CiteULike: suizan's neural_network

Connectivity and Dynamics of Neuronal Networks as Defined by the Shape of Individual Neurons

Sebastian Ahnert — 2008-05-13T17:08:05-00:00

(12 May 2008)

Neuronal networks constitute a special class of dynamical systems, as they are formed by individual geometrical components, namely the neurons. In the existing literature, relatively little attention has been given to the influence of neuron shape on the overall connectivity and dynamics of the emerging networks. The current work addresses this issue by considering simplified neuronal shapes consisting of circular regions (soma/axons) with spokes (dendrites). Networks are grown by placing these patterns randomly in the 2D plane and establishing connections whenever a piece of dendrite falls inside an axon. Several topological and dynamical properties of the resulting graph are measured, including the degree distribution, clustering coefficients, symmetry of connections, size of the largest connected component, as well as three hierarchical measurements of the local topology. By varying the number of processes of the individual basic patterns, we can quantify relationships between the individual neuronal shape and the topological and dynamical features of the networks. Integrate-and-fire dynamics on these networks is also investigated with respect to transient activation from a source node, indicating that long-range connections play an important role in the propagation of avalanches.

A model of grounded language acquisition: Sensorimotor features improve lexical and grammatical learning

Steve Howell — 2005-06-21T09:18:01-00:00

Journal of Memory and Language, Vol. 53, No. 2. (August 2005), pp. 258-276.

It is generally accepted that children have sensorimotor mental representations for concepts even before they learn the words for those concepts. We argue that these prelinguistic and embodied concepts direct and ground word learning, such that early concepts provide scaffolding by which later word learning, and even grammar learning, is enabled and facilitated. We gathered numerical ratings of the sensorimotor features of many early words (352 nouns, 90 verbs) using adult human participants. We analyzed the ratings to demonstrate their ability to capture the embodied meaning of the underlying concepts. Then using a simulation experiment we demonstrated that with language corpora of sufficient complexity, neural network (SRN) models with sensorimotor features perform significantly better than models without features, as evidenced by their ability to perform word prediction, an aspect of grammar. We also discuss the possibility of indirect acquisition of grounded meaning through "propagation of grounding" for novel words in these networks.

Neural network processing of natural language: I. Sensitivity to serial, temporal and abstract structure of language in the infant

PF Dominey — 2005-12-20T23:46:42-00:00

Language and Cognitive Processes, Vol. 15, No. 1. (1 February 2000), pp. 87-127.

Well before their first birthday, babies can acquire knowledge of serial order relations (Saffran et al., 1996a), as well as knowledge of more abstract rule-based structural relations (Marcus et al., 1999) between neighbouring speech sounds within 2 minutes of exposure. These early learners can likewise acquire knowledge of rhythmic or temporal structure of a new language within 5-10 minutes of exposure (Nazzi et al., 1998). All three of these types of knowledge likely play invaluable roles in "bootstrapping" language acquisition. Two important open questions that remain include: What are the mechanisms that provide this rapid learning ability, and how do they depend on pre-exposure to the environment? Here we show that a neurophysiologically validated temporal recurrent network simulates babies' capabilities to learn serial order and rhythmic structure. Indeed the recurrent network is capable of representing serial and temporal structure with no pre-exposure, and through exposure these internal representations can become bound to behavioural responses. In contrast, babies' performance in extracting abstract structure can only be simulated by a modified version of the model. We thus demonstrate how innate representational capabilities for serial and temporal structure of language could arise from a common neural architecture, distinct from that required for the representation of abstract structure, and we provide a predictive testable model of at least these aspects of the initial computational state of the language learner.