LLMs are used for activities such as automated essay review, tutoring, and content generation, potentially enabling subtle forms of manipulation and misconduct. The hidden prompt subtly alters the behaviour of the generative model, steering its output away from the intended results of the user’s LLM prompt. We analyze the structure of these attacks, evaluate the vulnerability and resilience of popular LLMs, and suggest potential countermeasures. We conclude by discussing the broader implications, evolving risks, and opportunities for securing LLM-based workflows.

In the field of machine learning and pattern recognition, the use of binary correlation indices is essential for accurate prediction and modelling. This work presents a novel evolutionary method to address the problem of discovering binary correlation indices in different application domains. The proposed approach introduces the concept of meta-correlation, a parametric formula representing classes of binary similarity indices, and optimizes it through an evolutionary scheme. The method has been experimented with and validated in the context of the link prediction problem based on local topological similarity (i.e. graph neighbourhood). A Differential Evolution optimization algorithm finds the evolved correlations that perform best in a given domain. Experiments conducted across different network domains have shown that the instances of the discovered meta-correlations generally outperform state-of-the-art binary correlation indices for all the experimented domains. This approach effectively explores the correlation space and can find a unique pattern that adapts to the domains under consideration. The meta-correlation classes can be applied to both topological and semantic similarity problems, taking into account local information without requiring complete knowledge of the graph.

Abstract Inter-species emotional relationships, particularly the symbiotic interaction between humans and dogs, are complex and intriguing. Humans and dogs share fundamental mammalian neural mechanisms including mirror neurons, crucial to empathy and social behavior. Mirror neurons are activated during the execution and observation of actions, indicating inherent connections in social dynamics across species despite variations in emotional expression. This study explores the feasibility of using deep-learning Artificial Intelligence systems to accurately recognize canine emotions in general environments, to assist individuals without specialized knowledge or skills in discerning dog behavior, particularly related to aggression or friendliness. Starting with identifying key challenges in classifying pleasant and unpleasant emotions in dogs, we tested advanced deep-learning techniques and aggregated results to distinguish potentially dangerous human--dog interactions. Knowledge transfer is used to fine-tune different networks, and results are compared on original and transformed sets of frames from the Dog Clips dataset to investigate whether DogFACS action codes detailing relevant dog movements can aid the emotion recognition task. Elaborating on challenges and biases, we emphasize the need for bias mitigation to optimize performance, including different image preprocessing strategies for noise mitigation in dog recognition (i.e., face bounding boxes, segmentation of the face or body, isolating the dog on a white background, blurring the original background). Systematic experimental results demonstrate the system’s capability to accurately detect emotions and effectively identify dangerous situations or signs of discomfort in the presence of humans.

In the framework of Adversarial Machine Learning, several detection and protection techniques are used to characterize specific attack-defense scenarios.In this paper, we present universal, unrestricted black-box adversarial attacks based on a multi-objective nested evolutionary algorithm able to incorporate the detection rate and a measure of image quality into the attack building phase.

Crossover operators are very important components in Evolutionary Computation. Here we are interested in crossovers for the permutation representation that find applications in combinatorial optimization problems such as the permutation flowshop scheduling and the traveling salesman problem. We introduce three families of permutation crossovers based on algebraic properties of the permutation space. In particular, we exploit the group and lattice structures of the space. A total of 34 new crossovers is provided. Algebraic and semantic properties of the operators are discussed, while their performances are investigated by experimentally comparing them with known permutation crossovers on standard benchmarks from four popular permutation problems. Three different experimental scenarios are considered and the results clearly validate our proposals.

Abstract Crowds express emotions as a collective individual, which is evident from the sounds that a crowd produces in particular events, e.g., collective booing, laughing or cheering in sports matches, movies, theaters, concerts, political demonstrations, and riots. A critical question concerning the innovative concept of crowd emotions is whether the emotional content of crowd sounds can be characterized by frequency-amplitude features, using analysis techniques similar to those applied on individual voices, where deep learning classification is applied to spectrogram images derived by sound transformations. In this work, we present a technique based on the generation of sound spectrograms from fragments of fixed length, extracted from original audio clips recorded in high-attendance events, where the crowd acts as a collective individual. Transfer learning techniques are used on a convolutional neural network, pre-trained on low-level features using the well-known ImageNet extensive dataset of visual knowledge. The original sound clips are filtered and normalized in amplitude for a correct spectrogram generation, on which we fine-tune the domain-specific features. Experiments held on the finally trained Convolutional Neural Network show promising performances of the proposed model to classify the emotions of the crowd.

