He has posted many of his articles, if not most, on his site as PDFs, which is very cool. I'll include one article relevant to this talk at the bottom. This article comes from Wired UK.
By Duncan Geere |16 July 2010
"I am my connectome" intones the audience in the Player's Theatre in Oxford at TEDGlobal 2010, under the orders of computational neuroscientist Sebastian Seung. "But you don't even know what a connectome is yet!" smirks Seung, before embarking on a precise, detailed explanation about what might be at the root of identity.
The connectome is a map. It's a map of the neurons in your brain and how they interact with each other -- which are connected, where, and how strongly. Some neuroscientists believe that it's this map that defines who we are -- associating memories with other things in your brain, in the same way that a smell of baking can conjure up an image of your kitchen.
Mapping those connections could give us the best chance yet at discovering how brains work, how memories are stored and accessed, and perhaps even changing them without causing irreparable damage. But it's a daunting task. Your neurons are tangled together tighter than headphone cables after a week in your desk drawer, and it's thought that there are millions of miles that need unravelling -- as task far behind current computational capabilities.
"How could we even dare to think we might understand this?" asks Seung. "It's the work of generations." So he and his colleagues are starting small, focusing on getting tiny, partial bits of the whole map, colouring them different shades in what is wonderfully called a "Brainbow". Even these small sections could contain valuable data on memories.
A sequence of muscle actions, like playing a series of notes on a piano, should represent itself as a series of connections between a pair of neurons that fire in turn. If those can be located, then the memory of what order the notes should be played in will be found, and could potentially be modified.
But, as someone ending a relationship might say, "people change". And their connectomes change, too. Over time, neurons will lose branches, grow new ones and twist into new and different shapes. Experiences, memories and emotions will shape your connectome, creating new links and causing old ones to die.
"The metaphor I like to use is a 'stream of consciousness'", says Seung. "The connectome is the bed of the stream, and your neural activity is the water. In daily use, the water flows along the stream bed, but over time, it forges new channels and twists the streambed into a new shape, leaving oxbow lakes and derelic,t backwaters behind it."
Seung ended his talk with a brief detour into the world of cryonics, raising the question of whether it would successfully preserve the connectome. There's evidence that the connectome disappears rapidly after death, and if that's true then cryonics would need to be fast, or it'd be impossible to resurrect someone with their memories and personality intact.
The only way to be sure would be to defrost someone, extract their brain, and see if the connectome is intact. Given we still aren't anywhere close to mapping a human connectome yet, it's going to be a while until we can find out.
Photo Credit: Flickr CC: Mike Blogs
Reading the Book of Memory: Sparse Sampling versus Dense Mapping of ConnectomesRead the whole article.
By H. Sebastian Seung
Howard Hughes Medical Institute; Brain and Cognitive Sciences Department, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
Correspondence: seung@mit.edu
Neuron 62, 17-29 (2009)
DOI 10.1016/j.neuron.2009.03.020
Many theories of neural networks assume rules of connection between pairs of neurons that are based on their cell types or functional properties. It is finally becoming feasible to test such pairwise models of connectivity, due to emerging advances in neuroanatomical techniques. One method will be to measure the functional properties of connected pairs of neurons, sparsely sampling pairs from many specimens. Another method will be to find a ‘‘connectome,’’ a dense map of all connections in a single specimen, and infer functional properties of neurons through computational analysis. For the latter method, the most exciting prospect would be to decode the memories that are hypothesized to be stored in connectomes.
In constructing a neural network model of brain function, it is standard to start from a mathematical description of spiking and synaptic transmission, make assumptions about how neurons are connected by synapses and then numerically simulate or analytically derive the activity patterns of the network. Success is declared if the model’s activity patterns reproduce those measured by neurophysiologists.
Initially, the model neurons used in such networks were highly simplified to the point of being naive. But they have become more sophisticated over the years, incorporating findings about intrinsic and synaptic currents in neurons. In contrast, many assumptions about neural connectivity have been used by theorists for decades without revision, because they have been difficult to test empirically.
It has been popular to assume that the connectivity between any pair of neurons is a function of variables associated with the neurons. These variables, which I dub cell labels, are attributes of a neuron that can be measured without determining its connectivity directly. The cell label can include what neuroanatomists call cell type, which is defined classically by shape and location (Bota et al., 2003; Masland, 2004). In the retina, photoreceptors make connections onto horizontal cells, a rule of connectivity based on cell type (Masland, 2001b). A cell label could also include some property that is determined by a neurophysiologist through activity measurements. For example, some models of the primary visual cortex assume that excitatory neurons with similar preferred orientations are connected (Somers et al., 1995; Ben-Yishai et al., 1995), so that the cell label is preferred orientation.
For testing such a pairwise model of neural connectivity, two standard neuroanatomical methods are available. Sparse reconstruction relies on light microscopy and sparse labeling of neurons, and dense reconstruction relies on electron microscopy and dense labeling. Both methods have been problematic. Axons can be less than 100 nm in diameter (Shepherd and Harris, 1998), and dendritic spine necks can be even narrower (Fiala and Harris, 1999). Since 100 nm is less than the wavelength of visible light, these structures cannot be resolved with a light microscope if they are entangled in a densely stained neuropil (but see Hell [2007] for exceptions to this rule). However, one can see a single neuron stained with dye, as long as the surrounding neurons are unstained and hence remain invisible. This trick was employed by Golgi, who invented a stain that marked a sparse subset of neurons in the brain.
Cajal used Golgi’s stain to reconstruct the branching patterns of neurons. If two neurons made contact with each other, Cajal inferred that they were connected. However, he could not rigorously prove this inference, because he could not see synapses. Contact suggests that a connection exists, but a synapse must be identified to prove it. In short, connection = contact + synapse.
In the 1970s, neuroanatomists began to use electron microscopy for dense reconstruction of neurons. In principle, this imaging method has enough spatial resolution to see all of the axons and dendrites in a densely labeled neuropil. It is also possible to identify synapses through telltale markers such as vesicles. Most famously, electron microscopy was used to map every connection in the nervous system of the nematode C. elegans (White et al., 1986). For every synapse between two neurites, the presynaptic and postsynaptic neurons were identified by tracing the neurites back to their parent cell bodies.
Tags:
No comments:
Post a Comment