Adaptive Cognitive Systems

This article surveys the development of mathematical laws, circuits, and architectures that model how our brains make our minds, and shows how these contributions provide a blueprint for developing general-purpose autonomous adaptive algorithms and robots for intelligent applications in engineering and technology. The mathematical laws for short-term memory, medium-term memory, and long-term memory that provide the contemporary foundation for all subsequent biological neural network models were published in 1968, followed by a steady stream of model discovery and development to the present time. The Cohen-Grossberg model and theorem that was published in this journal in 1983 was one step in this series of developments. It proved global limit theorems for a general class of neural networks using a Lyapunov function as one tool. These theorems provide a guarantee that learning in these networks generates stable memories. This article surveys additional mathematical foundations for neural network design and applications and describes a modeling method for the incremental discovery of models with increasingly powerful and general functional capabilities. This method is illustrated by the adaptive resonance theory, or ART, architectures which explain how our brains autonomously learn to attend, recognize, and predict objects and events in a changing world, along the way explaining how our brains become conscious, as well as the computational constraints on learning that forced evolution to discover conscious states of mind. Multiple types of resonances support different kinds of conscious awareness and enable the explanation and prediction of large psychological and neurobiological databases about perception, cognition, emotion, and action. Because ART can be derived from a thought experiment about a universal problem of error correction in a changing world, its expanding applications to the development of autonomously intelligent systems should transform future technologies.

Adaptive Resonance Theory, or ART, is a neural model that explains how normal and abnormal brains may learn to categorize and recognize objects and events in a changing world, and how these learned categories may be remembered for a long time. This article uses ART to propose and unify the explanation of diverse data about normal and abnormal modulation of learning and memory by acetylcholine (ACh). In ART, vigilance control determines whether learned categories will be general and abstract, or specific and concrete. ART models how vigilance may be regulated by ACh release in layer 5 neocortical cells by influencing after-hyperpolarization (AHP) currents. This phasic ACh release is mediated by cells in the nucleus basalis (NB) of Meynert that are activated by unexpected events. The article additionally discusses data about ACh-mediated tonic control of vigilance. ART proposes that there are often dynamic breakdowns of tonic control in mental disorders such as autism, where vigilance remains high, and medial temporal amnesia, where vigilance remains low. Tonic control also occurs during sleep-wake cycles. Properties of Up and Down states during slow wave sleep arise in ACh-modulated laminar cortical ART circuits that carry out processes in awake individuals of contrast normalization, attentional modulation, decision-making, activity-dependent habituation, and mismatch-mediated reset. These slow wave sleep circuits interact with circuits that control circadian rhythms and memory consolidation. Tonic control properties also clarify how Alzheimer’s disease symptoms follow from a massive structural degeneration that includes undermining vigilance control by ACh in cortical layers 3 and 5. Sleep disruptions before and during Alzheimer’s disease, and how they contribute to a vicious cycle of plaque formation in layers 3 and 5, are also clarified from this perspective.

The Department of Defense has pursued the integration of virtual reality simulation into medical training and applications to fulfill the need to train 100,000 military health care personnel annually. Medical personnel transitions, both when entering an operational area and returning to the civilian theater, are characterized by the need to rapidly reacquire skills that are essential but have decayed through disuse or infrequent use. Improved efficiency in reacquiring such skills is critical to avoid the likelihood of mistakes that may result in mortality and morbidity. We focus here on a study testing a theory of how the skills required for minimally invasive surgery for military surgeons are learned and retained. Our adaptive virtual reality surgical training system will incorporate an intelligent mechanism for tracking performance that will recognize skill deficiencies and generate an optimal adaptive training schedule. Our design is modeling skill acquisition based on a skill retention theory. The complexity of appropriate training tasks is adjusted according to the level of retention and/or surgical experience. Based on preliminary work, our system will improve the capability to interactively assess the level of skills learning and decay, optimizes skill relearning across levels of surgical experience, and positively impact skill maintenance. Our system could eventually reduce mortality and morbidity by providing trainees with the reexperience they need to help make a transition between operating theaters. This article reports some data that will support adaptive tutoring of minimally invasive surgery and similar surgical skills.

