Learning & development

Nobandegani, A. S., Shultz, T. R., & Rish, I. (2023). Cognitive Models as Simulators: Using Cognitive Models to Tap into Implicit Human Feedback. Interactive Learning with Implicit Human Feedback Workshop at International Conference on Machine Learning, (1-9). Honolulu, Hawaii, USA. In this work, we substantiate the idea of cognitive models as simulators, which is to have AI systems interact with, and collect feedback from, cognitive models instead of humans, thereby making the training process safer, cheaper, and faster. We leverage this idea in the context of learning a fair behavior toward a counterpart exhibiting various emotional states — as implicit human feedback. As a case study, we adopt the Ultimatum game (UG), a canonical task in behavioral and brain sciences for studying fairness. We show that our reinforcement learning (RL) agents learn to exhibit differential, rationally-justified behaviors under various emotional states of their UG counterpart. We discuss the implications of our work for AI and cognitive science research, and its potential for interactive learning with implicit human feedback.
Nobandegani, A. S., & Shultz, T. R. (2023). Neural Network Modeling of Pure Reasoning in Preverbal Infants. In M. Goldwater, F. K. Anggoro, B. K. Hayes, & D. C. Ong (Eds.), Proceedings of the 45th Annual Conference of the Cognitive Science Society (1530-1536). Recent empirical research has provided evidence for pure reasoning in infancy, a capacity permitting flexible integration of multiple sources of information to form rational expectations about novel events (Teglas et al., 2011). However, the neural underpinnings of this capacity have remained elusive. In this work, we present the first ecologically rational, neural-level account of these findings on pure reasoning in human infants. Our work bridges two dominant approaches in computational developmental psychology, i.e. neural-network and Bayesian, substantiating the view that intuitive physics in infancy might, at least partly, involve heuristics: a set of simple, fast, resource-efficient, approximation algorithms that yield sufficiently good results. pdf
Fan, Z., Wang, Z., Shultz, T. R. (2023). A Perceptual Front-End for Probability Learning: Object Detection with YOLO. In M. Goldwater, F. K. Anggoro, B. K. Hayes, & D. C. Ong (Eds.), Proceedings of the 45th Annual Conference of the Cognitive Science Society (3130-3136). Neural Probabilistic Learner and Sampler (NPLS) is an algorithm that has simulated children’s non-symbolic probability learning from visual stimuli such as collections of different colors of marbles. Although NPLS closely simulates the cognitive process of probability learning, the training of such learning algorithms often uses binary encoding of inputs that represent the perceived visual stimuli, avoiding simulation of the visual perception of the stimuli. Here, the computer vision technique You Only Look Once (YOLO) (Jocher et al., 2021; Redmon et al., 2016), is integrated into the workflow of an NPLS simulation of probability learning experiments with children. YOLO is a convolutional neural network (CNN) designed to detect objects. The model’s performance on marble datasets is tested through an analysis of precision and recall. Results indicate that the YOLO model, when trained sufficiently, outputs predictions on marble image datasets with high accuracy and precision. We also analyze YOLO’s suitability as a biologically plausible model of visual processing, interfering with YOLO’s training process by shortening the amount of training to examine the effects of perceptual errors on simulated probabilistic reasoning. pdf
Shultz, T. R., Nobandegani, A. S., & Wang, Z. (2023). A Neural Model of Number Comparison with Robust Generalization. In M. Goldwater, F. K. Anggoro, B. K. Hayes, & D. C. Ong (Eds.), Proceedings of the 45th Annual Conference of the Cognitive Science Society (2134-2140). We propose and implement a relatively simple computational neural-network model of number comparison. Training on paired comparisons of the integers 1-9 enables the model to efficiently and accurately simulate some fundamental empirical phenomena (distance and ratio effects on accuracy and response time). It also generalizes robustly to more advanced tasks involving multidigit integers, negativenumbers, and decimal numbers. The work demonstrates that small neural networks can sometimes efficiently learn a powerful system that exhibits extremely robust generalization to untrained items. Some important alternate models of number comparison are considered to establish a broader context. Several predictions and suggestions are made for future empirical and computational research in this area. pdf
Shultz, T. R., & Nobandegani, A. (2023). Computational models of developmental psychology. In R. Sun (Ed.), The Cambridge Handbook on Computational Cognitive Sciences (pp. 769-794). Cambridge: Cambridge University Press. This is a comparative review of the major types of models being applied to cognitive developmental issues. doi:10.1017/9781108755610.028
Nobandegani, A. S., & Shultz, T. R. (2022). Computational approaches to cognitive development: Bayesian and artificial-neural-network models. In O. Houdé & G. Borst (Eds.), The Cambridge Handbook of Cognitive Development (pp. 318-338). Cambridge: Cambridge University Press. This is a comparitive review of the two leading approaches in computational modeling of cognitive development. pdf
