Cognitive Systems Research最新文献

Biologically plausible neural networks have demonstrated efficiency in learning and recognizing patterns in data. This paper proposes a general online unsupervised algorithm for spatial data encoding using fast Hebbian learning. Inspired by the Hierarchical Temporal Memory (HTM) framework, we introduce the SpatialEncoder algorithm, which learns the spatial specialization of neurons’ receptive fields through Hebbian plasticity and k-WTA (k winners take all) inhibition. A key component of our model is a two-part synaptogenesis algorithm that enables the network to maintain a sparse connection matrix while adapting to non-stationary input data distributions. In the MNIST digit classification task, our model outperforms the HTM SpatialPooler in terms of classification accuracy and encoding stability. Compared to another baseline, a two-layer artificial neural network (ANN), our model achieves competitive classification accuracy with fewer iterations required for convergence. The proposed model offers a promising direction for future research on sparse neural networks with adaptive neural connectivity.

Serious games have a wide range of applications. Modeling virtual character behaviors and emotions is a challenging task in developing serious games. To generate real-time responses, behavioral and emotional models must be simple and effective. Existing studies have paid little attention to the semantic understanding of virtual characters to external stimuli and have not effectively linked perceived semantics and motivation. This paper proposes a cognitive structure for the virtual character. The structure contains multiple modules: perception, personality, motivation, behavior, and emotion. Based on psychological theory, a semantic table that connects external stimuli, motivations, behaviors, and emotions is designed for each virtual character. Perceptivity is introduced to measure the degree of perception. According to Maslow’s motivation theory, a quantitative description of motivation is given and a discriminating method is proposed to generate behaviors and emotions. A prototype of a serious game is developed to verify the validity of the proposed method. The experimental results show that the proposed method can simulate the behavior and emotion of virtual characters in real time and will enhance the immersion of serious games.

Simulating crowd motion in emergency scenarios remains a challenge in computer graphics due to crowd heterogeneity and environmental complexity. However, existing crowd simulation methods homogenize the agent model and simplify target selection and motion navigation of emergency crowds. To address these problems, we propose a multi-agent motion simulation method for emergency scenario deduction. First, we propose a multi-agent model to simulate crowd heterogeneity. This model includes a personality-based heterogeneous agent model and an agent perception model that considers vision, hearing, and familiarity with the environment. Second, we propose a target selection strategy based on the motion patterns of actual pedestrians. This strategy employs mathematical models and our agent perception model to guide agents in selecting appropriate targets. Finally, we propose a global navigation algorithm that combines random sampling with heuristic search methods. Concurrently, we use our multi-agent model to adjust the agent’s local motion planning to deduce the motion states of emergency crowds naturally. Experimental results validate that our method can realistically and reasonably simulate crowd motion in emergency scenarios.

In emergency situations, social media platforms produce a vast amount of real-time data that holds immense value, particularly in the first 72 h following a disaster event. Despite previous efforts, efficiently determining the geographical location of images related to a new disaster remains an unresolved operational challenge. Currently, the state-of-the-art approach for dealing with these first response mapping is first filtering and then submitting the images to be geolocated to a volunteer crowd, assigning the images randomly to the volunteers. In this work, we extend our previous paper (Ballester et al., 2023) to explore the potential of artificial intelligence (AI) in aiding emergency responders and disaster relief organizations in geolocating social media images from a zone recently hit by a disaster. Our contributions include building two different models in which we try to (i) be able to learn volunteers’ error profiles and (ii) intelligently assign tasks to those volunteers who exhibit higher proficiency. Moreover, we present methods that outperform random allocation of tasks, analyze the effect on the models’ performance when varying numerous parameters, and show that for a given set of tasks and volunteers, we are able to process them with a significantly lower annotation budget, that is, we are able to make fewer volunteer solicitations without losing any quality on the final consensus.

Empathy plays an important role in human conversations as an ability that enables individuals to understand the emotions and situations of others. Integrating empathy into dialogue systems is a crucial step in making them humanized. Relevant psychological studies have shown that a complete, high-quality empathetic dialogue should consist of the following two stages: (1) Empathetic Perception: the listener needs to perceive the emotional state of the speaker from both cognitive and affective aspects; (2) Empathetic Expression: the appropriate expression is chosen to respond to the perceived information. However, many existing studies on empathetic response generation only focus on one of these stages, resulting in incomplete and insufficiently empathetic responses. To this end, we propose the EmpCI, a two-stage empathetic response generation model that utilizes commonsense knowledge and mixed empathetic intent, respectively. Specifically, we use commonsense knowledge in the first stage to enhance the model’s perception of the user’s emotion and introduce mixed empathetic intent in the second stage to generate responses with appropriate expressions for the perceived information. Finally, we evaluated the EmpCI on the EmpatheticDialogues dataset, and extensive experiment results show that the proposed model outperforms the baselines in both perceiving users’ emotions and generating empathetic responses.

