Cognitive Scientist, Computer Scientist
I do a lot of work using different kinds of artificial neural networks, but I have also invented novel variants of artificial neural networks and novel neural networks based architectures. For example, I have invented some new variants of the self-organizing map invented by Teuvo Kohonen. A self-organizing map consists of a matrix of artificial neurons that after a learning phase can represent a certain type of stimulus in an orderly manner, for example, by exposition to a variety of colours, it can learn to represent these so that all blue shades are arranged in a certain part of the map, while all green and yellow lays represented in other parts. The transitions in the representations are gradual. One variant of the self-organizing map I have invented is the Associative Self-Organizing Map that can learn to associate the activity in its self-organized representation of input data with arbitrarily many sets of parallel inputs and with arbitrarily long time delays. Another variant is the Tensor Multiple Peak SOM that self-organizes into a mapping that approximates the mathematical operation tensor-product, while avoiding combinatorial explosion. When it comes to supervised neural networks I have, for example, been involved in work on creating novel self-organizing neural networks architectures based on self-organizing maps, and associative self-organizing maps for a recurrently connected supervised neural network architecture.
Some Related Publications
Gil, D., Garcia, J., Cazorla, M. and Johnsson, M. (2014). SARASOM - A Supervised Architecture based on the Recurrent Associative SOM. Neural Computing and Applications.
Johnsson, M. (2012). (Ed.) Applications of Self-Organizing Maps, InTech. ISBN: 978-953-51-0862-7.
Viejo, D., Garcia, J., Cazorla, M., Gil, D. and Johnsson, M. (2012). Using GNG to improve 3D features extraction - Application to 6DoF Egomotion. Neural Networks, 32, 138-146, ISSN 0893-6080.
Viejo, D., Garcia, J., Cazorla, M., Gil, D. and Johnsson, M. (2011). Using 3D GNG-Based Reconstruction for 6DoF Egomotion. Proceedings of the International Joint Conference on Neural Networks (IJCNN) 2011, San Jose, California, USA. 1207-1213. ISBN: 978-1-4244-9636-5.
Gil, D., Garcia, J., Cazorla, M. and Johnsson, M. (2011). Predictions Tasks with Words and Sequences: Comparing a Novel Recurrent Architecture with the Elman Network. Proceedings of the International Joint Conference on Neural Networks (IJCNN) 2011, San Jose, California, USA. 1042-1048. ISBN: 978-1-4244-9636-5.
Gil, D., and Johnsson, M. (2010). Supervised SOM Based Architecture versus Multilayer Perceptron and RBF Networks. In Bol, R. (Ed.) Proceedings of SAIS 2010, Uppsala, Sweden. 15-24.