Notwithstanding the integrative and interdisciplinary nature of Computational Neuroscience (and Artificial Intelligence) research in the field is still carried out by working neuro-scientists and working computer scientists who inevitably, and thankfully, bring with them their own distinctive methods, goals, theories and, of course, prejudices.
On taking on this editorial project we did not want to edulcorate this basic fact. On the contrary, we aimed at presenting the reader with a panorama of what people investigating in the area of Computational Neuroscience really do. More often than not the alleged commonalities get diluted when we get into the details of an algorithm or the intricacies of a neuro-psychological model. More often than not we discover that sharing an ontology hides important differences in the way Computational Neuroscience is understood and approached by the different communities that fall under that name.
We can thus claim that this volume gives a true account of the area’s heterogeneity. As a side-effect coherence in style and presentation were not a priority and for that we apologise to the readers: Following the chapters may be challenging but we believe it is a challenge worth taking.
This book is different from others in that we have not sought the collaboration of computational neuroscientists who dwell in middle ground, that is, of computer scientists who have never set foot in a lab or neuro-scientists who use but barely understand the complexities of computational models. But rather of specialists in neuro-science and computer science who for one reason or another are compelled to crisscross their areas of expertise.
As editors we have been as flexible as possible. Some chapters are predominantly technical and assume previous knowledge whereas other authors have followed a more didactic and openly interdisciplinary style. We have opted for a free-hands editorial approach: We understand that in so doing any reader would find something of interest in this book, be it a professional computer scientists or an undergraduate psychology student. This unorthodox approach in which surveys appear along with experimental reports and theoretical analyses makes the book demanding yet, we hope, entertaining.
The overall layout of the book is classical though: On the one hand, it exposes the extent to which neuro-scientists use computational methods and tools in building accurate neuro-psychological models; on the other hand, it reports on how computer scientists use neuro-psychological theories in developing efficient learning algorithms. Following this general scheme we have structured the book in four sections:
Section 1 presents examples of the kind of empirical work carried out by behavioral neuroscientists who use computational tools to facilitate the description and integration of results and to assist in elaborating new predictions.
We start this section with Honey and Grand’s chapter, in which a multi-layer artificial neural network (ANN) is used to model how animals may learn configural discriminations related to the XOR problem.
Cowell, Saksida and Bussey also use ANNs, this time to examine the organization and function of the ventral visual-perirhinal stream in the brain. More specifically they run computational simulations to demonstrate that the effects of brain damage on both visual discrimination and object recognition memory may not be due to an impairment in a specific function such as memory or perception, but are more likely due to compromised object representations in a hierarchical and continuous representational system.
However useful they might be, ANNs are not the only computational method neuro-scientists make us of. As an example, Jennings and collaborators introduce in the third chapter a model of overshadowing and temporal phenomena using Temporal Difference –a well-known algorithm in machine learning.
Section 2 is a compendium of different theoretical approaches to learning and behaviour built directly upon computational models and methods.
Schmajuk and Kutlu’s chapter presents a complex ANN that models attentional-associative processes and is able to describe a large number of classical paradigms.
Along the same line, McLaren proposes a connectionist model, the Adaptively Parameterised Error Correcting System (APECS) that provides insights in several learning problems from sequential learning to human contingency learning.
Vogel and Ponce give a theoretical overview on the mechanisms underlying Pavlovian conditioning, with an accent on how computational models have been shamelessly merged with psychological ones. The integration of various models in explaining such fundamental phenomena is a success story in computational neuroscience.
This chapter is ideally complemented with Ludvig, Bellemare and Pearson’s, who take a computational problem, Reinforcement Learning, and investigate it from different inter-related perspectives, computer science, neuroscience and even economics.
The third section of the book is dedicated to illustrate how neuro-science models have contributed to solve computational problems in Artificial Intelligence and Robotics.
Asghari-Oskoei and Hu’s chapter explores how in building robots for the disabled precise psychological models and methods must be taken into account. The authors advocate that in building such robots engineering or purely computational aspects are necessary but not sufficient.
Nehmzow's contribution offers an innovative approach to the modeling of computational entities, agents and robots, in which quantitative approaches take precedence. We think it is instructive to draw parallelisms between his work, inspired in the study of complex systems and their typically non-linear dynamics and recent studies on networks, neural or not, natural or artificial.
The third chapter, by Husbands, Philippides and Seth, reviews the use of neural systems in Robotics with particular emphasis on strongly biologically inspired neural networks and methods. It starts with an elegant historical introduction on the subject and then proceeds in a two-fold manner: First it gives examples of applications of artificial neural systems in Robotics then on Robotics tools used in neuroscience.
The fourth section reports work rarely accounted for in Computational Neuroscience handbooks that tend to focus on research carried out in academic institutions. We dedicate two chapters to the views of researchers working in Industry on how Computational Neuroscience can be applied to real-life domains. Interestingly, their opinions diverge.
Bisset’s contribution expresses serious doubts about the role played by Neural Computation in engineering and Robotics, a critical view that will no doubt provoke a lively and necessary debate about Computational Neuroscience research, its hype and claims, and how things work in down-to-earth industrial applications.
On the contrary, Ryman-Tubb reports on several neural approaches in Fraud Management Systems that are already in place in the banking sector with an emphasis on the integration of symbolic and connectionist methods –which proves that, at least in some cases, research is driven by business.
Finally, we have taken the liberty of adding a chapter of our own, since the content fitted the purpose of the book in an intriguing way. In it we argue that computational models in behavioral neuroscience must be taken with caution --a strange epilogue from the editors of a book on Computational Neuroscience-- and advocate for the study of mathematical models of existing theories as complementary to neuro-psychological models and computational models.
We would like to finish with a note of thanks to all who made this book a reality. To Joel Gamon, who promptly and efficiently answered our many questions, and of course to the authors.
Sadly, we shall close this preface with sorrowful news: Ulrich Nehmzow died of cancer shortly after submitting his final draft. His attitude and commitment were, to the very last moment, an inspiration. This book is a tribute to him. Requiescat in pace.