Tomorrow I’ll be attending the International joint Conference on Neural Networks. This is one of the most important events in the area of Artificial Intelligence, and will be performed in the City of Dallas, TX.
A huge variety of tutorials will be presented such as “Cognitive Computational Intelligence”, “Random Neural Network and Applications in Engineering and Biology”, and many more. Also, I’ll be attending the workshops: “What Language And Emotion Can Tell Us About the Brain: New Methods of Analysis” and “Perception and Cognition in the Brain: Integrating Single Cell Recordings, Spiking Neurons, and a Brain Theory”. Which are both quite interesting for my research in neural models.
In this event, I’ll present a new training method for Polynomial Cellular Neural Networks (PCNN): These models are capable of implementing powerful classification mechanisms, Implementations of cellular automata, and even the hypothesis that these models are capable of computational universality in the Turing Sense. In other words, If a polynomial cellular neural network is capable of achieving the computational universality, it is able to execute any algorithm, and be equivalent to any general purpose computer.
The first step toward this goal is to implement any cellular automata in a PCNN, using a single layer of neurons. But, the main challenge is the determination of the synaptic weights: If you are able to choose appropriately the synaptic weights of a PCNN, you can achieve any behavior you desire.
In the presentation, a root location training method will be described: The algorithm receives the behavior as an input and straightforwardly obtains the synaptic weights that implements your desired behavior. For now, the implementations are limited to the Totalistic Cellular Automata behaviors, but future work will extend this limit to the Semitotalistic and Universal Cellular Automata behaviors.
The paper (and proof of the theorem that was mentioned in the conference) can be found here: A root location training method for PCNN
After the presentation, the slides will be available here. (and perhaps a video ;D ).
Here is a brief list of another cool things you can model and build with Cellular Automata and Cellular Neural Networks (which are not my work, all credits goes to their respective authors) :
- Dentritic crystal growth: http://www.youtube.com/watch?v=6COiKEYBg84
- Road traffic simulation with cellular automata: http://www.youtube.com/watch?v=i6tyAiBnCcM
- A robot that learns with cellular neural networks: http://www.youtube.com/watch?v=mRn98UzDgiI
- Morphogenesis: common cheetah skin pattern generated with cellular automata: http://www.youtube.com/watch?v=a1I8enP3TS4