The book under review is a very important one, and gives an an excellent example how to combine computational neuroscience with the theory and practice of cognitive architectures and even hardware implementations. The book helps to neural modelers to understand the challenges coming from the field of cognitive systems, architectures including robots, and cognitive scientists and engineers to learn more about the methods of computational neuroscience to understand the underlying neural mechanisms and computational algorithms.
The first large part (Computational Neuroscience Models) contains ten papers. The general goal of this part was to collect works on presenting neural network models of different aspects of the perception—action cyclic process.
Perception, itself is studied in two papers. Tsotsos and Rothenstein argue that the brain does not solve the generic problem of visual search, but by the aid of attention it helps to create a computationally solvable problem. Ursino et al. reviews neural mechanisms of multisensory integration in audio-visual and visual-tactile systems respectively, constrained by neurophysiological data. It seems to be some general features behind integration, such as lateral inhibition and excitation, nonlinear processing, competitive mechanism etc. Memory and learning, as the most important cognitive functions, were reviewed from a computer scientist’s (Andrew Coward) perspective to find mapping between the psychological and physiological level descriptions.
One of the central points of this part is to review the underlying neural models of decision making processes. Levine offers a mechanism based on the interaction of a number of brain regions to model decision making under risky conditions. The main point is how emotional evaluation and numerical calculation interact. The role of error monitoring in managing conflict resolution and decision making was analyzed by Brown and Alexander. Specifically, the importance of a brain region, namely the anterior cingulate cortex was emphasized. Related to conscious and unconscious processing a neurally plausible network framework was suggested by Rolls based on probabilistic attractor network models.
Kinematic of robots is a featured topic in robotics, Gentili et al. presented a neural network model, showed simulations, and discussed the scope and limits of biological plausible models. I found very interesting the paper of Nishimoto and Tani. They connected brain modeling, schemata learning and robot applications by adopting the concept and model framework of multiple time-scale neural networks.
As always, John Taylor in his two papers combined broad conceptual overviews with specific details ready to applications. In the first paper he reviewed the main results of the EU GNOSYS project to build brain-inspired reasoning robot, in the second one he gives a comparative analysis of models of consciousness. It is not the reviewer’s task to decide whether or not the model called CODAM (established by Taylor’s own team) is the clear winner of the competitive comparison.
The second part is about cognitive architectures, algorithms and systems.
Another paper about CODAM (Rapantzikos et al.) emphasizes the necessity of integrating bottom up and top down algorithms, and the dynamic character of the attentional selection mechanisms to have a successful computational model of visual attention. In their implementation there is an interplay between a higher level control system and lower level cortical representation. Petridis and Perantonis applied a new methodology based on graph theory to extract semantics form multimedia data.
The cognitive architecture CLARION (Connectionist Learning with Adaptive Rule Induction On-line) elaborated by Ron Sun and his team was mentioned in two papers. Qi Zhang reviewed the algorithms of learning episodic and semantic memory. Sun and Wilson simulated motivational processes, and the challenge is to define appropriate motivational variables, and relations among them. The topic of error monitoring, conflict resolution and decision making came back in the paper of Lima, in particular in application to robots. Petri nets proved to be a good tool for plan representation.
The next two papers contribute to the theory and application of cognitive robots. Karigiannis et al. is interested in the ontogenetic development of learning. Different versions of reinforcement learning algorithms are successfully adopted to solve kinematic problems. Mohan et al. worked also in the framework of GNOSYS, and built algorithms for the internal models of real and imagined actions. One important aspect is the learning of the reward structure.
Computational algorithms for reasoning and knowledge representation are reviewed by d’Avila Garcez and Lamb. A connectionist algorithms was give to learn from experience. Tishby and Polani investigated the circular flow of information in the perception–action cycle by the classical syntactical information theory. Specifically, they are interested in analyzing reward-driven decision process. Chella and Manzotti reviews the ambitious project of artificial consciousness. I agree with the authors: “Although a complete theory of consciousness will very probably require some ground-breaking conceptual revisions, the deliberate attempt at mimicking consciousness will pave the way to a future scientific paradigm to understand the mind.
The third small, illustrative part demonstrates the possibilities of hardware implementations. Lim’s chapter (Smart sensor networks) shows applications of perception-reason-action networks for distributed multi-robot systems. Tang and Murray presented artificial neural networks, as appropriate methods for sensory data fusion. Data fusion should have a major role in wireless microsystems.
The book certainly provides an overview about the neural and cognitive architectures, algorithms and systems for the perception -action systems. I missed the mentioning of one concept, namely “circular causality”. I see the book as a well-edited whole, which offers a coherent perspective, and contains many remarkable details. I recommend to anyone who would like to learn more about the field.