Particle Swarm Optimization (PSO), though originally introduced for continuous search spaces, has been increasingly applied to combinatorial optimization problems. In this paper, we focus on the PSO applications to permutation-based problems. As far as we know, the most popular and general PSO sche mes for permutation solutions are those based on random key techniques. After highlighting the main criticalities of the random key approach, we introduce a discrete PSO variant for permutation-based optimization problems. By simulating search moves through a vector space, the proposed algorithm, Algebraic PSO (APSO), allows the original PSO design to be applied to the permutation search space. APSO directly represents both particle positions and velocities as permutations. The APSO search scheme is based on a general algebraic framework for combinatorial optimization based on strong mathematical foundations. However, in order to make this new scheme viable, some challenges have to be overcome: the choice of the order of the velocity terms, and the rationale behind the PSO inertial move. Design solutions have been proposed for both the issues. Furthermore, an alternative geometric interpretation of classical PSO dynamics allows to introduce a major APSO variant based on a novel concept of convex combination between permutation objects. In total, four APSO schemes have been introduced. Experiments have been held to compare the performances of the APSO schemes with respect to the random key based PSO schemes in literature. Widely adopted benchmark instances of four popular permutation problems have been considered. The experimental results clearly show that, with high statistical evidence, APSO outperforms its competitors and it reaches results comparable with state-of-the-art on most of the instances considered.

[No abstract available]

Crossover operators are very important tools in Evolutionary Computation. Here we are interested in crossovers for the permutation representation that find applications in combinatorial optimization problems such as the permutation flowshop scheduling and the traveling salesman problem. We introduce three families of permutation crossovers based on algebraic properties of the permutation space. In particular, we exploit the group and lattice structures of the space. A total of 14 new crossovers is provided. Algebraic and semantic properties of the operators are discussed, while their performances are investigated by experimentally comparing them with known permutation crossovers on standard benchmarks from four popular permutation problems. Three different experimental scenarios are considered and the results clearly validate our proposals.

In this paper an innovative dynamic dimensional morphing metaphor is introduced to monitor students' engagement and cohort dynamic. Teachers using eLearning monitoring tools find them usually lacking usability and inadequate to give productive feedback about their learning designs. Learning analytics tools mostly focus on after course analysis with the assumption that user competence in data analysis is high. The tool we propose is based on visual interface morphing: reshaping of the interface elements, such as learning objects' links and icons, is put in place to reflect some key performance indicators of learners' activities. Quantitative and temporal analytics data, aggregated using various functions, is used to present animated enhanced information to teachers. Experiments for the assessment of the effectiveness of the proposed tool has been conducted on data from higher education courses. Through logs' analysis and teachers' questionnaires the usability and validity of the proposed metaphor has been assessed. The proposed tool outperforms traditional monitoring techniques.

Particle Swarm Optimization (PSO), though being originally introduced for continuous search spaces, has been increasingly applied to combinatorial optimization problems. In particular, we focus on the PSO applications to permutation problems. As far as we know, the most popular PSO variants that produce permutation solutions are those based on random key techniques. In this paper, after highlighting the main criticalities of the random key approach, we introduce a totally discrete PSO variant for permutation-based optimization problems. The proposed algorithm, namely Algebraic PSO (APSO), simulates the original PSO design in permutations search space. APSO directly represents the particle positions and velocities as permutations. The APSO search scheme is based on a general algebraic framework for combinatorial optimization previously, and successfully, introduced in the context of discrete differential evolution schemes. The particularities of the PSO design scheme arouse new challenges for the algebraic framework: the non-commutativity of the velocity terms, and the rationale behind the PSO inertial move. Design solutions have been proposed for both the issues, and two APSO variants are provided. Experiments have been held to compare the performances of the APSO schemes with respect to the random key based PSO schemes in literature. Widely adopted benchmark instances of four popular permutation problems have been considered. The experimental results clearly show, with high statistical evidence, that APSO outperforms its competitors.