Shultz, T. R., & Nobandegani, A. S. (2022). A Computational Model of Infant Learning and Reasoning With Probabilities. Psychological Review, 129(6), 1281–1295. We present a novel computational system called Neural Probability Learner and Sampler (NPLS) that learns and reasons with probabilities, providing a computationally sufficient mechanism to explain infant probabilistic learning and inference. In 24 computer simulations, NPLS shows how probability distributions can emerge naturally from neural-network learning of event sequences, providing a novel explanation of infant probabilistic learning and reasoning. Three mathematical proofs show how and why NPLS simulates the infant results so accurately. The results are situated in relation to seven other active research lines. This work provides an effective way to integrate Bayesian and neural-network approaches to cognition. pdf
Shultz, T. & Nobandegani, A. (2022). Modeling the Learning and Use of Probability Distributions in Chimpanzees and Humans. In J. Culbertson, A. Perfors, H. Rabagliati, & V. Ramenzoni (Eds.), Proceedings of the 44th Annual Conference of the Cognitive Science Society, (1947-1953). Toronto, ON: Cognitive Science Society. - We present a neural-network computational model of a recent experiment revealing that chimpanzees show some ability to reason probabilistically. Specifically, we show that the neural probability learner and sampler (NPLS) system can account for both success by chimpanzees and better performance by human controls. NPLS effectively combines learning probability distributions with sampling from those learned distributions to guide action choices. Because NPLS also simulates learning and use of probability distributions by human infants, this brings us closer to a unifying model of probabilistic reasoning, across various age groups and species. pdf
Nobandegani, A. S., & Shultz, T. R. (2022). Computational approaches to cognitive development: Bayesian and artificial-neural-network models. In O. Houdé & G. Borst (Eds.), The Cambridge Handbook of Cognitive Development (pp. 318-338). Cambridge: Cambridge University Press. This chapter reviews research using the two of the most influential approaches to such modeling: Bayesian and artificial neural networks. The techniques are explained for a general audience and concrete examples are described, providing both an in-depth and broad coverage of these literatures. pdf
Montrey, M. & Shultz, T. (2022). Ingroup-Biased Copying Promotes Cultural Diversity and Complexity. In J. Culbertson, A. Perfors, H. Rabagliati, & V. Ramenzoni (Eds.), Proceedings of the 44th Annual Conference of the Cognitive Science Society, (1306-1313). Toronto, ON: Cognitive Science Society. - Studies have found that when innovation involves recombining cultural traits, partially-connected populations produce higher levels of cultural complexity than fully-connected populations by avoiding cultural homogenization. However, population connectedness is only one of many factors that could promote cultural diversity and thus cultural complexity. Here, we examine whether people's preference for copying members of their own social group could also fill this role. Our simulations reveal that even in fully-connected populations, ingroup-biased transmission results in greater cultural complexity than unbiased transmission. Moreover, in partially-connected populations, this bias interacts with population structure to produce even higher levels of cultural complexity than population structure alone. Finally, by incorporating population turnover into our model, we shed light on the trade-off between promoting cultural diversity versus limiting cultural loss. pdf
Montrey, M., & Shultz, T. R. (2022). Copy the in-group: group membership trumps perceived reliability, warmth, and competence in a social learning task. Psychological Science, 33(1), 165-174. Although several studies have shown that humans prefer to copy in-group members, these have failed to resolve whether group membership genuinely affects who is copied or if it merely correlates with other known factors, such as similarity and familiarity. Using the minimal group paradigm, we disentangle these effects in an online social learning game. We find a robust in-group copying bias that (1) is bolstered by a preference for observing in-group members; (2) overrides perceived reliability, warmth, and competence; (3) grows stronger when social information is scarce; and (4) even causes cultural divergence between intermixed groups. These results suggest that humans genuinely employ a copy-the-ingroup social learning strategy, which could help explain how inefficient behaviors spread through social learning and how humans maintain the cultural diversity needed for cumulative cultural evolution. pdf
Shultz, T. R., & Nobandegani, A. S. (2020). Probability without counting and dividing: a fresh computational perspective. In S. Denison, M. Mack, Y. Xu, & B. Armstrong (Eds.), Proceedings of the 42nd Annual Conference of the Cognitive Science Society (3405-3411). Toronto, ON: Cognitive Science Society. - Recent experiments show that preverbal infants can reason probabilistically. This raises a deep puzzle because infants lack the counting and dividing abilities presumably required to compute probabilities. In the standard way of computing probabilities, they would have to count or accurately estimate large frequencies and divide those values by their total. Here, we present a novel neural-network model that learns and uses probability distributions without explicit counting or dividing. Probability distributions emerge naturally from neural-network learning of event sequences, providing a computationally sufficient explanation of how infants could succeed at probabilistic reasoning. Several alternative explanations are discussed and ruled out. Our work bears on several other active literatures, and it suggests an effective way to integrate Bayesian and neural-network approaches to cognition. pdf
