A self-organising network that grows when required

By Stephen R. Marsland, Jonathan L. Shapiro, Ulrich Nehmzow. 2002

418 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Cognitive Neuroscience, Computer Vision and Pattern Recognition, Neural Architecture Search & AutoML

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Cognitive Neuroscience

In Computer Vision and Pattern Recognition

In Neural Architecture Search & AutoML

Citations in the corpus

On the role of neurogenesis in overcoming catastrophic forgetting (2018, methodology): “he weight vector of the BMU at t 1, and C k 2Rn is the global context of the network with C k(t 0) = 0. If K= 0, then Eq.2 resembles the learning dynamics of the standard GWR without temporal context [12]. For a given input x(t), the activity of the network, a(t), is deﬁned in relation to the distan”
Lifelong Learning of Spatiotemporal Representations With Dual-Memory Recurrent Self-Organization (2018, methodology): “, 2017) is a recurrent extension of the Grow-When-Required (GWR) self-organizing network (Marsland et al., 2002) that embeds a Gamma memory (Principe et al.”
An Incremental Self-Organizing Architecture for Sensorimotor Learning and Prediction (2017, methodology): “The proposed learning architecture consists of a hierarchy of GWR networks [15] for processing input data sequences and learning inherent spatiotemporal dependencies (Fig.”
Lifelong learning of human actions with deep neural network self-organization (2017, background): “Lifelong Learning In growing self-organizing networks such as the GWR (Marsland et al., 2002), 310”
A self-organizing neural network architecture for learning human-object interactions (2017, methodology): “Growing self-organizing approaches such as the GNG [50] and he GWR networks [17] are characterized by a dynamic topological tructure able to adapt toward the input data space through the echanism of the competitive Hebbian learning [66] .”
Knowledge Capture and Replay for Continual Learning (2020, neutral): “, 2016] [Marsland et al., 2002] [Hung et al.”
Emergence of multimodal action representations from neural network self-organization (2017, methodology): “Hierarchical learning from contiguous Growing When Required (GWR) networks (Marsland et al., 2002) shapes a functional hierarchy that processes spatiotemporal visual patterns with an increasing level of complexity by using neural activation trajectories from lower-level layers for training…”
Self-organizing neural integration of pose-motion features for human action recognition (2015, methodology): “The Growing When Required (GWR) algorithm by Marsland et al. (2002) decides when to add new neurons by evaluating the activity of that neuron that best matches the current input.”
Combining Self-organizing Maps with Mixtures of Experts: Application to an Actor-Critic Model of Reinforcement Learning in the Basal Ganglia (2006, background): “It concerns the classical Kohonen algorithm [12], which requires the number of experts to be a priori set; the Growing Neural Gas algorithm [6], improved by [9], which adds a new expert when an existing expert has a important error of classification; and the Growing When Required algorithm [15], whi”
Multi-modal integration of dynamic audiovisual patterns for an interactive reinforcement learning scenario (2016, methodology): “The standard unsupervised training algorithm was presented in [23].”
Online Task-free Continual Learning with Dynamic Sparse Distributed Memory (2022, background): “In the literature, there exist several expandable architectures [62, 40, 14] and they compare the to-BMU distance or similar metric to a pre-defined threshold to decide whether a new node or layer is added.”
Intentional motion on-line learning and prediction (2008, methodology): “This has motivated us to try a different algorithm: Grow When Required [ 26 ] which seems to build better representations of areas having low observation likelihood.”
Class Incremental Learning for Video Action Classification (2021, methodology): “The activated neurons are updated as follows: For i ∈ { B, S } , η i and µ i are constant learning rate. h i is the ﬁring counter of the neuron [21] and it represents how frequently the neuron is activated.”
Emotion Recognition from Body Expressions with a Neural Network Architecture (2017, methodology): “The GWR [15] is a growing self-organizing network that learns the prototype neural weights from a multi-dimensional input distribution.”