Nowadays, people spend a lot of time in the information space, communicating within various social platforms. Content of those platforms can influence people’s feelings and personalities, which is especially relevant for young people. In this research, we made an attempt to prove this hypothesis. For the experiment, we selected the VKontakte social network and analysed users profiles together with the results of the psychological tests passed by them. The goal of the experiment was to find correlations between the information provided within the social network communities and the users’ personalities. Moreover, in this paper, we made an attempt to enhance the results of the classifier accuracy using the sentiment analysis. The experiments were conducted to test the sentiment analysis models, to analyse the proposed feature based on posts’ sentiment, and test the classifier for the detection of the potentially destructive impacts. The analysis of the correlation of the proposed feature with the communities that have potentially destructive impacts on anxiety is conducted. The analysis of the obtained results is provided. During the experiments, the authors found out that consideration of the posts’ sentiment allows increasing accuracy of the classifier for anxiety destructive impacts on 12.24 %. Additionally, we analysed the relationship between the user sentiments metric and destructiveness. We confirmed that the assessment of the user’s posts’ sentiment can be used to compile his psychological characteristics and determine possibility of destructiveness.
When the sensor dynamically collects point cloud data for object or map reconstruction, the registration effect is poor and reconstruction application is difficult with a too low overlap rate of the collected point cloud data. The reason is that the objects are covered, the sensor rotation angle is too large and the speed of movement is too fast. Because of these problems, this paper proposes a point cloud registration algorithm based on FPFH feature matching, combined with second-order spatial measures. Firstly, using the FPFH feature extraction algorithm, the features of each point are extracted, and then feature matching is performed to generate the set of feature point pairs. Secondly, the second-order spatial measure is used to calculate the set of feature point pairs to obtain the second-order spatial measure matrix scores and sort them. Finally, the dichotomy method is used to find the appropriate second-order spatial measure scores for distinguishing the inner points (points in the overlap region) from the outer points (points that do not belong to the overlap region as well as the mismatched points and some disturbances). The contrast experiments between this algorithm and three common point cloud registration algorithms, FPFH-ICP, 4PCS-ICP, and NDT-ICP, on the Stanford dataset and 3DMatch dataset shows that the registration accuracy of the other algorithms decreases significantly with a low overlap rate. But this algorithm still has a high registration accuracy and is less affected by outliers than the other algorithms. Besides, this algorithm can still maintain a good registration effect on different data sets.
Pedestrian intent prediction is an essential task for ensuring the safety of pedestrians and vehicles on the road. This task involves predicting whether a pedestrian intends to cross a road or not based on their behavior and surrounding environment. Previous studies have explored feature-based machine learning and vision-based deep learning models for this task but these methods have limitations in capturing the global spatio-temporal context and fusing different features of data effectively. To address these issues, we propose a novel hybrid framework HSTGCN for pedestrian intent prediction that combines spatio-temporal graph convolutional neural networks (STGCN) and long short-term memory (LSTM) networks. The proposed framework utilizes the strengths of both models by fusing multiple features, including skeleton pose, trajectory, height, orientation, and ego-vehicle speed, to predict their intentions accurately. The framework’s performance have been evaluated on the JAAD benchmark dataset and the results show that it outperforms the state-of-the-art methods. The proposed framework has potential applications in developing intelligent transportation systems, autonomous vehicles, and pedestrian safety technologies. The utilization of multiple features can significantly improve the performance of the pedestrian intent prediction task.
Diversity plays a major role when a student is looking for a tutor to better understand some concepts or an entire course. In fact, algorithms for recommending potential tutors have to take into account several aspects of diversity that may be critical to successful tutoring. The tutor must have the appropriate competencies in the subject matter. In addition, he/she must be able to convey the knowledge and skills associated with the tutoring topic. Moreover, the personality traits of the tutee and the tutor can facilitate or hinder the learning process during tutoring. This study presents the experience of ‘SOS TUTORÍA UC’, a responsive web application aimed at facilitating academic assistance among students. Special emphasis is placed on the importance of incorporating dimensions of diversity that can inform the artificial intelligence algorithms of the potential tutor recommendation system. Indeed, competence in the topic tutored is the first diversity criterion for recommending more knowledgeable potential tutors. In addition, the tutee has to choose whether to look for tutors with personalities that are “different”, “similar” or “indifferent” to his or her own preferences for academic assistance on the specific topic. To achieve this, ‘SOS TUTORÍA UC’ is integrated with the WeNet platform, which provides user management services and user recommendation algorithms. The results of the testing of the recommendation system were positive with regard to the criterion of competence, while the criterion of personality should be addressed for improvement. In order to improve the tutor-tutee matching process, participants emphasized the importance of considering the criterion of personality traits in addition to competence. They also requested additional information and parameters to facilitate tutor selection.