This paper describes a framework for successful interaction between universities and schools. It is common for computing academics interested in outreach (computer science evangelism) to work with local schools, particularly in countries where the computing curriculum in K-12 is new or underdeveloped. However it is rare for these collaborations to be ongoing, and for resources created through these school-university links to be shared beyond the immediate neighborhood. We have achieved this, through shared resources, careful evaluation, and cross-country collaboration. The activities themselves are inspired by ideas from the Lifelong Kindergarten group at MIT, emphasizing playful exploration of computational concepts and interdisciplinary working.

In this work, we present SEMO, a Semantic Model for Emotion Recognition, which enables users to detect and quantify the emotional load related to basic emotions hidden in short, emotionally rich sentences (e.g. news titles, tweets, captions). The idea of assessing the semantic similarity of concepts by looking at the occurrences and co-occurrences of terms describing them in pages indexed by a search engine can be directly extended to emotions, and to the words expressing them in different languages. The emotional content associated to a particular emotion for a term can thus be estimated using web-based similarity measures, e.g. Confidence, PMI, NGD and PMING, aggregating the distance computed by a model of emotions, e.g. Ekman, Plutchik and Lovheim. Emotions are ranked based on their similarity to the analyzed text, describing each sentence through a vector of values of emotion load, which form the Vector Space Model for the chosen emotion model and similarity measures. The model is tested comparing experimental results to a ground truth in literature. SEMO takes care of both the phases of data collection and data analysis, to produce knowledge to be used in application domains such as social robots, recommender systems, and human-machine interactive systems.

Emotional affordances represent a recently introduced concept which model all the mechanisms used to collect/transmit emotional meaning in the context of human machine interaction. In this work, we introduce and formally define the cognitive role of emotional affordances in a collaboration human-machine dialogue as tools for triggering or recognizing planning-based activities of delegation, goal negotiation, state acquisition, plan prioritization, taking place with the interaction partner. The presented formal model is grounded in an emergency scenario where reacting to emotional affordances or transmitting an emotional content is instrumental to reach the goal of an effective collaborative response. The implementation issues of generation and recognition of emotional affordance are also discussed.

Topological link prediction is the task of assessing the likelihood of new future links based on topological properties of entities in a network at a given time. In this paper, we introduce a multistrain bacterial diffusion model for link prediction, where the ranking of candidate links is based on the mutual transfer of bacteria strains via physical social contact. The model incorporates parameters like efficiency of the receiver surface, reproduction rate and number of social contacts. The basic idea is that entities continuously infect their neighborhood with their own bacteria strains, and such infections are iteratively propagated on the social network over time. The probability of transmission can be evaluated in terms of strains, reproduction, previous transfer, surface transfer efficiency, number of direct social contacts i.e. neighbors, multiple paths between entities. The value of the mutual strains of infection between a pair of entities is used to rank the potential arcs joining the entity nodes. The proposed multistrain diffusion model and mutual-strain infection ranking technique have been implemented and tested on widely accepted social network data sets. Experiments show that the MSDM-LP and mutual-strain diffusion ranking technique outperforms state-of-the-art algorithms for neighbor-based ranking.

This paper introduces an original algebraic approach to differential evolution (DE) algorithms for combinatorial search spaces. An abstract algebraic differential mutation for generic combinatorial spaces is defined by exploiting the concept of a finitely generated group. This operator is specialized for the permutations space by means of an original randomized bubble sort algorithm. Then, a discrete DE algorithm is derived for permutation problems and it is applied to the permutation flowshop scheduling problem with the total flowtime criterion. Other relevant components of the proposed algorithm are: a crossover operator for permutations, a novel biased selection strategy, a heuristic-based initialization, and a memetic restart procedure. Extensive experimental tests have been performed on a widely accepted benchmark suite in order to analyze the dynamics of the proposed approach and to compare it with the state-of-the-art algorithms. The experimental results clearly show that the proposed algorithm reaches state-of-the-art performances and, most remarkably, it is able to find some new best known results. Furthermore, the experimental analysis on the impact of the algorithmic components shows that the two main contributions of this paper, i.e., the discrete differential mutation and the biased selection operator, greatly contribute to the overall performance of the algorithm.

A novel method for the automatic online extraction of contexts from collaborative explanation network is introduced. The method explore an unknown online collaborative network in order to find multiple explanatory paths between seed concepts. The exploration is driven by an online randomized walk informed by a heuristics based on semantic proximity measures. A pheromone-like model is then applied to the analysis of the relevance of concepts in multiple explanatory paths in order to extract the relevant contexts. Experiments held on the collaborative network Wikipedia and accepted datasets show that the proposed method is able to determine contexts with high degree of relevance which outperforms other methods. The methodology have general aim and it can be easily extended to other online collaborative networks and to non-textual domains.