How does the brain maintain stable fusion of 3D scenes when the eyes move? Every eye movement causes each retinal position to process a different set of scenic features, and thus the brain needs to binocularly fuse new combinations of features at each position after an eye movement. Despite these breaks in retinotopic fusion due to each movement, previously fused representations of a scene in depth appear stable. This is illustrated by moving the eyes after fusing binocular or monocular ("Magic Eye") stereograms. A neural model proposes how the brain does this by unifying concepts about how multiple cortical areas in the What and Where cortical streams interact to carry out 3D boundary and surface perception, spatial attention, invariant object category learning, predictive eye movements, and learned coordinate transformations. Data from single neuron studies and also from psychophysical studies of covert visual attention shifts prior to eye movements (Cavanagh et al., 2010; Duhamel and Goldberg, 1992; Gottlieb, 1992; Gottlieb and Snyder, 2010; Melcher, 2007; Rolfs et al., 2011) are explained. The model clarifies how perceptual, attentional, and cognitive interactions among multiple brain regions (e.g., LGN, V1, V2, V3A, V4, MT, MST, PPC, LIP, ITp, ITa, SC) may accomplish predictive remapping as part of the process whereby view-invariant object categories are learned. This model builds upon earlier neural models of 3D vision and figure-ground separation (e.g., Grossberg 1994; Grossberg and Yazdanbakhsh, 2005) and of the learning of invariant object categories as the eyes freely scan a scene (Cao, Grossberg, and Markowitz, 2011; Fazl, Grossberg, and Mingolla, 2009). A key process concerns how an object’s surface representation generates a form-fitting distribution of spatial attention, or attentional shroud, in parietal cortex that helps to maintain the stability of multiple perceptual and cognitive processes as the eyes freely scan a scene. Meeting abstract presented at VSS 2012

A neural model is described of how spontaneous retinal waves are formed in infant mammals, and how these waves organize activity-dependent development of a topographic map in the lateral geniculate nucleus, with connections from each eye segregated into separate anatomical layers. The model simulates the spontaneous behavior of starburst amacrine cells and retinal ganglion cells during the production of retinal waves during the first few weeks of mammalian postnatal development. It proposes how excitatory and inhibitory mechanisms within individual cells, such as Ca2+-activated K+ channels, and cAMP currents and signaling cascades, can modulate the spatiotemporal dynamics of waves, notably by controlling the after-hyperpolarization currents of starburst amacrine cells. Given the critical role of the geniculate map in the development of visual cortex, these results provide a foundation for analyzing the temporal dynamics whereby the visual cortex itself develops.

1 June 2011 The BRiMS Society and Conference (Behavioral Representation in Modeling and Simulation (BRiMS, brimsconference.org) promote cross-disciplinary communication for basic and applied scientific research in the realm of modeling and simulation of human behavior, with a particular emphasis on defense government-related tasks and behavior. Thus, the BRiMS conference brings together scientists, engineers, practitioners, and application users to discuss modeling behavior ranging from that of individuals to the behavior of whole societies, their interactions, and their implications. For a few days each year, we get to meet to share ideas and experiences, identify gaps in current capabilities, discuss new directions, highlight promising technologies, and showcase applications. This special issue is similar to our previous special issue (Kennedy, Ritter, & Best, 2010) in that it includes four papers based on the award winning conference papers of the 2010 annual conference, reviewed and extended to journal article length. The papers include a new model integrating top-down and bottom-up factors affecting visual target acquisition (Jungkunz & Darken, 2011), the application of a statistical methodology to modeling psychological and cognitive impacts of protective clothing (Mueller et al., 2011), the use of fuzzy cognitive mapping techniques to model situation awareness (Jones et al., 2011), and the challenge of exploration and optimization of cognitive models (Moore, 2011). Overall, they represent how the 2010 conference addressed modeling from small-scale models, for example, predicting eye movements, to large-scale parameter exploration using high-performance computing facilities. At the small-scale end of the range of these papers, Patrick Jungkunz and Christian Darken present models of eye movements during target acquisition in military simulations. They found that a relevance map performed better than a salience map and that scene locations that are semantically relevant predict human eye fixations better than just visual salience. However, the combined approach was not statistically better than the relevance map alone. Their work identified semantically relevant scene locations as the most significant factor in predicting eye fixations and they developed a novel method that supports direct extraction of that information directly from the simulation environment. This work is important for simulations because they too often assume that models can see everything (Ritter, Baxter, Jones, & Young, 2000). Not seeing everything or seeing things that are trying not to be seen is important in adversarial simulations (Best & Gerhart, 2011). The second paper concerns the predicting the impact of the psychological and cognitive stressors on performance. Shane Mueller, Benjamin Simpkins, George Anno, Corey Fallon, Gene McCellan, and Owen Price of Applied Research Associates, report on the

An ongoing debate in developmental cognitive neuroscience is whether individuals with autism are able to learn prototypical category representations from multiple exemplars. Prototype learning and memory were examined in a group of high-functioning autistic boys and young men, using a classic paradigm in which participants learned to classify novel dot patterns into one of two categories. Participants were trained on distorted versions of category prototypes until they reached a criterion level of performance. During transfer testing, participants were shown the training items together with three novel stimulus sets manifesting variable levels of physical distortion (low, medium, or high distortion) relative to the unseen prototypes. Two experiments were conducted, differing only in the manner in which the physical distortions were defined. In the first experiment, a subset of autistic individuals learned categories more slowly than controls, accompanied by an overall diminution in transfer-testing performance. The autism group did, however, manifest a typical pattern of performance across the testing conditions, relative to controls. In the second experiment, group means did not differ statistically in either the training or testing phases. Taken together, these data indicate that high-functioning autistic individuals do not manifest gross deficits in prototypical category learning. A theoretical discussion is given in terms of how perceptual grouping may interact with category learning.