Human action recognition has been widely used in fields such as human–computer interaction and virtual reality. Despite significant progress, existing approaches still struggle with effectively integrating hierarchical information and processing data beyond a certain frame count. To address these challenges, we introduce the Multi-AxisFormer (MAFormer) model, which is organized in terms of spatial, temporal, and channel dimensions of the action sequence, thereby enhancing the model’s understanding of correlations and intricate structures among and within features. Drawing on the Transformer architecture, we propose the Cross-channel Spatio-temporal Aggregation (CSA) structure for more refined feature extraction and the Multi-Axis Attention (MAA) module for more comprehensive feature aggregation. Moreover, the integration of Rotary Position Embedding (RoPE) boosts the model’s extrapolation and generalization abilities. MAFormer surpasses the known state-of-the-art on multiple skeleton-based action recognition benchmarks with the accuracy of 93.2% on NTU RGB+D 60 cross-subject split, 89.9% on NTU RGB+D 120 cross-subject split, and 97.2% on N-UCLA, offering a novel paradigm for hierarchical modeling in human action recognition.
A long-standing challenge in artificial intelligence is lifelong reinforcement learning, where learners are given many tasks in sequence and must transfer knowledge between tasks while avoiding catastrophic forgetting. Policy reuse and other multi-policy reinforcement learning techniques can learn multiple tasks but may generate many policies. This paper presents two novel contributions, namely 1) Lifetime Policy Reuse, a model-agnostic policy reuse algorithm that avoids generating many policies by optimising a fixed number of near-optimal policies through a combination of policy optimisation and adaptive policy selection; and 2) the task capacity, a measure for the maximal number of tasks that a policy can accurately solve. Comparing two state-of-the-art base-learners, the results demonstrate the importance of Lifetime Policy Reuse and task capacity based pre-selection on an 18-task partially observable Pacman domain and a Cartpole domain of up to 125 tasks.
Autonomous behaviors may raise ethical issues that agents must consider in their reasoning. Some approaches use deontic logics, while others consider a value-based argumentation framework. However, no work combines both modal logic and argumentation to reason about ethics. Hence, we propose a new argumentation framework where arguments are built from a n-ary multi-modal logic. It allows to express different kinds of operators, e.g. nullary choice or moral worth operators, dyadic deontic operators or mental states. However the standard attacks based on logical contradictions are no longer sufficient to catch an intuitive meaning for attacks. Hence, we enrich standard attacks by characterizing how oppositions between modal operators arise. Furthermore we show the standard logic-based attacks have a quasi-symmetry property, i.e. when an argument attacks another, this argument is necessarily attacked by another one. Our modal attacks do not have this property, which is highly relevant to decide a dilemma.
In this paper, a novel approach for transductive classification is proposed. Unlike existing methods that heavily rely on constructing the Laplacian matrix to capture data distribution, the proposed approach takes a unique path. It employs a linear transformation model to create local patches for each data point and then unifies them in an objective function to build the Laplacian matrix. Incorporating this Laplacian matrix into the transductive classification framework allows us to assign optimal class labels globally. The experimental results from toy data and real world databases demonstrate that the proposed approach achieves more efficient and stable performance, while this approach is insensitive to the parameters. Notably, our method exhibits robustness to parameter variations, making it highly adaptable to practical applications.
Swarm intelligence optimization algorithm has been proved to perform well in the field of parameter optimization. In order to further improve the performance of intelligent optimization algorithm, this paper proposes an improved and adaptive tunicate swarm algorithm (IMATSA) based on tunicate swarm algorithm (TSA). IMATSA improves TSA in the following four aspects: population diversity, local search convergence speed, jumping out of local optimal position, and balancing global and local search. Firstly, IMATSA adopts Tent map and quadratic interpolation to initialize population and enhance the diversity. Secondly, IMATSA uses Golden-Sine algorithm to accelerate the convergence of local search. Thirdly, in the process of global development, IMATSA adopts Levy flight and the improved Gauss disturbance method to adaptively improves and coordinates the ability of global development and local search. Then, this paper verifies the performance of IMATSA based on 14 benchmark functions experiment, ablation experiment, parameter optimization experiments of Support Vector Machine (SVM) and Gradient Boosting Decision Tree (GBDT), Wilcoxon signed rank test and image multi-threshold segmentation experiment with the performance metrics are convergence speed, convergence value, significance level P-value, Peak Signal-to-Noise Ratio (PSNR) and Standard Deviation (STD). Experimental results show that IMATSA performs better in three kinds of benchmark functions; each component of IMATSA has a positive effect on the performance; IMATSA performs better in parameter optimization experiments of SVM experiment and GBDT; there is significant difference between IMATSA and other algorithms by Wilcoxon signed rank test; in image segmentation, the performance is directly proportional to the number of thresholds, and compared with other algorithms, IMATSA has better comprehensive performance.
