A basic aspect of understanding the neural code of a neuron (or a neural system), is the ability to form a dictionary from the stimuli presented to the neuron (or the system) and the patterns of spikes that the neuron responds with. As neurons may respond unreliably to their stimuli, such a dictionary will be stochastic by nature. If the neuron responds to many different stimuli in a similar way (i.e. the nuber of stimulus features that the neuron "cares about" is small), then the dictionary can be compressed, without a significant loss of its properties. Here the authors apply the agglomerative information bottleneck algorithm to study the properties of the dictionary (and nerual code) of the identified H1 neuron in the fly visual system. They find that the neural code dictionaries of different files are highly compressible, suggesting that a small number of features are the key compenents of the H1 neural code. They also compare the encoded features of the different flies and find similar general structure, but differences in the details.

1:30pm -3:00pm, HSE 810.