Shultz, T. R., Montrey, M., & Aplin, L. M. (2017). Modelling the spread of innovation in wild birds. Journal of the Royal Society Interface, 14: 20170215. - We apply three plausible algorithms in agent-based computer simulations to recent experiments on social learning in wild birds. Although some of the phenomena are simulated by all three learning algorithms, several manifestations of social conformity bias are simulated by only the approximate majority (AM) algorithm, which has roots in chemistry, molecular biology and theoretical computer science. The simulations generate testable predictions and provide several explanatory insights into the diffusion of innovation through a population. The AM algorithm’s success raises the possibility of its usefulness in studying group dynamics more generally, in several different scientific domains. Our differential-equation model matches simulation results and provides mathematical insights into the dynamics of these algorithms. pdf
Shultz, T. R. (2017). Constructive artificial neural-network models for cognitive development. In N. Budwig, E. Turiel, & P. D. Zelazo (Eds.), New Perspectives on Human Development (pp. 13-26). Cambridge: Cambridge University Press. - This chapter provides an overview of research using constructive neural networks to simulate phenomena in cognitive development. These algorithms are described, and their application to developmental issues and phenomena are reviewed in both breadth and some depth. When contrasted against fully designed static neural networks or symbolic rule-based systems, constructive neural networks provided superior coverage of the psychological data. Theoretical implications are discussed and constructive networks are situated within the current computational-developmental literature. pdf
Shultz, T. R. (2015). Connectionist models of development. In J. D. Wright (Ed.), International Encyclopedia of the Social & Behavioral Sciences (2 ed., pp. 1–8). Oxford: Elsevier, ISBN 9780080970875, - Survey of connectionist models of psychological development. Description of important modeling techniques. Summary and evaluation of how this modeling has addressed fundamental issues in developmental psychology: the distinction between development and learning, progression through stages, and the debate between symbols and connections. Applications to the study of concept acquisition, education, and disordered development. Emerging trends. pdf
Shultz, T. R., & Doty, E. (2014). Knowing when to quit on unlearnable problems: another step towards autonomous learning. In J. Mayor & P. Gomez (Eds.), Computational Models of Cognitive Processes (pp. 211-221). London: World Scientific. - Learning in SDCC is abandoned when network error fails to change by more than a specified threshold for a specified number of consecutive learning cycles. Here we explore the space defined by threshold (sensitivity to error change) and patience parameters on problems of different degrees of learnability. pdf
Coldren, J. T., & Shultz, T. R (May, 2014). A neural network simulation of cognitive control. Poster, Association for Psychological Science, San Francisco, CA. - We use a hybrid SDCC and constraint-satisfaction model to simulate several empirical phenomena on the Dimensional Change Card Sort task.
Coldren, J. T., Colombo, J., & Shultz, T. R. (May, 2014). Empirical and neural network evidence of infant dimensional learning. Poster, Association for Psychological Science, San Francisco, CA. - We show that 4-month-olds and SDCC networks find an intra-dimensional shift easier than an extra-dimensional shift.
Dandurand, F., & Shultz, T. R. (2014). A comprehensive model of development on the balance-scale task. Cognitive Systems Research, 31-32, 1-25. - A tripartite neural-network model featuring both KBCC and SDCC that captures all four replicable balance-scale stages (including a genuine torque rule), the torque-difference effect, response times, and overlapping waves of development. A selection module decides whether to deliver the quick response of the intuitive module (based on confidence) or whether to invoke the more deliberative approach of computing and comparing torques. Appendices document the tendency of Latent Class Analysis to find small, random classes that fail to replicate. pdf
Kharratzadeh, M. & Shultz, T. R. (2013). Neural-network modelling of Bayesian learning and inference. In M. Knauff, M. Pauen, N. Sebanz, & I. Wachsmuth (Eds.), Proceedings of the 35th Annual Conference of the Cognitive Science Society (pp. 2686-2691). Austin, TX: Cognitive Science Society. - A modular neural-network implementation of Bayesian learning and inference. Provides a novel explanation of base-rate neglect as connection-weight decay. pdf
Shultz, T. R. (2013). Computational models in developmental psychology. In P. D. Zelazo (Ed.), Oxford handbook of developmental psychology, Vol. 1: Body and mind (pp. 477-504). New York: Oxford University Press. - A general review of all the major approaches to modeling psychological development.
