Deep Sequence Models: Learning to Generate Data and Adversarial Attacks

This study proposes a dependency grammar-based self-attention multilayered bidirectional long short-term memory (DG-M-Bi-LSTM) model for subject–predicate–object (SPO) tuple recognition from natural language (NL) sentences. To add recent knowledge to the knowledge base autonomously, it is essential to extract knowledge from numerous NL data. Therefore, this study proposes a high-accuracy SPO tuple recognition model that requires a small amount of learning data to extract knowledge from NL sentences. The accuracy of SPO tuple recognition using DG-M-Bi-LSTM is compared with that using NL-based self-attention multilayered bidirectional LSTM, DG-based bidirectional encoder representations from transformers (BERT), and NL-based BERT to evaluate its effectiveness. The DG-M-Bi-LSTM model achieves the best results in terms of recognition accuracy for extracting SPO tuples from NL sentences even if it has fewer deep neural network (DNN) parameters than BERT. In particular, its accuracy is better than that of BERT when the learning data are limited. Additionally, its pretrained DNN parameters can be applied to other domains because it learns the structural relations in NL sentences.

Artificial intelligence (AI) is increasingly influencing medical education by enabling adaptive learning, AI-assisted assessment, and scalable instructional tools. Natural language processing, machine learning, and generative large language models offer innovative ways to support teaching and learning, yet their integration raises ethical, pedagogical, and infrastructural challenges. This viewpoint article aims to examine the current applications, benefits, and challenges of AI in medical education and propose strategies for responsible and effective integration. AI tools such as chatbots, virtual patients, and intelligent tutoring systems enhance personalized and immersive learning. Automated grading and predictive analytics support efficient evaluations, while AI-assisted writing tools streamline content creation. Despite these advances, concerns persist around data privacy, algorithmic bias, unequal access, and diminished critical thinking. Key solutions include AI literacy training, data oversight, equitable infrastructure, and curriculum reform. The FACETS framework offers 6 dimensions (ie, form, application, context, instructional mode, technology, and the SAMR [substitution, augmentation, modification, redefinition model]) to evaluate AI integration effectively. AI offers substantial opportunities to transform medical education, but its adoption must be ethical, equitable, and pedagogically grounded. Strategic frameworks such as FACETS, combined with institutional governance and cross-sector collaboration, are essential to guide implementation so that AI enhances learning outcomes while preserving the humanistic foundations of medical practice.

In this paper, we present a system that translates first order logic (FOL) formulas into natural language (NL) sentences. The motivation comes from an intelligent tutoring system teaching logic as a knowledge representation language, where it is used as a means for feedback to the users. FOL to NL conversion is achieved by using a rule-based approach, where we exploit the pattern matching capabilities of rules. So, the system consists of a rule-based component and a lexicon. The rule-based unit implements the conversion process, which is based on a linguistic analysis of a FOL sentence, and the lexicon provides lexical and grammatical information that helps in producing the NL sentences. The whole system is implemented in Jess, a java-based expert system shell. The conversion process currently covers a restricted set of FOL formulas.

Action recognition, a field within human-centered computing, is vital for identifying and understanding human actions, benefiting applications like surveillance, autonomous vehicles, and human-computer interaction. Artificial intelligence’s enduring goal is to develop robust models for perceiving and understanding the visual world around us. Deep neural networks have shown exceptional performance in various tasks, profoundly impacting real-world applications, including action prediction and recognition, which have advanced significantly in recent years. Most work in visual human action recognition focuses on single viewpoints or modalities from complete observation. Yet, the real significance lies in predicting future actions from incomplete observations to prevent real-world tragedies. With the availability of multiple cameras and data from multiple modalities (RGB, Depth, and Skeleton) available today, it becomes possible to model human action in multi-view and multi-modality context, minimizing the data loss due to occlusions and signal quality issues to improve recognition accuracy on the strength of state-of-the-art deep learning models. However, deep neural network models are susceptible to adversarial attacks, where imperceptible perturbations can compromise action recognition model performance. This thesis focuses on identifying latent vulnerabilities and proposing a defense mechanism against such threats in a multi-modality and multi-view setting. This work introduces an efficient and effective attack mechanism that perturbs skeleton data by targeting key joints and segments while employing a graph attention mechanism that learns the semantics to perturb other modalities. Additionally, an approach has been developed that not only adds noise but also alters the visual spatial structure of skeleton data through generative modeling. Furthermore, this dissertation introduces a defense mechanism known as the Collaborative Knowledge Distillation Network, which leverages graph attention and knowledge distillation techniques. This network leverages the knowledge from compromised multi-view data and integrates information from clean data to address incomplete observations and noisy action videos, enhancing the robustness of action recognition models for real-world applications.

