Tumble Tree - Reducing Complexity of the Growing Cells Approach

We propose a data structure that decreases complexity of unsupervised competitive learning algorithms which are based on the growing cells structures approach. The idea is based on a novel way of ordering the cells in a tree like data structure in a way that random access during training is replaced by tree traversals. Overall time complexity is reduced from O(n2 ) to O(n log n) which opens new application fields to the growing cells structures approach. Key words: neural networks, unsupervised learning, reinforcement learning, growing cells structures, growing neural gas.

• Type: Conference
• Year: 2010
• Where: ICANN
• Authors: Hendrik Annuth, Christian-A. Bohn
• Keywords: Cells Structures, Data Structure, ICANN 2010, Neural Networks, Unsupervised Competitive Learning