In this work, the Linear Ordering Problem (LOP) has been approached using a discrete algebraic-based Differential Evolution for the Linear Ordering Problem (LOP). The search space of LOP is composed by permutations of objects, thus it is possible to use some group theoretical concepts and methods. Indeed, the proposed algorithm is a combinatorial Differential Evolution scheme designed by exploiting the group structure of the LOP solutions in order to mimic the classical Differential Evolution behavior observed in continuous spaces. In particular, the proposed differential mutation operator allows to obtain both scaled and extended differences among LOP solutions represented by permutations. The performances have been evaluated over widely known LOP benchmark suites and have been compared to the state-of-the-art results.

We present several interrelated technical and empirical contributions to the problem of emotion-based music recommendation and show how they can be applied in a possible usage scenario. The contributions are (1) a new three-dimensional resonance-arousal-valence model for the representation of emotion expressed in music, together with methods for automatically classifying a piece of music in terms of this model, using robust regression methods applied to musical/acoustic features; (2) methods for predicting a listener’s emotional state on the assumption that the emotional state has been determined entirely by a sequence of pieces of music recently listened to, using conditional random fields and taking into account the decay of emotion intensity over time; and (3) a method for selecting a ranked list of pieces of music that match a particular emotional state, using a minimization iteration method. A series of experiments yield information about the validity of our operationalizations of these contributions. Throughout the article, we refer to an illustrative usage scenario in which all of these contributions can be exploited, where it is assumed that (1) a listener’s emotional state is being determined entirely by the music that he or she has been listening to and (2) the listener wants to hear additional music that matches his or her current emotional state. The contributions are intended to be useful in a variety of other scenarios as well.

Recent findings show that online reviews, blogs, and discussion forums on chronic diseases and drugs are becoming important supporting resources for patients. Extracting information from these substantial bodies of texts is useful and challenging. We developed a generative probabilistic aspect mining model (PAMM) for identifying the aspects/topics relating to class labels or categorical meta-information of a corpus. Unlike many other unsupervised approaches or supervised approaches, PAMM has a unique feature in that it focuses on finding aspects relating to one class only rather than finding aspects for all classes simultaneously in each execution. This reduces the chance of having aspects formed from mixing concepts of different classes; hence the identified aspects are easier to be interpreted by people. The aspects found also have the property that they are class distinguishing: They can be used to distinguish a class from other classes. An efficient EM-algorithm is developed for parameter estimation. Experimental results on reviews of four different drugs show that PAMM is able to find better aspects than other common approaches, when measured with mean pointwise mutual information and classification accuracy. In addition, the derived aspects were also assessed by humans based on different specified perspectives, and PAMM was found to be rated highest.

Effective sharing of diverse social media is often inhibited by limitations in their search and discovery mechanisms, which are particularly restrictive for media that do not lend themselves to automatic processing or indexing. Here, we present the structure and mechanism of an adaptive search engine which is designed to overcome such limitations. The basic framework of the adaptive search engine is to capture human judgment in the course of normal usage from user queries in order to develop semantic indexes which link search terms to media objects semantics. This approach is particularly effective for the retrieval of multimedia objects, such as images, sounds, and videos, where a direct analysis of the object features does not allow them to be linked to search terms, for example, nontextual/icon-based search, deep semantic search, or when search terms are unknown at the time the media repository is built. An adaptive search architecture is presented to enable the index to evolve with respect to user feedback, while a randomized query-processing technique guarantees avoiding local minima and allows the meaningful indexing of new media objects and new terms. The present adaptive search engine allows for the efficient community creation and updating of social media indexes, which is able to instill and propagate deep knowledge into social media concerning the advanced search and usage of media resources. Experiments with various relevance distribution settings have shown efficient convergence of such indexes, which enable intelligent search and sharing of social media resources that are otherwise hard to discover.

This paper introduces the Asynchronous Differential Evolution (ADE) scheme which generalizes the classical Differential Evolution (DE) approach along the dimension of Synchronization Degree (SD). SD regulates the synchrony of the evolution of the current population, i.e. how fast it is replaced by the newly generated population. The definition of the ADE scheme is given and different synchronization strategies are discussed. The introduction of SD parameter allows the tuning of the differential evolution from a completely asynchronous behavior to a super-synchronous behavior. Experiments show that a low SD generally improves the convergence speed and the convergence probability with respect to the classical synchronous DE. Moreover the ordering strategies introduced in ADE seem to improve the performances of the only already known asynchronous variant of DE (the Dynamical Differential Evolution Strategy).