The language is a symbolic system unique to human being. The acquisition of language, which has its meanings in the real world, is important for robots to understand the environment and communicate with us in our daily life. This paper proposes a novel approach to establish a fundamental framework for the robots which can understand language through their whole body motions. The proposed framework is composed of three modules: “motion symbol”, “motion language model”, and “natural language model”. In the motion symbol module, motion data are symbolized by Hidden Markov Models (HMMs). Each HMM represents abstract motion patterns. Then the HMMs are defined as motion symbols. The motion language model is stochastically designed for links between motion symbols and words. This model consists of three layers of motion symbols, latent states and words. The connections between the motion symbol and the latent state, and between the latent state and the words are denoted by two kinds of probabilities respectively. One connection is represented by the probability that the motion symbol generates the latent state, and the other connection is represented by the probability that the latent state generates the word. Therefore, the motion language model can connect the motion symbols to the words through the latent state. The natural language model stochastically represents sequences of words. In this paper, a bigram, which is a special case of N-gram model, is adopted as the natural language model. This model has the words as nodes and transitions between two words as edges. Therefore sentence structure is expressed as transitions among words. The integration of the motion language model and natural language model can be implemented by the search computation for sentences corresponding to motions and for motions corresponding to sentences. Especially, the usage of the bigram as the natural language model provides a simple search computation so that appropriate and fast bidirectional computation between the motions and language can be achieved. Our approach makes it possible for humanoid robots not only to interpret motions as sentences but also to generate motions from sentences. The tests by using various motions and words validate our framework for the language acquisition of humanoid robots.

Natural language understanding (NLU) and natural language generation (NLG) are two fundamental and related tasks in building task-oriented dialogue systems with opposite objectives: NLU tackles the transformation from natural language to formal representations, whereas NLG does the reverse. A key to success in either task is parallel training data which is expensive to obtain at a large scale. In this work, we propose a generative model which couples NLU and NLG through a shared latent variable. This approach allows us to explore both spaces of natural language and formal representations, and facilitates information sharing through the latent space to eventually benefit NLU and NLG. Our model achieves state-of-the-art performance on two dialogue datasets with both flat and tree-structured formal representations. We also show that the model can be trained in a semi-supervised fashion by utilising unlabelled data to boost its performance.

Macroeconomic forecasting remains difficult because aggregate dynamics are nonlinear, regime-dependent, and increasingly informed by high-dimensional, mixed-frequency data, conditions under which traditional econometric models often lose accuracy, especially around turning points. This study systematically reviews how artificial intelligence (AI) is being used to address these challenges and what reliable evidence exists on its forecasting value. Using a PRISMA-aligned protocol, we searched Web of Science, Scopus, and ScienceDirect, identifying 1,627 records; after de-duplication and multi-stage screening, 178 studies were retained for qualitative synthesis. The review develops an evidence-based taxonomy of AI model families in macroeconomic forecasting and synthesizes their motivations, performance patterns, and limitations across targets such as gross domestic product (GDP) growth, inflation, and labor-market indicators. This study shows that AI methods, particularly tree-based ensembles, deep sequence models, and hybrid econometric–AI systems, frequently improve short-horizon forecasting and nowcasting in data-rich settings and during crisis regimes, but do not dominate uniformly across variables or horizons. The study's contributions are threefold: a structured taxonomy that organizes a fragmented field, a consolidated assessment of where AI gains are robust versus conditional, and a clear articulation of the main methodological and policy-relevant constraints shaping future macro-AI forecasting research.

