This chapter provides a selective review of the issues that have dominated computational models of associative learning in recent decades. Associative learning research concerns the simplest and most fundamental processes by which humans and other animals come to predict events in their environment based on past experience. It has far-reaching implications for understanding adaptive and maladaptive human behavior. With a focus on Pavlovian conditioning and adjacent subdisciplines, this chapter explores how the prediction error learning algorithm has shaped understanding of competitive learning, selective attention, stimulus representation, and learning about absent events. A number of alternative computational approaches will be introduced, along with some remaining challenges in the computational modeling of human and animal associative learning.

Analogy is a core cognitive capacity encompassing basic similarity (“this is like that”), relational similarity (proportional analogies of the form A:B::C:x), and complex system mappings, in which the elements of one situation are structurally aligned with the elements of another. The latter permits complex inferences from a known source situation to a less familiar target situation. Because of its centrality in human thinking, analogy has been the subject of numerous computational modeling efforts. Models of similarity come from multiple traditions in cognitive science, including associationist approaches (such as connectionist models), “traditional” symbolic approaches (such as graph matching and production systems), and hybrid symbolic/connectionist approaches. This chapter reviews and evaluates several models from these various approaches in terms of their ability to simulate basic similarity, relational similarity, and system mapping.

Motor neuroscience centers on characterizing human movement, and the way it is represented and generated by the brain. A key concept in this field is that despite the rich repertoire of human movements and their variability across individuals, both the behavioral and neuronal aspects of movement are highly stereotypical, and can be understood in terms of basic principles or low dimensional systems. Highlighting this concept, this chapter outlines three core topics in this research field: (1) Trajectory planning, where prominent theories based on optimal control and geometric invariance aim at describing end-effector kinematics using basic unifying principles; (2) Compositionality, and specifically the ideas of motor primitives and muscle synergies that account for motion generation and muscle activations, using hierarchical low-dimensional structures; and (3) Neural control models, which regard the neural machinery that gives rise to sequences of motor commands, exploiting dynamical systems and artificial neural network approaches.

Dynamical systems thinking originated from the sensory-motor domain, but is hypothesized to reach all forms of cognition.Dynamic field theory (DFT) is a mathematically specific, neurally grounded formalization of dynamical systems thinking. Stable states of neural activation, realized as localized activation patterns in low-dimensional neural fields are the units of representation. Their dynamic instabilities lead to the emergence of events at discrete moments in time from continuous-time dynamics. These enable sequences of neural processing steps and flexible binding of multiple localist representations within neural dynamic architectures. Stability enables linking DFT accounts to sensory-motor systems and closed-loop behavior. Instabilities and coordinate transforms are key to reaching the flexibility and productivity of higher cognition. This chapter discusses the relationship between DFT and other approaches to cognition.

This chapter provides an overview of approaches to formal modeling in the domain of categorization. The core psychological processes addressed by models are: generating a classification decision in response to a stimulus and constructing category representations based on supervised experience. A taxonomy is provided that organizes the formal models in terms of their use of a fixed, combined, or constructed approach to predicting categories under either a cue-based or item-based framework. The chapter gives in-depth coverage of a leading approach (exemplar models) as well as an emerging alternative: a constructed cue-based model (DIVA) that differs from competing accounts by learning to reconstruct the input features via sets of category-specific weights and using the degree of reconstructive success (i.e., goodness-of-fit to the category) to determine the likelihood of membership.

Creativity is typically defined as producing something that is novel, useful, and surprising. Such endeavor plays a critical role in the arts and scientific discovery. However, not all creativity is groundbreaking or historically important. As a common cognitive activity, creativity is amenable to scientific investigation leading to a process-based understanding, so it should be possible to propose models and write computer programs simulating the creativity process. However, the path from cognitive models to computational models is still not trodden as often as would be beneficial. This chapter reviews common concepts underlying many computational creativity efforts, namely idea generation, search, and evaluation. Two example computational models are described in more detail, namely the explicit-implicit interaction theory and the CreaCogs architecture. The chapter concludes with a discussion of current shortcomings and future directions for computational creativity as well as discussing promising avenues and successes of current models.

