FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W93035400> ?p ?o ?g. }

• W93035400 abstract "An Embodied Dynamical Approach to Relational Categorization Paul L. Williams 1 (plw@indiana.edu) Randall D. Beer 1,2,3 (rdbeer@indiana.edu) Michael Gasser 1,2 (gasser@cs.indiana.edu) 1 Cognitive Science Program, 2 Dept. of Computer Science, 3 Dept. of Informatics Indiana University, Bloomington, IN 47406 USA Abstract us to take seriously the view that cognition is situated, em- bodied, and dynamical. On this view, cognition is a contin- uous, ongoing interaction between a brain, a body, and an environment. After successfully evolving agents, we apply the tools of dynamical systems theory to analyze the result- ing brain/body/environment systems. The ability of dynamical neural circuits to perform rela- tional tasks has been demonstrated several times in previous work. For example, in one study a simple recurrent network was trained to recognize string sequences of the form a n b n , and thus to identify a same count relationship between se- quences of inputs (Rodriguez, Wiles, & Elman, 1999). How- ever, in this study the relational task was disembodied and computational in nature, whereas the work here is concerned with relational behavior in situated embodied agents. In another study, a neural model was proposed that captured findings from a relational task performed by nonhuman pri- mates (Miller, Brody, Romo, & Wang, 2003). In this case, though, the relational mechanism was hand-designed, while in the work presented here we employ evolutionary tech- niques, thereby attempting to minimize a priori assumptions about how the relational mechanisms should work. This paper has two primary aims. The first is to contribute to a growing body of research on minimally cognitive behav- ior, which studies simple behaviors of cognitive interest in or- der to develop and refine our conceptual and analytical tools for understanding cognition (Beer, 1996; Slocum, Downey, & Beer, 2000). The second is to show how the approach used here may inform the study and modeling of relational mecha- nisms more generally. To this end, we discuss how relational categorization is addressed by other models and identify some of the specific advantages of our approach. The rest of this paper is organized as follows. In the next section, we discuss the basic challenge posed by rela- tional categorization tasks and how this challenge has been addressed in previous models. The following section then describes the agent, neural network model, and evolutionary algorithm used here. Next, we briefly discuss the results of the evolutionary simulations. After, we present an analysis of the relational mechanism used by the best evolved agent. Fi- nally, we conclude with some general remarks, and an outline of ongoing and future work. This paper presents a novel approach to the study of relational categorization based on the evolution of simulated agents in a relational task. In contrast to most previous models of rela- tional categorization, which begin by assuming abstract rep- resentations and role-filler binding mechanisms, we begin by studying relational behavior in embodied dynamical agents, which results in a wider range of possibilities for relational mechanisms. The mathematical tools of dynamical systems theory are used to analyze the relational mechanism of the best evolved agent, and we then identify some of the insights of- fered by this analysis. Keywords: continuous-time recurrent neural networks; dy- namical systems; genetic algorithms; relational categories Introduction Recent research in cognitive science has seen flourishing in- terest in relations and relational categories (e.g., Gentner & Kurtz, 2005; Markman & Stilwell, 2001). Relational cate- gories are categories determined by common relational struc- ture among category members, rather than intrinsic similar- ities between category members. For example, same and smaller are instances of relational categories. There is a vast body of research on relational categorization in a wide range of species, including humans (Kurtz & Boukrina, 2004), pi- geons (Wills, 1999), rats (Saldanha & Bitterman, 1951), and insects (Giurfa, Zhang, Jenett, Menzel, & Srinivasan, 2001). Moreover, relational categories are of particular importance for cognitive science because they are fundamental to many topics, such as analogy and language. In recent years, a number of authors have proposed connec- tionist (Gasser & Colunga, 1998; Tomlinson & Love, 2006) and hybrid symbolic-connectionist (Hummel & Holyoak, 1997) models of relational categorization. These models put forth various mechanisms for the implementation of re- lational categories in cognitive systems. However, because these models are largely hand-crafted, they are limited to the space of possible mechanisms considered by their designers. In contrast, the work presented here takes an entirely dif- ferent approach to the study of relational mechanisms. We use an evolutionary algorithm to evolve dynamical neural controllers for simulated agents in a relational categorization task. There are numerous benefits of using this kind of ap- proach to study cognition (Beer, 1996; Harvey, Di Paolo, Wood, Quinn, & Tuci, 2005), two of which are central to our purposes. First, by evolving agents we are able to make minimal prior assumptions about how various behav- iors should be implemented. Second, this approach allows Relational Categorization Relational categorization tasks can be described in terms of two sets of features of the related objects (sometimes, but not" @default.
• W93035400 created "2016-06-24" @default.
• W93035400 creator A5014235111 @default.
• W93035400 creator A5063726569 @default.