In the digital age, people prefer digital content, but screen-related health concerns like eye strain and blue light emerge. Legibility gains importance in digital text, especially in fields like optometry and for those with low vision. Therefore, having good letter recognition ensures better readability of words and written language in general. This work focuses on defining three typeface legibility indices from the judgements of a group of 31 observers. Those indices are based on statistics, confusion matrices, and power indices from game theory. As far as we know, this is the first time that typeface legibility indices have been defined using game theory. These indices help us to globally assess how legible is a typeface. We apply them to three commonly used typefaces (Roboto, Helvetica and Georgia), and to a new one developed for the authors (Optotipica 5 v2022). This comparison helps us understand which typefaces are more legible according to the defined indices on digital screens. The major conclusions are: (1) The three indices are highly consistent pairwise; (2) Helvetica is the most legible typeface for uppercase letters, whilst Optotipica is the most legible for lowercase; (3) the two cases of Helvetica exhibit uniform high legibility metrics, ensuring optimal recognition regardless of letter case.

In 2020, Mc Fadden published an article in which he discusses how algorithms can be encoded in time and space. By analyzing the topology of the cytoarchitecture of the brain, cognitive architectures can understand the underlying mechanisms that have led to the development of human intelligence in space. In this study, our focus lies in investigating temporal heterogeneity as a mechanism that the brain could have developed not solely as a biological constraint, but also as an evolutionary advantage. To accomplish this, we employed virtual agents within a virtual environment and constructed a prototype cognitive architecture. Subsequently, we compared the benefits and drawbacks of having this cognitive architecture operate under a model of temporal heterogeneity versus one characterized by temporal homogeneity. At the conclusion of the article, we present the results obtained from two perspectives. From a quantitative standpoint, we contrast the agents’ adaptation to the environment based on the cognitive architecture model employed by each agent. On this front, we found evidence that temporal heterogeneity might be useful in finding parameter optimizations faster, amongst other benefits. From a qualitative perspective, we examine the potential of this model to explore the cognitive processes of the virtual agents, concluding that a different representation of percepts is needed, which we further discuss.

Digital Twin (DT) will transform digital healthcare and push it far beyond expectations. DT creates a virtual representation of a physical object reflecting its current state using real-time converted data. Nowadays, Women’s health is more frequently impacted by cervical cancer, but early detection and rapid treatment are critical factors in the cure of cervical cancer. This paper proposes and implements an automated cervical cancer detection DT framework in healthcare. This framework is a valuable approach to enhance digital healthcare operations. In this proposed work, the SIPaKMeD dataset was used for multi-cell classification. There were 1013 images (Input size 224 × 224 × 3) in the collection, from which 4103 cells could be extracted. As a result, the CervixNet classifier model is developed using machine learning to detect cervical problems and diagnose cervical disease. Using pre-trained recurrent neural networks (RNNs), CervixNet extracted 1172 features, and after that, 792 features were selected using an independent principal component analysis (PCA) algorithm. The implemented models achieved the highest accuracy for predicting cervical cancer using different algorithms. The collected information has shown that integrating DT with the healthcare industry will enhance healthcare procedures by integrating patients and medical staff in a scalable, intelligent, and comprehensive health ecosystem. Finally, the suggested method produces an impressive 98.91 % classification accuracy in all classes, especially for support vector machines (SVM).

As humans, we have an excellent performance when perceiving the environment. In the artificial world, it is important for machines to perceive their environment so they can make correct decisions and act accordingly. An essential process to accomplish perception is to identify objects in a scene, but, as in reality, these objects can appear as ambiguous, and additionally, those objects are embedded into a particular scene. For our proposal, we created an architecture to identify ambiguous objects by using scene information to guide the identification process. The design is based on the human cortical systems that participate in object and scene recognition. In our study, we validate this proposal by analyzing a prior human experiment that demonstrates and quantifies the impact of scene information on ambiguous objects. Our findings demonstrate that employing the presented architecture on an object recognition task results in superior machine performance with familiar scenes, as opposed to unfamiliar or absent ones, consistent with human behavior.

Automation of teaching people new skills requires modeling of human reasoning because human cognition involves active reasoning over the new subject domain to acquire skills that will later become automatic. The article presents Thought Process Trees — a language for modeling human reasoning that was created to facilitate the development of intelligent tutoring systems, which can perform the same reasoning that is expected of a student and find deficiencies in their line of thinking, providing explanatory messages and allowing them to learn from performance errors. The methodology of building trees which better reflect human learning is discussed, with examples of design choices during the modeling process and their consequences. The characteristics of educational modeling that impact building subject-domain models for intelligent tutoring systems are discussed. The trees were formalized and served as a basis for developing a framework for constructing intelligent tutoring systems. This significantly lowered the time required to build and debug a constraint-based subject-domain model. The framework has already been used to develop five intelligent tutoring systems and their prototypes and is being used to develop more of them.