Cognitive Systems Research最新文献

Whenever someone in a team tries to help others, it is crucial that they have some understanding of other team members’ goals. In modern teams, this applies equally to human and artificial (“bot”) assistants. Understanding when one does not know something is crucial for stopping the execution of inappropriate behavior and, ideally, attempting to learn more appropriate actions. From a statistical point of view, this can be translated to assessing whether none of the hypotheses in a considered set is correct. Here we investigate a novel approach for making this assessment based on monitoring the maximum a posteriori probability (MAP) of a set of candidate hypotheses as new observations arrive. Simulation studies suggest that this is a promising approach, however, we also caution that there may be cases where this is more challenging. The problem we study and the solution we propose are general, with applications well beyond human–bot teaming, including for example the scientific process of theory development.

This article presents the use of second-order adaptive network models of hospital teams consisting of doctors and nurses, interacting together. A variety of scenarios are modelled and simulated, in relation with respiratory distress of a neonate, along with the integration of an AI-Coach for monitoring and support of such teams and of organizational learning. The research highlights the benefits of introducing a virtual AI-Coach in a hospital setting. The practical application setting revolves around a medical team responsible for managing neonates with respiratory distress. In this setting an AI-Coach act as an additional team member, to ensure correct execution of medical procedure. Through simulation experiments, the adaptive network models demonstrate that the AI-Coach not only aids in maintaining correct medical procedure execution but also facilitates organizational learning, leading to significant improvements in procedure adherence and error reduction during neonatal care.

The integration of robotics into everyday life is increasing and these complex systems are exposed to complex faults that require rapid identification for seamless repair and continuous operation. These faults have a complex impact on cognitive aspects such as perception, decision-making and behavioral execution in robots. Robotic fault detection and diagnosis research (FDD) focuses primarily on individual robot scenarios, which lack a comprehensive investigation in multi-robot systems (MRSs). Our paper introduces a robotic control method to control operations in a wide range of production systems. The control system architecture developed by multiple robots provides a local and global cognitive system that is shared between them. Internal dynamics, represented by finite state machines, represent different operating scenarios. The rigorous formal methodology such as Petri Nets and Computer Tree Logic (CTL) validates the accuracy of control architectures and fault management strategies. Building a model of trust based on the historical interactions between intelligent robots facilitates the creation of a global cognitive system that enables adaptation in the management of errors. Our research is launching a trust estimation model, especially the collaboration between reliable robots, and increasing the fault flexibility of multirobot control systems. The contributions include the design of multi-robot control architectures, the management of failures of control robots, and the formulation of trust models.

Compositionality can be considered as finding (or creating) the correct meaning of the constituents of a non-simple language expression or visual image. The Causal Cognitive Architecture is a brain-inspired cognitive architecture (BICA). It is not a traditional artificial neural network architecture, nor a traditional symbolic AI system but instead uses spatial navigation maps as its fundamental circuits. In previously described versions of the architecture, sensory inputs are compared in each existing sensory system against previous stored navigation maps for that sensory system, and the best navigation map is chosen and then updated with the new sensory inputs and a best multisensory navigation map is similarly created and used as the working navigation map. Instinctive and learned small procedures are triggered by input sensory inputs as well as matched navigation maps, and in the Navigation Module operate on the working navigation map and produce an output signal. By feeding back intermediate results in the Navigation Module it has been shown previously how causal and analogical behaviors emerge from the architecture. In new work, the Navigation Module is duplicated in a biologically plausible manner. It becomes possible to compositionally process information in the duplicated Navigation Module, and as a result compositional language comprehension and behavior readily emerge. A formalization and simulation of the architecture is presented. A demonstration example, and its negation, are explored of solving a compositional problem requiring the placement of an object in a specific location with regard to other objects. Future work is discussed using large language models to create navigation maps. Given the mammalian brain inspiration of the architecture, it suggests that it is indeed feasible for modest genetic changes to have allowed the emergence of compositional language in humans.

Vector database management systems have emerged as an important component in modern data management, driven by the growing importance for the need to computationally describe rich data such as texts, images and video in various domains such as recommender systems, similarity search, and chatbots. These data descriptions are captured as numerical vectors that are computationally inexpensive to store and compare. However, the unique characteristics of vectorized data, including high dimensionality and sparsity, demand specialized solutions for efficient storage, retrieval, and processing. This narrative literature review provides an accessible introduction to the fundamental concepts, use-cases, and current challenges associated with vector database management systems, offering an overview for researchers and practitioners seeking to facilitate effective vector data management.

Rett Syndrome (RTT) is a rare neurodevelopmental disorder caused by mutation in the MECP2 gene. No cures are still available, but several clinical trials are ongoing. Here we examine neurophysiological correlates of auditory processing for ability to differentiate patients with RTT from typically developing (TD) peers applying standard machine learning (ML) methods and pipelines. Capitalized on the available event-related potential (ERP) data recorded in response to tone presented at different rates (stimulus onset asynchrony 900, 1800 and 3600 ms) from 24 patients with RTT and 27 their TD peer. We considered the most common ML models that are widely used for classification tasks. These include both linear models (logistic regression, support-vector machine with linear kernel) and tree-based nonlinear models (random forest, gradient boosting). Based on these methods we were able to differentiate RTT from TD children with high accuracy (with up to 0.94 ROC-AUC score), which was evidently higher at the fastest presentation rate. Importance analysis and perturbation importance pointed out that the most important feature for classification is P2-N2 peak-to-peak amplitude, consistently across the approaches and blocks with different presentation rate. The results suggest the unique pattern of ERP characteristics for RTT and points to features of importance. The results might be relevant for establishing outcome measures for clinical trials.