• W93035400 creator A5083713233 @default.
• W93035400 date "2008-01-01" @default.
• W93035400 modified "2023-09-24" @default.
• W93035400 title "An embodied dynamical approach to relational categorization" @default.
• W93035400 cites W119251950 @default.
• W93035400 cites W1564759995 @default.
• W93035400 cites W1706366769 @default.
• W93035400 cites W1972085588 @default.
• W93035400 cites W1995921430 @default.
• W93035400 cites W1996950710 @default.
• W93035400 cites W2013494846 @default.
• W93035400 cites W2022029210 @default.
• W93035400 cites W2109840317 @default.
• W93035400 cites W2122186032 @default.
• W93035400 cites W2157306293 @default.
• W93035400 cites W2163324304 @default.
• W93035400 cites W2334384411 @default.
• W93035400 cites W2550491819 @default.
• W93035400 cites W2581390987 @default.
• W93035400 cites W287077210 @default.
• W93035400 hasPublicationYear "2008" @default.
• W93035400 type Work @default.
• W93035400 sameAs 93035400 @default.
• W93035400 citedByCount "6" @default.
• W93035400 countsByYear W930354002013 @default.
• W93035400 countsByYear W930354002018 @default.
• W93035400 countsByYear W930354002021 @default.
• W93035400 crossrefType "journal-article" @default.
• W93035400 hasAuthorship W93035400A5014235111 @default.
• W93035400 hasAuthorship W93035400A5063726569 @default.
• W93035400 hasAuthorship W93035400A5083713233 @default.
• W93035400 hasConcept C100609095 @default.
• W93035400 hasConcept C103683099 @default.
• W93035400 hasConcept C127413603 @default.
• W93035400 hasConcept C132829578 @default.
• W93035400 hasConcept C154945302 @default.
• W93035400 hasConcept C15744967 @default.
• W93035400 hasConcept C169760540 @default.
• W93035400 hasConcept C169900460 @default.
• W93035400 hasConcept C188147891 @default.
• W93035400 hasConcept C201995342 @default.
• W93035400 hasConcept C2780451532 @default.
• W93035400 hasConcept C41008148 @default.
• W93035400 hasConcept C80944243 @default.
• W93035400 hasConcept C94124525 @default.
• W93035400 hasConceptScore W93035400C100609095 @default.
• W93035400 hasConceptScore W93035400C103683099 @default.
• W93035400 hasConceptScore W93035400C127413603 @default.
• W93035400 hasConceptScore W93035400C132829578 @default.
• W93035400 hasConceptScore W93035400C154945302 @default.
• W93035400 hasConceptScore W93035400C15744967 @default.
• W93035400 hasConceptScore W93035400C169760540 @default.
• W93035400 hasConceptScore W93035400C169900460 @default.
• W93035400 hasConceptScore W93035400C188147891 @default.
• W93035400 hasConceptScore W93035400C201995342 @default.
• W93035400 hasConceptScore W93035400C2780451532 @default.
• W93035400 hasConceptScore W93035400C41008148 @default.
• W93035400 hasConceptScore W93035400C80944243 @default.
• W93035400 hasConceptScore W93035400C94124525 @default.
• W93035400 hasIssue "30" @default.
• W93035400 hasLocation W930354001 @default.
• W93035400 hasOpenAccess W93035400 @default.
• W93035400 hasPrimaryLocation W930354001 @default.
• W93035400 hasRelatedWork W119251950 @default.
• W93035400 hasRelatedWork W1509953508 @default.
• W93035400 hasRelatedWork W1650154193 @default.
• W93035400 hasRelatedWork W1758907577 @default.
• W93035400 hasRelatedWork W1876456565 @default.
• W93035400 hasRelatedWork W1983993791 @default.
• W93035400 hasRelatedWork W2012391952 @default.
• W93035400 hasRelatedWork W2022029210 @default.
• W93035400 hasRelatedWork W2041533382 @default.
• W93035400 hasRelatedWork W2058580716 @default.
• W93035400 hasRelatedWork W2085529605 @default.
• W93035400 hasRelatedWork W2097571405 @default.
• W93035400 hasRelatedWork W2103084048 @default.
• W93035400 hasRelatedWork W2142040995 @default.
• W93035400 hasRelatedWork W2163324304 @default.
• W93035400 hasRelatedWork W2169382901 @default.
• W93035400 hasRelatedWork W2397638991 @default.
• W93035400 hasRelatedWork W2402841949 @default.
• W93035400 hasRelatedWork W2550491819 @default.
• W93035400 hasRelatedWork W2772557520 @default.
• W93035400 hasVolume "30" @default.
• W93035400 isParatext "false" @default.
• W93035400 isRetracted "false" @default.
• W93035400 magId "93035400" @default.
• W93035400 workType "article" @default.