Berthiaume, V. G., Shultz, T. R., & Onishi, K. H. (2013). A constructivist connectionist model of transitions on false-belief tasks. Cognition, 126 (3), 441-458. - SDCC networks simulate the major transitions in children's false-belief development. pdf
Shultz, T. R. (2012). A constructive neural-network approach to modeling psychological development. Cognitive Development, 27, 383-400. - A rare chance to review our simulations with cascade-correlation (CC) and sibling-descendant CC (SDCC). Received an Editor's Choice award. pdf
Shultz, T. R., Doty, E., & Dandurand, F. (2012). Knowing when to abandon unproductive learning. In N. Miyake, D. Peebles, & R. P. Cooper (Eds.), Proceedings of the 34th Annual Conference of the Cognitive Science Society (pp. 2327-2332). Austin, TX: Cognitive Science Society. - An SDCC model that monitors its learning progress and quits when there is no change in network error for some number of learning cycles, thus simulating results in recent infant experiments, with implications for the modeling of autonomous learning. pdf
Shultz, T. R. (2011). Computational modeling of infant concept learning: The developmental shift from features to correlations. In L. M. Oakes, C. H. Cashon, M. Casasola & D. H. Rakison (Eds.), Infant perception and cognition: Recent advances, emerging theories, and future directions (pp. 125-152). New York: Oxford University Press. - A comparative analysis of various computational models of Les Cohen’s infant experiments on this pervasive shift, including a new SDCC model.
Dandurand, F., & Shultz, T. R. (2011). A fresh look at vocabulary spurts. In C. Hoelscher, T. F. Shipley, & L. Carlson (Eds.), Proceedings of the 33rd Annual Conference of the Cognitive Science Society (pp. 1134-1139). Boston, MA: Cognitive Science Society. - Application of our Functional Data Analysis technique (called Automatic Maxima Detection) shows that children exhibit multiple vocabulary spurts of varying intensity and location, not just one spurt as commonly believed. pdf
Shultz, T. R., & Kuipers, B. (2011). Understanding psychological development in biological and artificial agents. IEEE Transactions on Autonomous Mental Development, 3 (1), 4-5. - One of four surveys of best papers at a 2010 conference of developmental researchers from psychology and robotics. pdf
Shultz, T. R., & Kuipers, B. (2011). Cognitive development in humans and developmental robots. Cognitive Development, 26, 82-85.
Shultz, T. R., & Kuipers, B. (2010). Development and learning. The Reasoner, 4 (12), 178-179.
Shultz, T. R., & Kuipers, B. (2010). Report of the Ninth IEEE International Conference on Development and Learning (IEEE ICDL 2010). In Pierre-Yves Oudeyer (Ed.), IEEE AMD Newsletter: The Newsletter of the Autonomous Mental Development Technical Committee, 7 (2), 11-12.
Dandurand, F., & Shultz, T. R. (2010). Automatic detection and quantification of growth spurts. Behavior Research Methods, 42 (3), 809-823. - A new Functional Data Analysis technique for systematically finding spurts in individual growth curves.
Shultz, T. R. (2010). Connectionism and learning. In P. Peterson, E. Baker, & B. McGaw, (Eds.), International Encyclopedia of Education, 5, 476-484. Oxford: Elsevier. - A survey of work that applies connectionist models to educational issues in reading and mathematics.
Evans, V. C., Berthiaume, V. G., & Shultz, T. R. (2010). Toddlers’ transitions on non-verbal false-belief tasks involving a novel location: a constructivist connectionist model. Proceedings of the Ninth IEEE International Conference on Development and Learning (pp. 225-230) Ann Arbor, MI: IEEE. - An SDCC model of an experiment on 25-month-olds correctly anticipating that an actress would search according to her false belief. pdf
Shultz, T. R., Berthiaume, V. G., & Dandurand, F. (2010). Bootstrapping syntax from morpho-phonology. Proceedings of the Ninth IEEE International Conference on Development and Learning (pp. 52-57). Ann Arbor, MI: IEEE. - A hybrid competitive-learning/SDCC model showing how it is possible for syntax to be learned from positive evidence alone, simulating recent adult experiments. Although this may not work with isolating languages (having a low morpheme-to-word ratio) or with analytic languages like Mandarin (where inflections do not indicate syntax), it could work with the many well-attested synthetic languages (having high or moderate morpheme-to-word ratios). In any case, the simulation demonstrates the power of using unsupervised CL to generate targets for error-driven SDCC learning, which could be extended beyond language learning. pdf
Berthiaume, V. G., Shultz, T. R., & Dammann, O. (2010). White- and grey-matter damage differentially impair learning and generalization in a computational model of the raven matrices task. In S. Ohlsson & R. Catrambone (Eds.), Proceedings of the 32nd Annual Conference of the Cognitive Science Society (pp. 682-687). Austin, TX: Cognitive Science Society. - An SDCC model of the brain damage associated with preterm birth and its effect on performance on IQ test items. pdf
Baetu, I., & Shultz, T. R. (2010). Development of prototype abstraction and exemplar memorization. In S. Ohlsson & R. Catrambone (Eds.), Proceedings of the 32nd Annual Conference of the Cognitive Science Society (pp. 814-819). Austin, TX: Cognitive Science Society. - An SDCC model that integrates prototype and exemplar effects in concept learning and reconciles apparently conflicting findings on the development of these effects. pdf
Metz, A., & Shultz, T. R. (2010). Spatial factors in social and asocial learning. In S. Ohlsson & R. Catrambone (Eds.), Proceedings of the 32nd Annual Conference of the Cognitive Science Society (pp. 1685-1690). Austin, TX: Cognitive Science Society. - The proportion of asocial innovative learners increases boundary surface length between social and asocial learners which, in turn, increases social imitative learning.
