Cognitive Systems Research最新文献

In the last years, the System 1/System 2 cognitive architecture, proposed by psychologist Daniel Kahneman, raised the interest of many researchers in the field. System 1 is an intuitive, automatic, and fast-thinking system working effortlessly, without conscious effort. System 2 is a deliberate, analytical, and slower-thinking system employing conscious effort and attention. This work proposes an innovative approach that exploits techniques typical of information retrieval (the trie data structure) to efficiently encode the solutions’ repository at the border between System 2 and System 1. This repository stores the solutions (successful plans) the agent has already used and can re-enact to achieve the goals. System 2 conceives new plans and delegates System 1 to execute them. If the plan is successful (and so it becomes a solution), System 1 stores that in the repository to quickly retrieve any solution that may help fulfil the goals deliberated by System 2 in the future.

This article presents a transdisciplinary analysis of the challenges in fusing neuroscience concepts with artificial intelligence (AI) to create AI systems inspired by biological cognition. We explore the structural and functional disparities between the neocortex’s canonical microcircuits and existing AI models, focusing on architectural differences, learning mechanisms, and energy efficiency. The discussion extends to adapting non-goal-oriented learning and dynamic neuronal connections from biological brains to enhance AI’s flexibility and efficiency. This work underscores the potential of neuroscientific insights to revolutionize AI development, advocating for a paradigm shift towards more adaptable and brain-like AI systems. We conclude that there is major room for bioinspiration by focusing on developing architecture, objective functions, and learning rules using a local instead of a global approach.

The Human decision-making process works by recollecting past sequences of observations and using them to decide the best possible action in the present. These past sequences of observations are stored in a derived form which only includes important information the brain thinks might be useful in the future, while forgetting the rest. we propose an architecture that tries to mimic the human brain and improve the memory efficiency of transformers by using a modified TransformerXL architecture which uses Automatic Chunking which only attends to the relevant chunks in the transformer block. On top of this, we use ForgetSpan which is technique to remove memories that do not contribute to learning. We also theorize the technique of Similarity based forgetting to remove repetitive memories. We test our model in various tasks that test the abilities required to perform well in a human–robot collaboration scenario.

As part of cognitive intelligence, concept cognition for knowledge graphs aims to clearly grasp the typical characteristics of the things referred to by the concept, which can provide prior knowledge for machine understanding and thinking. Different from concept learning and formal concept analysis that learn new concepts from data and the general decision rule that comes from an independent decision table, this paper cognizes an existing concept by decision rules that come from multiple granularities. Specifically, 1) concept cognition for knowledge graphs is realized from the perspective of mining multi-granularity decision rule. 2) Decision tables corresponding to four granularities form a multi-granularity decision table group, and then the result from coarser granularity can guide and help obtaining the result from finer granularity. 3) We propose a framework for mining multi-granularity decision rules, which involves going from a multi-granularity decision table group to the frequent maximal attribute patterns to the decision rules to the credible decision rules. Finally, we verified effectiveness of dividing positive and negative data, monotonicity of attribute patterns in a multi-granularity decision table group, and downward monotonicity of credibility, and observed the impact of the parameter min_cov and min_conf on execution times.

Music has been used to express and communicate emotional states through its different dimensions such as tone, rhythm, melody, and harmony. Consonant harmonies consistently are rated as pleasant whereas dissonant are considered unpleasant. The aim of this study was to explore the effect of consonant and dissonant musical harmonies used as prime on the emotional classification of images, as indexed by event-related potentials. Thirty volunteers (ages 21–27, 50 % women) were presented with a task consisting of 4 musical intervals in the C major scale, divided into consonant and dissonant harmonies, followed by 180 positive, negative, or neutral images from the International Affective Picture System (IAPS). Participants had to rate the images as pleasant or unpleasant. We found a bias effect on negative images rated as positive when preceded by a consonant musical interval. A N200 component, non-sensible to the valence of the images, was found. On the other hand, a significant difference was found in the amplitude of the P300 component, with a greater amplitude in the consonant-positive images condition compared to the dissonant-positive images. Lastly, a late positivity component around 500–700 ms was found in both negative conditions dissonant and consonant, but with a larger amplitude for the consonant condition when followed by a negative image. These results indicate that additionally to the P300 processing the relevance of the stimulus there are processes like recognition memory involved. As part of the novelty effect this late positive activity may also be related to the emotional content integration of the relevant stimulus.