Computer models of the acquisition of cognitive skills build on a long and progressive tradition of research. Since 1979, a wide range of psychologically plausible mechanisms for learning during skill practice have been implemented in computational models. This repertoire of mechanisms goes a long way towards answering the questions implied by Fitts’ (1964) division of practice into three phases: How does skill practice get started? How is a partially learned skill improved during practice? How does a skill change as practice is extended beyond mastery? Nine distinct modes of learning are identified. Each can be implemented in several different ways. The majority of models explain the speed-up of task completion that occurs during practice. There are fewer attempts to model the origin, consequences, and ultimate elimination of errors.

This chapter describes computational models developed to represent basic and applied phenomena of interest to I-O psychology. The basic phenomena of interest relate to motivational, learning, and decision-making processes. The applied phenomena relate to selecting, training, evaluating, retaining, and managing employees. These employees may work in teams, be leaders of others, or engage in action, information sharing, and decision making relevant to organizational outcomes. A computational control systems architecture is used in many of the more basic models, and agent-based modeling as well as control systems modeling are used for the more applied models.

This chapter discusses cognitive social simulation, which lies at the intersection of cognitive modeling and social simulation – two forms of computational modeling. Cognitive modeling focuses on producing precise computational models of individual mental processes, while social simulation centers on models of social processes (such as interaction of individuals or collective decision making). By combining cognitive and social models, cognitive social simulation is poised to address issues concerning both individuals and society. Detailed simulation enables precise analysis of possible scenarios and outcomes (social or individual). A number of examples of cognitive social simulation are sketched in this chapter. Issues involved are discussed. Some promising directions are outlined.

What counts as a philosophical issue in computational cognitive science? This chapter briefly reviews possible answers before focusing on a specific subset of philosophical issues. These surround challenges that have been raised by philosophers regarding the scope of computational models of cognition. The arguments suggest that there are aspects of human cognition that may, for various reasons, resist explanation or description in terms of computation. The primary targets of these “no go” arguments have been semantic content, phenomenal consciousness, and central reasoning. This chapter reviews the arguments and considers possible replies. It concludes by highlighting the differences between the arguments, their limitations, and how they might contribute to the wider project of estimating the value of ongoing research programs in computational cognitive science.

Progress in the computational cognitive sciences depends critically on model evaluation. This chapter provides an accessible description of key considerations and methods important in model evaluation, with special emphasis on evaluation in the forms of validation, comparison, and selection. Major sub-topics include qualitative and quantitative validation, parameter estimation, cross-validation, goodness of fit, and model mimicry. The chapter includes definitions of an assortment of key concepts, relevant equations, and descriptions of best practices and important considerations in the use of these model evaluation methods. The chapter concludes with important high-level considerations regarding emerging directions and opportunities for continuing improvement in model evaluation.

This chapter introduces deep learning (DL) in the framework of experimentalism, taking inspiration from Pierre Oleron’s explanation of human intellectual activities in terms of long (or, deep) circuits. A history of DL is presented, from its origin in the mid-twentieth century to the breakthrough of deep neural networks (DNNs) in the last decades. Architectural and representational issues are then discussed in depth. Convolutional neural networks, the most popular and successful DL algorithm to date, are reviewed in detail. Finally, adaptive activation functions in DNNs are presented in the context of homeostatic neuroplasticity, surveyed, and analyzed.

Many of the problems that human minds need to solve – including learning concepts, causal relationships, and languages – require making informed inferences from limited data. Bayesian models of cognition consider how an ideal agent should solve these problems, drawing on ideas from probability theory, statistics, machine learning, and artificial intelligence.  The resulting models can then be used to understand human behavior, identifying in formal terms the knowledge that human minds draw on when solving these problems and identifying potential mechanisms by which their solutions might be implemented. This chapter provides an introduction to Bayesian models of cognition, starting with the basic principles of probability theory and then considering more advanced topics such as graphical models, causal learning, hierarchical Bayesian models, and Markov chain Monte Carlo. The chapter ends with a brief review of recent theoretical developments.