Verbuk, A., & Shultz, T. (2010). Acquisition of relevance implicatures: A case against a rationality-based account of conversational implicatures. Journal of Pragmatics, 42 (8), 2297-2313. - In a first-language acquisition experiment with children, we compared the language-based and rationality-based accounts of how relevance implicatures are computed, and found support only for the former.
Schlimm, D., & Shultz, T. R. (2009). Learning the structure of abstract groups. In N. A. Taatgen & H. van Rijn (Eds.), Proceedings of the 31st Annual Conference of the Cognitive Science Society (pp. 2950-2955). Austin, TX: Cognitive Science Society. - Contradicting some views, knowledge-based CC (KBCC) networks learn to recognize mathematical groups consisting of up to 4 elements by abstracting the structure of subgroups, exhibiting the advantages of knowledge-based learning. pdf
Dandurand, F., & Shultz, T. R. (2009). Modeling acquisition of a torque rule on the balance-scale task. In N. A. Taatgen & H. van Rijn (Eds.), Proceedings of the 31st Annual Conference of the Cognitive Science Society (pp. 1541-1546). Austin, TX: Cognitive Science Society. - Contradicting recent claims, KBCC networks progress through all four stages seen in children, ending with a genuine torque rule that can solve problems only solvable by comparing torques. pdf
Shultz, T. R., & Sirois, S. (2008). Computational models of developmental psychology. In R. Sun (Ed.), The Cambridge handbook of computational psychology (pp. 451-476). New York: Cambridge University Press. - A comparative review of the major types of models being applied to developmental issues.
Verbuk, A., & Shultz, T. R. (2008). Acquisition of relevance implicatures: toward isolating the linguistic component of reasoning. Chicago Linguistic Society (CLS 44), 65-79. Chicago: University of Chicago. - Preliminary conference version of Verbuk & Shultz (2010). pdf
Shultz, T. R., Thivierge, J. P., & Laurin, K. (2008). Acquisition of concepts with characteristic and defining features. In B. C. Love, K. McRae, & V. M. Sloutsky (Eds.), Proceedings of the 30th Annual Conference of the Cognitive Science Society (pp. 531-536). Austin, TX: Cognitive Science Society. - SDCC networks cover several phenomena associated with learning concepts along the probabilistic to defining-features continuum, including the shift from the former to the latter. Neural networks are not supposed to be able to deal with crisp concepts with defining features. pdf
Berthiaume, V. G., Onishi, K. H., & Shultz, T. R. (2008). A computational developmental model of the implicit false belief task. In B. C. Love, K. McRae, & V. M. Sloutsky (Eds.), Proceedings of the 30th Annual Conference of the Cognitive Science Society (pp. 825-830). Austin, TX: Cognitive Science Society. - SDCC networks simulate the transition from omniscient to representational understanding of implicit false beliefs found in 15-month-olds. pdf
Jamrozik, A., & Shultz, T. R. (2007). Learning the structure of a mathematical group. In D. McNamara & G. Trafton (Eds.), Proceedings of the Twenty-ninth Annual Conference of the Cognitive Science Society (pp. 1115-1120). Mahwah, NJ: Lawrence Erlbaum. - SDCC networks learn the abstract structure of the Klein 4-group, simulating adult performance, and contradicting claims that this is outside the ability of neural networks. pdf
Shultz, T. R. (2007). Connectionist learning and education: applications and challenges. In D. McNamara & G. Trafton (Eds.), Proceedings of the Twenty-ninth Annual Conference of the Cognitive Science Society (pp. 1497-1502). Mahwah, NJ: Lawrence Erlbaum. - Preliminary conference version of Shultz (2010).
