Ruggero Micheletto

Yokohama University

A method for the visualization and interaction of simulated neural networks in real time

 

In this talk I want to introduce myself briefly, then show the research topics that we are currently carrying out in my laboratory in the department of nanobiosciences, Yokohama City University, Yokohama, Japan. Mainly we are dealing with digital simulation of spiking neural networks, and doing minor experimental work on perception with real human subjects. I will introduce briefly several experimental activities related to visual perception, perception of optical illusion and describe you a particular new optical illusion that we found originally in our laboratory. Then, I will concentrate on the description our work on simulation of complex neural networks, especially I will focus on our visualization and interaction method we developed. With this, we are able to simulate a neural network and present on the computer screen in real time a representation of the network electrical status. Each neuron is represented by a block and the color and size of the block represents its electrical status that change in real time (of course not an exact physical representation of time, but a near-real-time representation proportional to that). Moreover, we devised a methodology to interact with the network, we can press keyboard keys and alter the status of one or several neurons at will and during the simulation, and see the effects of these alteration on the network in real time. Why we do this? Because the behaviour of neural networks is extremely complex and counter-intuitive if seen after the event by analyzing statistical data, however it can become slightly more interesting and intuitive if observed and interacted in real-time. Afterall we are brains trying to understand simulated brains (!) Also, I will introduce our insight on open problems and give the audience our feeling for a possible explanation of still unexplained neural-network questions, as for example "how is information coded?", "is timing the key?", "spiking rate is important or not?", "resonance is important?" or other pretty difficult questions. We cannot answer to those, but I will try to discuss our intuitive feeling of possible answers