The hypothesis-driven methodology is a cognitive activity used in expertise processes to solve problems with limited knowledge and understanding. Although some organizations have standardized this approach to guide humans in carrying out expertise in enterprises, it lacks appropriate tools to assist experts in carrying out this cognitive activity, tracking understanding, or capturing the reasoning steps and the knowledge produced during the process.

To acquire, share and reuse experts’ knowledge applied during expertise processes while assisting humans in bringing understanding to complex problems, this study introduces a human–machine collaborative framework that formalizes experts’ knowledge from the hypothesis-driven methodology described in the France standard NF X50-110 of “Quality of expertise activity”. This framework utilizes Hypothesis Theory extended with qualitative doubt and a systematic reasoning process to generate a hypothesis exploratory graph (HEG).

The proposed approach makes it easier to carry out expertise processes through a human–machine collaboration, offers a means to share and reuse knowledge from expertise, and provides expertise processes evaluation mechanisms. Furthermore, an experiment conducted on a use-case of expertise process verifies the feasibility and effectiveness of the approach.

Deep learning approaches to mind, brain, and behavior raise new philosophical and methodological questions about the nature of artificial intelligence (AI) and its relationship to biological cognitive systems. The articles in this Special Issue combine insights, results and methodologies from philosophy, psychology, AI, neuroscience, linguistics, and cognitive science more generally, to explore some of those questions, including the relation between deep learning models and the brain, the testability, transparency and explanatory power of deep learning models, and their abilities for inductive reasoning, language processing and semantic understanding. By engaging with these foundational questions, the Special Issue as a whole contributes to illuminate deep learning, illustrating the need for, and fruitfulness of, interdisciplinary perspectives in cognitive systems research.

The growing field of explainable Artificial Intelligence (xAI) has given rise to a multitude of techniques and methodologies, yet this expansion has created a growing gap between existing xAI approaches and their practical application. This poses a considerable obstacle for data scientists striving to identify the optimal xAI technique for their needs. To address this problem, our study presents a customized decision support framework to aid data scientists in choosing a suitable xAI approach for their use-case. Drawing from a literature survey and insights from interviews with five experienced data scientists, we introduce a decision tree based on the trade-offs inherent in various xAI approaches, guiding the selection between six commonly used xAI tools. Our work critically examines six prevalent ante-hoc and post-hoc xAI methods, assessing their applicability in real-world contexts through expert interviews. The aim is to equip data scientists and policymakers with the capacity to select xAI methods that not only demystify the decision-making process, but also enrich user understanding and interpretation, ultimately advancing the application of xAI in practical settings.

We propose a model that conceptualises diagrammatic sensemaking and reasoning as blends of image schemas – patterns derived from our perceptual and embodied experiences and interactions with the environment – with the geometric structure of the diagram. Our ultimate goal is to develop an algorithmic method for determining several potential blends that hold cognitive value for observers. Building upon our formal, category-theoretic approach to conceptual blending, we extend it by formalising two governing principles of blending. These principles serve as guides for the blending process, directing the cognitive construction of the blend. As these principles may compete with each other and favour different blend structures, we argue that their combination leads to cognitively useful blends. Through examples of several alternative blends of the geometric configuration of a particular Hasse diagram with the SCALE image schema, we demonstrate the implications of these competing pressures on diagrammatic reasoning. Consequently, this work disambiguates and operationalises the intricacies of conceptual blending, advancing its applicability in computational systems.

Despite the recent success of neural network models in mimicking animal performance on various tasks, critics worry that these models fail to illuminate brain function. We take it that a central approach to explanation in systems neuroscience is that of mechanistic modeling, where understanding the system requires us to characterize its parts, organization, and activities, and how those give rise to behaviors of interest. However, it remains controversial what it takes for a model to be mechanistic, and whether computational models such as neural networks qualify as explanatory on this approach.

We argue that certain kinds of neural network models are actually good examples of mechanistic models, when an appropriate notion of mechanistic mapping is deployed. Building on existing work on model-to-mechanism mapping (3M), we describe criteria delineating such a notion, which we call 3M++. These criteria require us, first, to identify an abstract level of description that is still detailed enough to be “runnable”, and then, to construct model-to-brain mappings using the same principles as those employed for brain-to-brain mapping across individuals.

Perhaps surprisingly, the abstractions required are just those already in use in experimental neuroscience and deployed in the construction of more familiar computational models — just as the principles of inter-brain mappings are very much in the spirit of those already employed in the collection and analysis of data across animals.

In a companion paper, we address the relationship between optimization and intelligibility, in the context of functional evolutionary explanations. Taken together, mechanistic interpretations of computational models and the dependencies between form and function illuminated by optimization processes can help us to understand why brain systems are built they way they are.