Shultz, T. R., & Takane, Y. (2007). Rule following and rule use in the balance-scale task. Cognition, 103, 460-472. - A rejoinder to a critique of our CC model of the development of balance-scale knowledge, emphasizing misuse of Latent Class Analysis of stages.
Shultz, T. R. (2007). The Bayesian revolution approaches psychological development. Developmental Science, 10, 357-364. - Critical, but supportive review of five articles applying Bayesian ideas to psychological development.
Shultz, T. R., Mysore, S. P., & Quartz, S. R. (2007). Why let networks grow? In D. Mareschal, S. Sirois, G. Westermann, & M. H. Johnson (Eds.), Neuroconstructivism: Perspectives and prospects (Vol. 2, pp. 65-98). Oxford: Oxford University Press. - Review of neuroscience and computational evidence for constructive neural networks.
Shultz, T. R. (2006). Constructive learning in the modeling of psychological development. In Y. Munakata & M. H. Johnson (Eds.), Processes of change in brain and cognitive development: Attention and performance XXI (pp. 61-86). Oxford: Oxford University Press. - Review of evidence that constructive networks are better than static ones at learning and covering developmental phenomena, including a new SDCC model of conservation acquisition.
Shultz, T. R., & Bale, A. C. (2006). Neural networks discover a near-identity relation to distinguish simple syntactic forms. Minds and Machines, 16, 107-139. - Constructive neural networks cover the results of artificial-grammar learning in infants, despite claims to the contrary, with a more extensive knowledge-representation analysis and new simulations.
Sirois, S., & Shultz, T. R. (2006). Preschoolers out of adults: Discriminative learning with a cognitive load. Quarterly Journal of Experimental Psychology, 59, 1357-1377. - Consistent with the predictions of earlier CC simulations, adults given a cognitive load perform like preschoolers in discrimination-shift learning.
Westermann, G., Sirois, S., Shultz, T. R., & Mareschal, D. (2006). Modeling developmental cognitive neuroscience. Trends in Cognitive Sciences, 10, 227-232. - Review of neural network modeling that involves either network construction or integration of modules. pdf
Thivierge, J. P., Shultz, T. R., & Balaban, E. (2005). A unified model of thalamocortical axon guidance. Proceedings of the Twentieth AAAI Annual Conference (pp.3-14). Menlo Park, California: AAAI. - A model of cortical map formation involving both activity-independent and activity-dependent processes.
Shultz, T. R. (2005). Generalization in a model of infant sensitivity to syntactic variation. Proceedings of the Twenty-seventh Annual Conference of the Cognitive Science Society (pp. 2009-2014). Mahwah, NJ: Erlbaum. - CC networks cover the essential features of infant learning of artificial grammars, generalizing by both extrapolation and interpolation. This was supposed to be the exclusive domain of symbolic rules. pdf
Shultz, T. R., & Gerken, L. A. (2005). A model of infant learning of word stress. Proceedings of the Twenty-seventh Annual Conference of the Cognitive Science Society (pp. 2015-2020). Mahwah, NJ: Erlbaum. - SDCC networks cover the ability of 9-month-olds to distinguish the word-stress patterns of two artificial languages, by making transitive inferences from known to unknown constraints. pdf
Thivierge, J.-P., Titone, D., & Shultz, T. R. (2005). Simulating frontotemporal pathways involved in lexical ambiguity resolution. Proceedings of the Twenty-seventh Annual Conference of the Cognitive Science Society (pp. 2178-2183). Mahwah, NJ: Erlbaum. - Rule-based CC (RBCC) model of the neural pathways involved in resolving lexical ambiguities.
Shultz, T. R., & Vogel, A. (2004). A connectionist model of the development of transitivity. Proceedings of the Twenty-sixth Annual Conference of the Cognitive Science Society (pp. 1243-1248). Mahwah, NJ: Erlbaum. - Hybrid CC/constraint-satisfaction model captures six important phenomena in the development of transitive inference. pdf
Shultz, T. R., & Cohen, L. B. (2004). Modeling age differences in infant category learning. Infancy, 5, 153-171. - CC networks simulate the shift from detecting features to detecting relations among features.
Shultz, T. R. (2003). Computational developmental psychology. Cambridge, MA: MIT Press. - “A manifesto for a more scientific approach to cognitive development in which the focus is firmly on the mechanisms of change. Packed with detailed examples, this book is essential reading for advanced students and researchers in cognitive development and will be of interest to cognitive scientists more generally.” Mark Johnson. book website
Sirois, S., & Shultz, T.R. (2003). A Connectionist perspective on Piagetian development. In P. T. Quinlan (Ed.), Connectionist models of development: Developmental processes in real and artificial neural networks (pp. 13-41). New York: Psychology Press. - Reinterpretation of Piaget’s theory in terms of neural networks.