Formalization of natural language metaphors is a notorious problem in artificial intelligence and in other overlapping domains. It is semantic vagueness that makes metaphors resistant to formulaic or algorithmic descriptions. Great effort has been invested into modeling metaphors computationally but the issue remains methodologically uncertain and needs further research. This paper works on a practical solution to the problem how metaphorical meaning can be represented in a way suitable for computation. The research agenda of this paper is interdisciplinary; it brings together an algebraic heuristic-driven theory for metaphors developed in artificial intelligence and an applied theory of meaning that comes from cognitive linguistics. This agenda postpones theoretical speculation and argument and is solely solution-focused, which contributes to the value of this paper’s attempt to bridge the cognitive science disciplines whose compatibility, though declared, is seldomdemonstrated in a piece of practical research.This paper works with metaphors of human emotions that are linguistically manifested in modern English discourse. Emotions by virtue of their ineffability as qualia are rich in metaphorical conceptualizations and serve the research agenda well. This paper in a meaningful way exposes and ranks designated properties of the FEAR, SADNESS, HAPPINESS, and RELAXATION/SERENITY concepts and arranges these properties into general-purpose ontologies that explicitly specify metaphorically preferred emotion conceptualizations and are good candidates for computation. In prospect, this paper will account for some theoretical aspects of the research and probe the algorithmic and repetitive nature of schemas that license metaphorical expressions in natural language.

The rising sophistication of adversarial attacks poses significant threats to the integrity and reliability of real-time financial fraud monitoring systems. As machine learning (ML) and deep learning models become more integrated into financial security infrastructures, they are increasingly vulnerable to subtle perturbations that can mislead fraud detection mechanisms. This review explores the intersection of explainable artificial intelligence (XAI) and generative models—such as Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs)—in fortifying financial systems against adversarial threats. The study categorizes existing adversarial attack vectors targeting transaction-level anomaly detection and evaluates the limitations of conventional defense techniques. Furthermore, it assesses the role of XAI methods, such as SHAP and LIME, in interpreting model decisions to uncover adversarial behavior in a transparent and auditable manner. Generative models are reviewed both as tools for generating adversarial examples and for enhancing model robustness through adversarial training and anomaly simulation. Real-time constraints are also examined, including latency, scalability, and system responsiveness. The paper concludes by identifying research gaps and proposing an integrated framework that combines interpretability, real-time detection, and generative defense strategies for resilient and accountable fraud monitoring in financial ecosystems.

Logic as a knowledge representation and reasoning language is a fundamental topic of an Artificial Intelligence (AI) course and includes a number of sub-topics. One of them, which brings difficulties to students to deal with, is converting natural language (NL) sentences into first-order logic (FOL) formulas. To assist students to overcome those difficulties, we developed the NLtoFOL system and equipped it with a strong assistance and feedback mechanism. In this work, first, we present that feedback mechanism. The mechanism can provide assistance before an answer is submitted, if requested, but mainly it provides assistance after an answer is submitted. To that end, it characterizes the answer in terms of completeness and accuracy to determine the level of incorrectness, based on an answer categorization scheme, introduced in this paper. The automatically generated natural language feedback sequences grow from general to specific and can include statements on a student’s metacognitive state. Feedback is provided as natural language sentences automatically generated through a template-based natural language generation mechanism. Second, we present an extensive evaluation of the effectiveness of the assistance and feedback mechanism on students’ learning. The evaluation of feedback with students showed that full feedback sequences lead to greater learning gains than sequences consisting of only flag feedback and bottom-out hints (n = 226), and that generic, template-based feedback sequences are comparable to the utility of problem-specific hints generated by human tutors (n = 120).

Artificial Intelligence (AI) is emerging as a transformative force in both drug development and orthodontics. This paper provides an overview of key AI techniques employed in these fields. In drug development, AI is accelerating discovery, development, and delivery through generative models, predictive modeling, image analysis, and natural language processing. Orthodontics benefits from AI through image analysis, treatment planning, and patient experience enhancement. Common AI techniques in both domains include machine learning, deep learning, natural language processing, and computer vision. By leveraging these technologies, researchers and clinicians can achieve significant advancements in drug discovery, treatment efficacy, and patient care.