In this work, we explore the links between the Borda voting rule and belief merging operators. More precisely, we define two families of merging operators inspired by the definition of the Borda voting rule. We also introduce a notion of cancellation in belief merging, inspired by the axiomatization of the Borda voting rule proposed by Young. This allows us to provide a characterization of the drastic merging operator and of a family of merging operators defined in a way which is similar to the Borda rule.
We present an approach to autonomous drone racing inspired by how a human pilot learns a race track. Human pilots drive around the track multiple times to familiarise themselves with the track and find key points that allow them to complete the track without the risk of collision. This paper proposes a three-stage approach: exploration, navigation, and refinement. Our approach does not require prior knowledge about the race track, such as the number of gates, their positions, and their orientations. Instead, we use a trained neural pilot called DeepPilot to return basic flight commands from camera images where a gate is visible to navigate an unknown race track and a Single Shot Detector to visually detect the gates during the exploration stage to identify points of interest. These points are then used in the navigation stage as waypoints in a flight controller to enable faster flight and navigate the entire race track. Finally, in the refinement stage, we use the methodology developed in stages 1 and 2, to generate novel data to re-train DeepPilot, which produces more realistic manoeuvres for when the drone has to cross a gate. In this sense, similar to the original work, rather than generating examples by flying in a full track, we use small tracks of three gates to discover effective waypoints to be followed by the waypoint controller. This produces novel training data for DeepPilot without human intervention. By training with this new data, DeepPilot significantly improves its performance by increasing its flight speed twice w.r.t. its original version. Also, for this stage 3, we required 66 % less training data than in the original DeepPilot without compromising the effectiveness of DeepPilot to enable a drone to autonomously fly in a racetrack.
We present an epistemic language for representing an artificial player’s beliefs and actions in the context of the Yōkai board game. Yōkai is a cooperative game which requires a combination of Theory of Mind (ToM), temporal and spatial reasoning to be played effectively by an artificial agent. We show that the language properly accounts for these three dimensions and that its satisfiability problem is NP-complete. This opens up the possibility of exploiting SAT techniques for automating reasoning of an artificial player in the context of the Yōkai board-game.
The exponential growth of technology in recent decades has led to the emergence of some challenges inherent to this growth. One of these challenges is the enormous amount of data collected by the different sensors in our society, namely in management processes such as Wastewater Treatment Plants (WWTPs). These infrastructures comprise several processes to treat wastewater and discharge clean water in water courses. Therefore, the concentration of pollutants must be below the allowable emissions limits. In this work, anomaly detection models were conceived, tuned and evaluated to monitor essential parameters such as nitrate and ammonia concentrations and pH to improve WWTP management. Four Machine Learning models were considered, particularly Local Outlier Fraction, Isolation Forest, One-Class Support Vector Machines and Long Short-Term Memory-Autoencoders (LSTM-AE), to detect anomalies in the three parameters mentioned. Through the different experiments, it was possible to verify that, in terms of F1-Score, the best candidate model for the three analyzed parameters was LSTM-AE-based, with a value consistently higher than 97%.
Static node embedding algorithms applied to snapshots of real-world applications graphs are unable to capture their evolving process. As a result, the absence of information about the dynamics in these node representations can harm the accuracy and increase processing time of machine learning tasks related to these applications. Aiming at fill the gap regarding the inability of static methods to capture evolving processes on dynamic networks, we propose a biased random walk method named Evolving Node Embedding (EVNE). EVNE leverages the sequential relationship of graph snapshots by incorporating historic information when generating embeddings for the next snapshot. It learns node representations through a neural network, but differs from existing methods as it: (i) incorporates previously run walks at each step; (ii) starts the optimization of the current embedding from the parameters obtained in the previous iteration; and (iii) uses two time-varying parameters to regulate the behavior of the biased random walks over the process of graph exploration. Through a wide set of experiments we show that our approach generates better embeddings, outperforming baselines by up to 20% in a downstream node classification task. EVNE’s embeddings achieve better performance than others, based on experiments with four classifiers and five datasets. In addition, we present seven variations of our model to show the impact of each of EVNE’s mechanisms.
