|
|
|
 |
|
Howard Hughes Medical Institute and MIT
Friday 26 June 2009
14.30
Seminar Room B10 (Basement)
Alexandra House, 17 Queen Square, London, WC1N 3AR
The computational challenges of connectomics
Recent innovations in 3d nanoscale imaging are expected to produce teravoxel and petavoxel-sized images of the brain's neural networks. These images pose some fascinating computational problems. Among these, the most immediate challenge is accurate tracing of the "wires" of the brain, its axons and dendrites, through the 3d images. This is an example of a long-standing problem in computer vision known as image segmentation. I will describe the first method for supervised learning based on a genuine measure of image segmentation performance. When applied to tracing neurons, this improves accuracy relative to conventional learning methods. If the tracing problem were solved, it would become possible to find "wiring diagrams" or “connectomes of brain tissue, which in turn would pose further challenges for computational neuroscience“. I will describe how algorithms for graph analysis could be applied to connectomes to test neural network theories of brain function. One of the most exciting prospects would be to decode the memories that are hypothesized to be stored in
connectomes.