Intelligent transportation systems, especially Autonomous Vehicles (AVs), are emerging as a paradigm with the potential to change modern society. However, with this, there is a strong need to ensure the security and privacy of such systems. AV ecosystems depend on machine learning algorithms to autonomously control their operations. Given the amount of personal information AVs collect, coupled with the distributed nature of such ecosystems, there is a movement to employ federated learning algorithms to develop secure decision-making models. Although federated learning is a viable candidate for data privacy, it is vulnerable to adversarial attacks, particularly data poisoning attacks, where malicious vectors would be injected in the training phase. Additionally, hyperparameters play an important role in establishing an efficient federated learning model that can be resilient against adversarial attacks. In this paper, to address these challenges, we propose a novel Optimized Quantum-based Federated Learning (OQFL) framework to automatically adjust the hyperparameters of federated learning using various adversarial attacks in AV settings. This work is innovative in two ways: first, a quantum-behaved particle swarm optimization technique is used to update the hyperparameters of the learning rate, local and global epochs. Second, the proposed technique is utilized within a cyber defense framework to defend against adversarial attacks. The performance of the proposed framework was evaluated using two benchmark datasets: MINST and Fashion-MINST, where they include images that would be extracted from smart cameras of AVs. This framework is shown to be more resilient against various adversarial attacks compared with peer techniques.

Visual scene recognition is an important problem in artificial intelligence with applications in areas such as autonomous vehicles, visually impaired people assistance, augmented reality, and many other pattern recognition areas. Visual scene recognition has been tackled in recent years by means of image descriptors such as the popular Speeded-Up Robust Features (SURF) algorithm. The problem consists in analyzing the scenes in order to produce a compact representation based on a set of so called regions of interest (ROIs) and then finding the largest number of matches among a dataset of reference images that include non-affine transformations of the scenes. In this paper, a new form of descriptors based on moment invariants from Gegenbauer orthogonal polynomials is presented. Their computation is efficient and the produced feature vector is compact, containing only a couple dozens of values. Our proposal is compared against SURF by means of the recognition rate computed on a set of two hundred scenes containing challenging conditions. The experimental results show no statistically significant difference between the performances of the descriptors.

Recommendation models typically follow a discriminative paradigm, predicting whether items should be retrieved. While effective, the expressive capabilities of these recommender systems are limited. Users can only passively browse the recommended items rather than actively express their needs and engage in an interactive experience. With recent advances in generative models such as large language models, a paradigm shift is happening in the study of recommender systems. Researchers propose building generative recommendation models either by aligning pre-trained generative models with user behaviors or designing recommendation models within a generative framework. These models enable the systems to receive and deliver more human-like content, such as natural language, images, and beyond. In this tutorial, we first provide an overview of the latest progress in generative recommendation models, covering approaches based on large language models, semantic IDs, diffusion models, and more. We then make an in-depth discussion on the challenges, open questions, and potential future directions in developing generative recommendation models.

Generative artificial intelligence (AI) has the potential to transform creative industries through supporting human creative ideation—the generation of new ideas1, 2, 3, 4–5. However, limitations in model capabilities raise key challenges in integrating these technologies more fully into creative practices. Iterative tweaking and divergent thinking remain key to enabling creativity support using technology6,7, yet these practices are insufficiently supported by state-of-the-art generative AI models. Using game development as a lens, we demonstrate that we can make use of an understanding of user needs to drive the development and evaluation of generative AI models in a way that aligns with these creative practices. Concretely, we introduce a state-of-the-art generative model, the World and Human Action Model (WHAM), and show that it can generate consistent and diverse gameplay sequences and persist user modifications—three capabilities that we identify as being critical for this alignment. In contrast to previous approaches to creativity support tools that required manually defining or extracting structure for relatively narrow domains, generative AI models can learn relevant structure from available data, opening the potential for a much broader range of applications.