We investigate the information retained in neuronal activity after a stimulus has been briefly presented. We employ an artificial neuron (simulated on a computer) to extract information from the activity recorded from a number of real neurons simultaneously. As shown in the figure, the attached readout neuron uses a computer-implemented algorithm to learn the weights of connections (strengths of ‘synapses’) to the real neurons. These connections need to be optimized for detection of a certain visual stimulus.
These experiments have several goals. For one, important insights have been obtained about the nature of iconic memory. We discovered that the brain has a one-back memory for visual stimuli. Neural responses to an image contain as much information about that image as about another image presented immediately before.
Also, we learn about the mechanisms by which information is represented in distributed neuronal activity. We can investigate the conditions under which the information is available in neuronal activity and the time course of the available information. By doing so, we can investigate the current theories about cortical information processing (e.g., liquid state machine).
These experiments have also important practical implications for advancement of neuroprostetic technologies. To connect electronic or robotic devices to the brains of impaired human patients, it is necessary to understand the principles of information coding, the types of information detectable from neuronal activity, the pitfalls and the limitations of such technologies. Our experiments investigate these questions by extracting stimulus-related information from the primary visual cortex (Nikolić et al., 2009).
Nikolić, D.*, S. Häusler*, W. Singer and W. Maass (2007)
Temporal dynamics of information content carried by neurons in the primary visual cortex.
Advances in Neural Information Processing Systems (NIPS), 19