In this special issue of AI Communications, we converge upon a selection of extended manuscripts emanating from the 17th Ibero-American Conference on Artificial Intelligence.Each manuscript reflects the scholarly rigor and quest for applied innovation in our field.The chosen contributions embody incremental advancements in AI and present solutions to some of the most pressing challenges in the domain.Carrillo, de Wolff, Martí, and Sanchez-Pi's paper presents a novel multi-objective optimization framework for Physics-Informed Neural Networks (PINNs), heralding a refined approach that synergizes data with physical laws to advance predictive models -a vital intersection of AI and physics.In tackling the dynamism of network data, Enes, Nunes, Murai, and Pappa introduce a temporal embedding method that captures the evolution of networks, offering a solution to the stagnant nature of traditional embeddings and ushering in improved processing for machine learning applications.Yokoyama, Ferro, and Schulze's research navigates the multi-faceted objectives of machine learning efficiency, making strides towards a more sustainable AI with their genetic algorithm-based hyperparameter optimization -an essential consideration in an era of environmental consciousness.Addressing data-driven environmental management, Oliveira, Duarte, and Novais apply anomaly detection models to wastewater treatment monitoring, showcasing AI's potential to support and enhance ecological compliance and sustainability.Rojas-Perez and Martinez-Carranza's work on autonomous drone navigation parallels the human experience of learning and adaptation, translating it into AI to push the boundaries of autonomous vehicular technology and its real-world applicability.These papers collectively advance our understanding of AI's role in diverse problem-solving contexts, reinforcing the conference's reputation as a crucible for innovation.As these extended versions attest, the research presented at IBERAMIA 2022 transcends the theoretical, offering meaningful, actionable intelligence for the complex world that AI is poised to navigate and transform.
In a recently published book, the French writer and comedian François Rollin has discussed various aspects of the notion of stupidity, including artificial stupidity, the stupid counterpart of artificial intelligence. His claim is that a system of artificial stupidity is a system that provides wrong answers to any task it should solve, leading to absurd solutions in most cases. We believe that this claim is (at least partially) false and that designing artificial stupidity is not as trivial as it seems. In this article, we discuss why and how one could design a system of artificial stupidity. We believe that such a reflection on (artificial) stupidity can bring about some interesting insights about (artificial) intelligence.
This paper presents a multi-objective optimization approach for developing efficient and environmentally friendly Machine Learning models. The proposed approach uses Genetic Algorithms to simultaneously optimize the accuracy, time-to-solution, and energy consumption simultaneously. This solution proposed to be part of an Automated Machine Learning pipeline and focuses on architecture and hyperparameter search. A customized Genetic Algorithm scheme and operators were developed, and its feasibility was evaluated using the XGBoost ML algorithm for classification and regression tasks. The results demonstrate the effectiveness of the Genetic Algorithm for multi-objective optimization, indicating that it is possible to reduce energy consumption while minimizing predictive performance losses.
The majority of existing rotating target detectors inherit the horizontal detection paradigm and design the rotational regression loss based on the inductive paradigm. But the loss design limitation of the inductive paradigm makes these detectors hardly detect effectively tiny targets with high accuracy, particularly for large-aspect-ratio objects. Therefore, in view of the fact that horizontal detection is a special scenario of rotating target detection and based on the relationship between rotational and horizontal detection, we shift from an inductive to a deductive paradigm of design to develop a new regression loss function named Gauss–Wasserstein scattering (GWS). First, the rotating bounding box is transformed into a two-dimensional Gaussian distribution, and then the regression losses between Gaussian distributions are calculated as the Wasserstein scatter; By analyzing the gradient of centroid regression, centroid regression is shown to be able to adjust gradients dynamically based on object characteristics, and small targets requiring high accuracy detection rely on this mechanism, and more importantly, it is further demonstrated that GWS is scale-invariant while possessing an explicit regression logic. The method is performed on a large public remote sensing dataset DOTA and two popular detectors and achieves a large accuracy improvement in both large aspect ratio targets and small targets detection compared to similar methods.
