How we learn things we don’t know already: A theory of learning structured representations from experience (original) (raw)
Related papers
Learning Abstract Relations Through Analogy to Concrete Exemplars
We present a model that bridges work in analogy and category learning. The model, Building Relations through Instance Driven Gradient Error Shifting (BRIDGES), extends ALCOVE, an exemplar-based connectionist model of category learning . Unlike ALCOVE which is limited to featural or spatial representations, BRIDGES can appreciate analogical relationships between stimuli and stored predicate representations of exemplars. Like ALCOVE, BRIDGES learns to shift attention over the course of learning to reduce error and, in the process, alters its notion of similarity. A shift toward relational sources of similarity allows BRIDGES to display what appears to be an understanding of abstract domains, when in fact performance is driven by similarity-based structural alignment (i.e., analogy) to stored exemplars. Supportive simulations of animal, infant, and adult learning are provided.
A theory of relation learning and cross-domain generalization
Psychological review, 2022
People readily generalize knowledge to novel domains and stimuli. We present a theory, instantiated in a computational model, based on the idea that cross-domain generalization in humans is a case of analogical inference over structured (i.e., symbolic) relational representations. The model is an extension of the Learning and Inference with Schemas and Analogy (LISA; Hummel & Holyoak, 1997, 2003) and Discovery of Relations by Analogy (DORA; Doumas et al., 2008) models of relational inference and learning. The resulting model learns both the content and format (i.e., structure) of relational representations from nonrelational inputs without supervision, when augmented with the capacity for reinforcement learning it leverages these representations to learn about individual domains, and then generalizes to new domains on the first exposure (i.e., zero-shot learning) via analogical inference. We demonstrate the capacity of the model to learn structured relational representations from a ...
Developing structured representations
Behavioral and Brain Sciences, 2008
The development of analogical reasoning has traditionally been understood in terms of theories of adult competence. This approach emphasizes structured representations and structure mapping. In contrast, we argue that by taking a developmental perspective, analogical reasoning can be viewed as the product of a substantially different cognitive ability -relational priming. To illustrate this, we present a computational (here connectionist) account where analogy arises gradually as a by-product of pattern completion in a recurrent network. Initial exposure to a situation primes a relation that can then be applied to a novel situation to make an analogy. Relations are represented as transformations between states. The network exhibits behaviors consistent with a broad range of key phenomena from the developmental literature, lending support to the appropriateness of this approach (using low-level cognitive mechanisms) for investigating a domain that has normally been the preserve of high-level models. Furthermore, we present an additional simulation that integrates the relational priming mechanism with deliberative controlled use of inhibition to demonstrate how the framework can be extended to complex analogical reasoning, such as the data from explicit mapping studies in the literature on adults. This account highlights how taking a developmental perspective constrains the theory construction and cognitive modeling processes in a way that differs substantially from that based purely on adult studies, and illustrates how a putative complex cognitive skill can emerge out of a simple mechanism.
2005
Relational reasoning is central in human cognition. Numerous computational models address the component processes of relational reasoning, however these models require the modeler to hand-code the vocabulary of relations on which the model operates. The acquisition of relational concepts remains poorly understood. We present a theory of relation discovery instantiated in a symbolic-connectionist model, which learns structured representations of attributes and relations from unstructured distributed representations of objects by a process of comparison, and subsequently refines these representations through a process of mapping-based schema induction.
A Computational Theory for the Learning of Equivalence Relations
Frontiers in Human Neuroscience, 2011
Equivalence relations (ERs) are logical entities that emerge concurrently with the development of language capabilities. In this work we propose a computational model that learns to build ERs by learning simple conditional rules.The model includes visual areas, dopaminergic, and noradrenergic structures as well as prefrontal and motor areas, each of them modeled as a group of continuous valued units that simulate clusters of real neurons. In the model, lateral interaction between neurons of visual structures and top-down modulation of prefrontal/premotor structures over the activity of neurons in visual structures are necessary conditions for learning the paradigm. In terms of the number of neurons and their interaction, we show that a minimal structural complexity is required for learning ERs among conditioned stimuli. Paradoxically, the emergence of the ER drives a reduction in the number of neurons needed to maintain those previously specific stimulus-response learned rules, allowing an efficient use of neuronal resources.
Producing and Recognizing Analogical Relations
Journal of the Experimental Analysis of Behavior, 2009
Analogical reasoning is an important component of intelligent behavior, and a key test of any approach to human language and cognition. Only a limited amount of empirical work has been conducted from a behavior analytic point of view, most of that within Relational Frame Theory (RFT), which views analogy as a matter of deriving relations among relations. The present series of four studies expands previous work by exploring the applicability of this model of analogy to topography-based rather than merely selectionbased responses and by extending the work into additional relations, including nonsymmetrical ones. In each of the four studies participants pretrained in contextual control over nonarbitrary stimulus relations of sameness and opposition, or of sameness, smaller than, and larger than, learned arbitrary stimulus relations in the presence of these relational cues and derived analogies involving directly trained relations and derived relations of mutual and combinatorial entailment, measured using a variety of productive and selection-based measures. In Experiment 1 participants successfully recognized analogies among stimulus networks containing same and opposite relations; in Experiment 2 analogy was successfully used to extend derived relations to pairs of novel stimuli; in Experiment 3 the procedure used in Experiment 1 was extended to nonsymmetrical comparative relations; in Experiment 4 the procedure used in Experiment 2 was extended to nonsymmetrical comparative relations. Although not every participant showed the effects predicted, overall the procedures occasioned relational responses consistent with an RFT account that have not yet been demonstrated in a behavior-analytic laboratory setting, including productive responding on the basis of analogies.
A Symbolic-Connectionist Model of Relation Discovery - eScholarship
Proceedings of the Annual Meeting of the Cognitive Science Society, 2005
Relational reasoning is central in human cognition. Numerous computational models address the component processes of relational reasoning, however these models require the modeler to hand-code the vocabulary of relations on which the model operates. The acquisition of relational concepts remains poorly understood. We present a theory of relation discovery instantiated in a symbolic-connectionist model, which learns structured representations of attributes and relations from unstructured distributed representations of objects by a process of comparison, and subsequently refines these representations through a process of mapping-based schema induction.
Comparison and Mapping Facilitate Relation Discovery and Predication
PLoS ONE, 2013
Relational concepts play a central role in human perception and cognition, but little is known about how they are acquired. For example, how do we come to understand that physical force is a higher-order multiplicative relation between mass and acceleration, or that two circles are the same-shape in the same way that two squares are? A recent model of relational learning, DORA (Discovery of Relations by Analogy; Doumas, Hummel & Sandhofer, 2008), predicts that comparison and analogical mapping play a central role in the discovery and predication of novel higher-order relations. We report two experiments testing and confirming this prediction.
Trained and Derived Relations With Pictures Versus Abstract Stimuli as Nodes
The Psychological record
Earlier studies have shown divergent results concerning the use of familiar picture stimuli in demonstration of equivalence. In the current experiment, we trained 16 children to form three 3-member classes in a many-to-one training structure. Half of the participants were exposed first to a condition with all abstract stimuli and then to a condition with new abstract stimuli as samples and 3 picture stimuli as comparisons (and nodes). The other participants were given the 2 conditions in the reverse order. The results, regardless of order, showed that the condition with picture stimuli as nodes was more effective in producing responding in accord with equivalence than stimulus sets with abstract stimuli only. In addition, more participants responded in accord with equivalence when they were trained with picture stimuli first. Reaction time to the comparison stimuli showed a greater increase with abstract stimuli than with pictures as nodes.