| ECVision indexed and annotated bibliography of cognitive computer vision publications This bibliography was created by Hilary Buxton and Benoit Gaillard, University of Sussex, as part of ECVision Specific Action 8-1 The complete text version of this BibTeX file is available here: ECVision_bibliography.bib
ABSTRACT
This book provides mechanistic descriptions of simple and complex cognitive behaviors. Part I describes several neural network theories applied to classical conditioning. Through classical conditioning, animals build a model of the world that predicts future events and, thereby, they become able to minimize the impact of environmental perturbations, replacing a negative feedback system that partially decreases the environmental effect by a more powerful feedforward mechanism that substantially diminishes it. Chapter 2 describes a neural network that modifies the internal model of the world when predicted events differ from observed events. In consequence, more reliable CSs prevent less reliable or salient ones from becoming associated with the US. When the difference between predicted and observed events cannot be reduced by modifying CS-US associations, the model assumes that elementary CSs are combined into internal compound representations that are associated with the US to solve the problem at hand. Chapter 3 describes a model of the world that combines classical conditioning associations (declarative representations of knowledge according to Dickinson, 1980), thereby infering new knowledge from previously experienced situations. The model increases the processing of those CSs associated with novelty, i.e., those associated with an increase of environmental information (defined as the inverse of probability of an event). Chapter 4 presents a neural network that assumes that CSs compete for a limited short-term memory capacity according to their salience and the value of their associations with the US. In consequence, the most reliable or salient predictors are selected to become associated with the US or other CSs. Therefore, competition for a limited short-term memory is an alternative to the mismatch mechanism described in Chapters 2 and 3 to reduce redundancy. Although in many cases the CR is considerably analogous to the UR, there are some cases in which the CR might be the opposite of the UR, i.e., an opponent process. Chapter 5 describes a neural network that predicts whether a CR will be analogous or opposite to the UR. Finally, Chapter 6 introduces a neural network in which the model of the world, in addition to predicting that event A will follow event B, describes what the time this occurs. Part II describes a neural network applied to operant conditioning. Chapter 8 introduces a neural network that learns different escape and avoidance responses by combining classical conditioning with a {"}response-selection{"} mechanism. Part III describes neural networks applied to animal cognition. Chapter 10 describes a neural network that accomplishes place learning by assuming that animal behavior is controlled by a goal-seeking mechanism that approaches a US or a CS that predicts the US making use of a spatial map built with classical conditioning mechanisms. Place learning is interpreted as an property of the model of the world that forecasts what event is going to occur, when in time, and where in space. Chapter 11 shows a neural network that depicts maze learning (and some problem solving tasks) by using a goal-seeking mechanism that approches the US or a CS that predicts the US making use of a cognitive map built with classical inference mechanisms as those described in Chapter 4. Part IV shows how neural network models permit to simultaneously develop psychological theories and models of the brain. Chapter 13 addresses the physiological basis of animal learning by showing how the network introduced in Chapter 3 can be mapped onto different regions of the brain including the hippocampus, cortex, septum, and cerebellum. |
|
| Site generated on Friday, 06 January 2006
|