Shultz, T. R. (2001). Connectionist models of development. In N. J. Smelser & P. B. Baltes (Eds.), International Encyclopedia of the Social and Behavioral Science (Vol. 4, pp. 2577-2580). Oxford: Pergamon. - A general review.
Shultz, T. R., & Bale, A. C. (2001). Neural network simulation of infant familiarization to artificial sentences: Rule-like behavior without explicit rules and variables. Infancy, 2, 501-536. - Contradicting claims to the contrary, unstructured CC networks simulate infant ability to distinguish simple, artificial grammars.
Shultz, T. R. (2001). Assessing generalization in connectionist and rule-based models under the learning constraint. Proceedings of the Twenty-third Annual Conference of the Cognitive Science Society (pp. 922-927). Mahwah, NJ: Erlbaum. - Under the reasonable constraint that models must be able to learn their knowledge representations, neural-network models consistently show superior generalization to symbolic-rule models. pdf
Buckingham, D., & Shultz, T. R. (2000). The developmental course of distance, time, and velocity concepts: A generative connectionist model. Journal of Cognition and Development, 1, 305-345. - CC networks capture the rule-like stages in children’s acquisition of concepts for moving objects: identity, additive, and multiplicative.
Sirois, S., Buckingham, D., & Shultz, T. R. (2000). Artificial grammar learning by infants: An auto-associator perspective. Developmental Science, 4, 442-456. - Contradicting claims to the contrary, unstructured auto-associator networks simulate infant ability to distinguish simple, artificial grammars.
Shultz, T. R., & Bale, A. C. (2000). Infant familiarization to artificial sentences: Rule-like behavior without explicit rules and variables. Proceedings of the Twenty-second Annual Conference of the Cognitive Science Society (pp. 459-463). Mahwah, NJ: Erlbaum. - Preliminary conference version of Shultz & Bale (2001).
Oshima-Takane, Y., Takane, Y., & Shultz, T. R. (1999). The learning of first and second pronouns in English: Network models and analysis. Journal of Child Language, 26, 545-575. - CC network learning speed and knowledge graphs confirm the importance of overheard speech, multiple speakers, and knowledge of the category PERSON in acquisition of personal pronouns.
Shultz, T. R. (1999). Rule learning by habituation can be simulated in neural networks. Proceedings of the Twenty-first Annual Conference of the Cognitive Science Society (pp. 665-670). Mahwah, NJ: Erlbaum. - CC networks cover acquisition of simple grammars by infants, showing that such coverage is not exclusive to symbolic rules.
Sirois, S., & Shultz, T. R. (1999). Learning, development, and nativism: Connectionist implications. Proceedings of the Twenty-first Annual Conference of the Cognitive Science Society (pp. 689-694). Mahwah, NJ: Erlbaum. - Analysis of why static networks implement learning, while constructive networks implement learning and development.
Mareschal, D., & Shultz, T. R. (1999). Development of children's seriation: A connectionist approach. Connection Science, 11, 149-186. - Modular CC networks cover several important aspects of seriation development.
Sirois, S., & Shultz, T. R. (1998). Neural network modeling of developmental effects in discrimination shifts. Journal of Experimental Child Psychology, 71, 235-274. - CC networks capture psychological regularities in the vast literature on discrimination learning shifts better than psychological theories do and suggest a novel interpretation of developmental change based on spontaneous rehearsal.
Sirois, S., & Shultz, T. R. (1998). Neural network models of discrimination shifts. Proceedings of the Twentieth Annual Conference of the Cognitive Science Society (pp. 980-985). Mahwah, NJ: Erlbaum. - Preliminary conference version of Sirois & Shultz (1998, JECP).
Shultz, T. R. (1998). A computational analysis of conservation. Developmental Science, 1, 103-126. - CC networks simulate several major psychological regularities in conservation acquisition: the problem size, length bias, and screening effects, and sudden jumps in performance, while providing novel explanations. Analysis of network knowledge representations (PCA of network contributions) supports Piaget’s view of a shift from perception to reasoning.
Shultz, T. R., & Mareschal, D. (1997). Rethinking innateness, learning, and constructivism: Connectionist perspectives on development. Cognitive Development, 12, 563-586. - Essay review of the rethinking-innateness books. Lots of interesting material on different ways to be innate, but not constructivist at the level of modeling.
