In this book the author presents a method for implementing delay learning in artificial neural networks. She shows that such a system bears some behavioural (and possibly, at a deeper level, conceptual) similarities to the bilogical methods by which living creatures accomplish the same tasks. The book begins with an overview of the existing neural network paradigms which address some features of delay learning. The attention-driven buffering approach is described; this allows a system of finite size to learn about action-reinforcement associations, even in situations where reinforcements are delayed indefinitely and may be interleaved with others. This approach is shown to be applicable to an operant conditioning task - which is also an example of delay learning in animals. the biological relevance of the attention-driving buffering approach is discussed in later chapters where it is compared in detail with the learning system in one specific animal - the octopus - and with higher animals such as mammals.
Reihe
Sprache
Verlagsort
Zielgruppe
Für höhere Schule und Studium
Für Beruf und Forschung
Illustrationen
Maße
Höhe: 234 mm
Breite: 156 mm
Gewicht
ISBN-13
978-0-412-45050-1 (9780412450501)
Copyright in bibliographic data is held by Nielsen Book Services Limited or its licensors: all rights reserved.
Schweitzer Klassifikation
Part 1 Introduction: definitions of learning and reinforcement; exploratory learning in a neural network. Part 2 Neural networks and learning with delayed reinforcement: mapping onto pattern association tasks; history maintenance; prediction-driven reinforcement; scope for new models. Part 3 RAM-based nodes and networks: introduction and motivation; bit-addressable RAM-nodes; the probabilistic logic node; Omega-state PLNs and exploratory learning algorithms; pRAMs, RAM-nodes and continuous values; PLN parameters; Iota - the number of inputs to a node; Phi-rho - the output probability function; omega - the cardinality of the stored value alphabet; PLN parameters - conclusions; RAM-based nodes - biological connections. Part 4 Attention-driven buffering: the ADB approach; an example; the location of the buffer; an example "bug"; comparison with other delay learning methods. Part 5 Analysis of parameters: reinforcement delay and buffer size; training set and learning rule; learning system topology; construction of nodes; generalization tests; attention-driven buffering - conclusions. Part 6 An ADB system for operant conditioning: the operant conditioning task; description of the simulation; the unspecific effect; operant conditioning experiments with OVSIM; simple discrimination tasks; delay to attack stimuli; task interference and relearning; transfer of discrimination learning; tasks involving multiple discrimination; OVSIM as an operant conditioning model. Part 7 OVSIM and the octopus - the case for modelling: an overview of the octopus visual learning system; OVSIM as a model of the octopus visual attack learning system; damage learning experiments; another model of the octopus visual learning system. Part 8 ADB and delay learning in higher animals: physiological relevance of ADB objectives; possible methods; possible mechanisms for ADB.