Categorization is the process of assigning an object or event to a behaviorally relevant group. Before the 1990s, almost nothing was known about the neural networks and processes that mediate human categorization. As a result, theories of categorization were dominated by purely cognitive descriptions. The cognitive neuroscience revolution dramatically increased our understanding of the neural bases of human categorization. As a result, models grounded in neuroscience are becoming increasingly popular. Virtually all of these models assume that different neural systems mediate learning in different types of categorization tasks. Collectively, these models have already made profound contributions to our understanding of human categorization, by widening the empirical domain of categorization research, and by motivating experiments that might not otherwise have been run. Furthermore, this trend should increase in the future, as methods for studying the functioning human brain improve and the database of human brain function during categorization grows.

Reinforcement learning (RL) is a computational framework for an active agent to learn behaviors on the basis of a scalar reward feedback. The theory of reinforcement learning was developed in the artificial intelligence community with intuitions from psychology and animal learning theory and mathematical basis in control theory. It has been successfully applied to tasks like game playing and robot control. Reinforcement learning gives a theoretical account of behavioral learning in humans and animals and underlying brain mechanisms, such as dopamine signaling and the basal ganglia circuit. Reinforcement learning serves as the “common language” for engineers, biologists, and cognitive scientists to exchange their problems and findings in goal-directed behaviors. This chapter introduces the basic theoretical framework of reinforcement learning and reviews its impacts in artificial intelligence, neuroscience, and cognitive science.

This chapter reviews contemporary computational models of psychological development in a historical context, including those based on symbolic rules, artificial neural networks, dynamic systems, robotics, and Bayesian ideas. Emphasis is placed on newer work and the insights that simulation can provide into developmental mechanisms. Within space limitations, coverage is both sufficiently broad to provide a general overview of the field and sufficiently detailed to facilitate understanding of important techniques. Prospects for integrating the dominant approaches of neural networks and Bayesian methods are explored. There is also speculation about how deep-learning networks might begin to impact developmental modeling by increasing the realism of training patterns, particularly in visual perception.

After a brief orientation to logic-based (computational) cognitive modeling, the necessary preliminaries are discussed in this chapter (e.g., what a logic is, and what it is for one to “capture" human cognition are explained). Three “microworlds" or domains that all readers should be comfortably familiar with (natural numbers and arithmetic; everyday vehicles; and residential schools, e.g., colleges and universities) are introduced, in order to facilitate exposition in the chapter. Then the ever-expanding universe U of formal logics, with an emphasis on three categories therein, is given: deductive logics having no provision for directly modeling cognitive states; nondeductive logics suitable for modeling rational belief through time without machinery to directly model cognitive states such as believes and knows; and finally, nondeductive logics that enable the kind of direct modeling of cognitive states absent from the first two types of logic. Then, there follows a focus spcifically on two important aspects of human-level cognition to be modeled in logic-based fashion: the processing of quantification, and defeasible (or nonmonotonic) reasoning. Finally, a brief evaluation of logic-based cognitive modeling is provided, together with some comparison to other approaches to cognitive modeling, and the future of the discipline is considered. The chapter presupposes nothing more than high-school mathematics of the standard sort on the part of the reader.

While psychiatry has made great strides in recent decades toward integrating our increasing understanding of the biological bases of cognition, it nonetheless continues to suffer from imprecise diagnostics and blunt treatment options. Recent advances in computational neuroscience have the potential to address these issues, with a range of neural and cognitive models offering the possibility of a more precise psychiatric nosology with more targeted therapeutics. Here we review a variety of these models, with a special emphasis on their application to addiction, psychosis, anxiety disorders, depression, obsessive-compulsive disorder, autism spectrum disorder, and attention-deficit hyperactivity disorder. We then close with a discussion of potential challenges in incorporating these insights and methods into a clinical setting.