Shultz, T. R. (1996). Models of cognitive development. In V. Rialle & D. Fisette, (Eds.), Penser l'esprit: Des sciences de la cognition à une philosophie cognitive (pp. 393-405). Grenoble: Presses Universitaires de Grenoble. - Early forecast of the merits of constructivist networks.
Mareschal, D., & Shultz, T. R. (1996). Generative connectionist networks and constructivist cognitive development. Cognitive Development, 11, 571-603. - Computational analysis of constructive networks. They, but not static networks, can escape from Fodor’s paradox about the impossibility of learning anything genuinely new.
Buckingham, D., & Shultz, T. R. (1996). Computational power and realistic cognitive development. Proceedings of the Eighteenth Annual Conference of the Cognitive Science Society (pp. 507-511). Mahwah, NJ: Erlbaum. - Static back-propagation networks cannot capture children’s stages on integration of the concepts of velocity, time, and distance. They are either too weak to reach the terminal stage or too strong to develop the intermediate stages.
Shultz, T. R., Buckingham, D., Oshima-Takane, Y. (1996). Generative connectionist models of cognitive development: Why they work. In Proceedings of Workshop on Computational Cognitive Modeling: Source of the Power. American Association of Artificial Intelligence. - Constructive networks perform well in simulating psychological development because they grow while learning and are sensitive to variation in input amounts.
Shultz, T. R., Schmidt, W. C., Buckingham, D., & Mareschal, D. (1995). Modeling cognitive development with a generative connectionist algorithm. In T. J. Simon & G. S. Halford (Eds.), Developing cognitive competence: New approaches to process modeling (pp. 205-261). Hillsdale, NJ: Erlbaum. Early review of CC models of several phenomena in cognitive and language development. - An interesting argument in support of the realism of batch training over per-pattern training in terms of hippocampal training of cortex.
Tetewsky, S. J., Shultz, T. R., & Takane, Y. (1995). Training regimens and function compatibility: Implications for understanding the effects of knowledge on concept learning. Proceedings of the Seventeenth Annual Conference of the Cognitive Science Society (pp. 304-309). Mahwah, NJ: Erlbaum. - Sub-tasking facilitates learning when the initial subtask involves learning a function compatible with that characterizing the rest of the task, and inhibits learning when the initial subtask involves a function incompatible with the rest of the task.
Shultz, T. R., Buckingham, D., & Oshima-Takane, Y. (1994). A connectionist model of the learning of personal pronouns in English. In S. J. Hanson, T. Petsche, M. Kearns, & R. L. Rivest (Eds.), Computational learning theory and natural learning systems, Vol. 2: Intersection between theory and experiment (pp. 347-362). Cambridge, MA: MIT Press. - CC networks simulate known psychological regularities in the acquisition of personal pronouns, supporting the idea that reversal errors result from listening to directly addressed speech, while success results from listening to overheard speech.
Buckingham, D., & Shultz, T. R. (1994). A connectionist model of the development of velocity, time, and distance concepts. Proceedings of the Sixteenth Annual Conference of the Cognitive Science Society (pp. 72-77). Hillsdale, NJ: Erlbaum. - Stage progressions in understanding concepts related to moving objects are covered by CC networks: identity, additive, and finally multiplicative rules.
Shultz, T. R., Mareschal, D., & Schmidt, W. C. (1994). Modeling cognitive development on balance scale phenomena. Machine Learning, 16, 57-86. - CC networks simulate stages and the torque-difference effect in knowledge of the balance-scale.
Mareschal, D., & Shultz, T R. (1993). A connectionist model of the development of seriation. Proceedings of the Fifteenth Annual Conference of the Cognitive Science Society (pp. 676-681). Hillsdale, NJ: Erlbaum. - Preliminary conference version of Mareschal & Shultz (1999).
Schmidt, W. C. & Shultz, T. R. (1992). An investigation of balance scale success. Proceedings of the Fourteenth Annual Conference of the Cognitive Science Society (pp. 72-77). Hillsdale, NJ: Erlbaum. - Back-propagation networks need to slow convergence in order to cover balance-scale stages.
Shultz, T. R., & Schmidt, W. C. (1991). A Cascade-Correlation model of balance scale phenomena. Proceedings of the Thirteenth Annual Conference of the Cognitive Science Society (pp. 635-640). Hillsdale, NJ: Erlbaum. - Preliminary conference version of Shultz et al. (1994).
Shultz, T. R. (1991). Simulating stages of human cognitive development with connectionist models. In L. Birnbaum & G. Collins (Eds.), Machine learning: Proceedings of the Eighth International Workshop (pp. 105-109). San Mateo, CA: Morgan Kaufmann. - Stages in connectionist models occur as networks solve part of a problem before solving the whole problem.