Adaptive search engines (ASE), used in the retrieval of multimedia objects adapt their behavior depending on the user feedback in order to eventually converge to the optimal answer. The adaptive architecture has been shown to improve the performance in case of multimedia objects retrieval, when pre-indexing techniques are costly or can be applied only partially. The continuous user feedbacks on the lists of returned objects are used to filter out irrelevant objects and promote the relevant ones. This work propose an original dealer/opponent game model for ASE. The system/user interactive process which takes place in ASE can be modeled as a discovery game between a dealer, the user community which holds a secret consisting in the optimal answer to a query, and an opponent, i.e. the system, which tries to discover the secret by submitting tentative solutions on which it receives the user/dealer feedback. It is shown how the complexity of the game can be related to known games. An evolutionary approach to solve the ASE game is also presented. Experimental results shows convergence to the optimal solution with acceptable performance for real domain size. The proposed schema is quite general and can fit other adaptive search architectures which appear in e-business and e-commerce applications.

This work introduces a framework for a representation of web-pages by means of multiple knowledge structures in order to improve semantic usage modelling by multiple semantic information. The proposed approach generalizes existing models, developed mainly for e-commerce systems, in order to include and integrate features of hypertext-structured educational systems such as e-learning systems.

The diffusion of domotic and ambient intelligence systems have introduced a new vision in which autonomous deliberative agents operate in environments where reactive responses of devices can be cooperatively exploited to fulfill the agent's goals. In this article a model for automated planning in reactive environments, based on numerical planning, is introduced. A planner system, based on mixed integer linear programming techniques, which implements the model, is also presented. The planner is able to reason about the dynamic features of the environment and to produce solution plans, which take into account reactive devices and their causal relations with agent's goals by exploitation and avoidance techniques, to reach a given goal state. The introduction of reactive domains in planning poses some issues concerning reasoning patterns which are briefly depicted. Experiments of planning in reactive domains are also discussed.

This paper present a technique based on genetic algorithms for generating online adaptive services. Online adaptive systems provide flexible services to a mass of clients/users for maximizing some system goals; they dynamically adapt the form and the content of the issued services while the population of clients evolve over time. The idea of online genetic algorithms (online GAs) is to use the online clients response behavior as a fitness function in order to produce the next generation of services. The principle implemented in online GAs, "the application environment is the fitness", allow to model highly evolutionary domains where both services providers and clients change and evolve over time. The flexibility and the adaptive behavior of this approach seems to be very relevant and promising for applications characterized by highly dynamical features such as in the web domain (online newspapers, e-markets, websites and advertising engines). Nevertheless the proposed technique has a more general aim for application environments characterized by a massive number of anonymous clients/users which require personalized services, such as in the case of many new IT applications.

This paper shows how current planning technology can be used in order to model online educational activity. In particular the automated planning approach allows to model the interactive process of designing the structure of educational tools and entities, such as courses, curricula studiorum, examples and training exercises, handbooks. Planning models allow to exploit educational knowledge bases in order to produce flexible and adaptive tools which are suited on the individual educational needs of the users. In addition to the generation of educational support material, state-of-the-art planning systems can also be used to monitor the interactive learning task of the user. Recent advancements on planning models based on resources management allow to refine and tune user modelling goals in order to realize a personalized service. An example of educational domain model based on planning with resources, is shown, in order to support the dynamical generation and monitoring of web based online courses.

A linear [n,k,d]/sub q/ code C is called near maximum-distance separable (NMDS) if d(C)=n-k and d(C/sup /spl perp//)=k. The maximum length of an NMDS [n,k,d]/sub q/ code is denoted by m'(k,q). In this correspondence, it has been verified by a computer-based proof that m'(5,8)=15, m'(4,9)=16,m'(5,9)=16, and 20/spl les/m'(4,11)/spl les/21. Moreover, the NMDS codes of length m'(4,8), m'(5,8), and m'(4,9) have been classified. As the dual code of an NMDS code is NMDS, the values of m'(k,8), k=10,11,12, and of m'(k,9),k=12,13,14 have been also deduced.