Abstract
The model NETNET (NEUronen NETz) was developed in 1974 by Prof. Reichl at the Johannes Kepler University Linz und publicized in 1975.
This neural-net-similar associative memory model NEUNET is able to store and reproduce patterns reading from its receptors. The model was improved continuously and NEUNET-3 was the first enhancement, which was able to store patterns without losses of information. This paper describes an enhancement of the model, termed Selective NEUNET, to associate faulty inputs with weighted original patterns. The network is self-learning, the correction of inputs, the storage of new information and the calculation of weights which enables the model to store information with various priorities is self-organized. The priority of information is increased if it is frequently used and in case of the association of unknown inputs it is more like to be associated with a high-priority original pattern. This usage of weights is the main enhancement of the model described in this contribution. Previously the network learned new information by enhancing its network-structure, now the weights can be adjusted in addition. It enables the model to separate important information from a huge amount of any data.
Translated title of the contribution | Selective association with associative memory model NEUNET |
---|---|
Original language | German (Austria) |
Publication status | Published - Nov 2006 |
Fields of science
- 102001 Artificial intelligence
- 102006 Computer supported cooperative work (CSCW)
- 102010 Database systems
- 102014 Information design
- 102015 Information systems
- 102016 IT security
- 102028 Knowledge engineering
- 102019 Machine learning
- 102022 Software development
- 102025 Distributed systems
- 502007 E-commerce
- 505002 Data protection
- 506002 E-government
- 509018 Knowledge management
- 202007 Computer integrated manufacturing (CIM)
- 102033 Data mining
- 102035 Data science