Posts tagged neural networks
Articles and news from the latest research reports.
Low-Power Chips to Model a Billion Neurons
It’s a little sobering, actually. The average human brain packs a hundred billion or so neurons—connected by a quadrillion (1015) constantly changing synapses—into a space the size of a cantaloupe. It consumes a paltry 20 watts, much less than a typical incandescent lightbulb. But simulating this mess of wetware with traditional digital circuits would require a supercomputer that’s a good 1000 times as powerful as the best ones we have available today. And we’d need the output of an entire nuclear power plant to run it.
Fortunately, we don’t have to rely on traditional, power-hungry computers to get us there. Scattered around the world are at least half a dozen projects dedicated to building brain models using specialized analog circuits. Unlike the digital circuits in traditional computers, which could take weeks or even months to model a single second of brain operation, these analog circuits can model brain activity as fast as or even faster than it really occurs, and they consume a fraction of the power. But analog chips do have one serious drawback—they aren’t very programmable. The equations used to model the brain in an analog circuit are physically hardwired in a way that affects every detail of the design, right down to the placement of every analog adder and multiplier. This makes it hard to overhaul the model, something we’d have to do again and again because we still don’t know what level of biological detail we’ll need in order to mimic the way brains behave.
To help things along, my colleagues and I are building something a bit different: the first low-power, large-scale digital model of the brain. Dubbed SpiNNaker, for Spiking Neural Network Architecture, our machine looks a lot like a conventional parallel computer, but it boasts some significant changes to the way chips communicate. We expect it will let us model brain activity with speeds matching those of biological systems but with all the flexibility of a supercomputer.
Another team, led by Dharmendra Modha at IBM Almaden Research Center, in San Jose, Calif., works on supercomputer models of the cortex, the outer, information-processing layer of the brain, using simpler neuron models. In 2009, team members at IBM and Lawrence Livermore National Laboratory showed they could simulate the activity of 900 million neurons connected by 9 trillion synapses, more than are in a cat’s cortex. But as has been the case for all such models, its simulations were quite slow. The computer needed many minutes to model a second’s worth of brain activity.
One way to speed things up is by using custom-made analog circuits that directly mimic the operation of the brain. Traditional analog circuits—like the chips being developed by the BrainScaleS project at the Kirchhoff Institute for Physics, in Heidelberg, Germany—can run 10 000 times as fast as the corresponding parts of the brain. They’re also fabulously energy efficient. A digital logic circuit may need thousands of transistors to perform a multiplication, but analog circuits need only a few. When you break it down to the level of modeling the transmission of a single neural signal, these circuits consume about 0.001 percent as much energy as a supercomputer would need to perform the same task. Considering you’d need to perform that operation 10 quadrillion times a second, that translates into some significant energy savings. While a whole brain model built using today’s digital technology could easily consume more than US $10 billion a year in electricity, the power bill for a similar-scale analog system would likely come to less than $1 million.
A group of researchers has developed some exciting new techniques for imaging neuronal and synaptic networks using the hard synchrotron x-rays provided by the U.S. Department of Energy Office of Science’s Advanced Photon Source (APS).
These techniques provide images with unprecedented detail and resolution, and open the door to three-dimensional tomographic reconstructions, a vital tool for studying the complex tree-like branching nature of neuronal networks.
Understanding intricate neuronal and synaptic networks, particularly in more complex mammalian brains, requires high-resolution mapping of large volumes of tissue, preferably in three dimensions in order to capture all the subtle structural details.
"Mapping neuron networks has been providing a very significant understanding of how the brain works," said Yeukwang Hwu of Academia Sinica in Taipei, Taiwan, lead author of the paper on this new study, which was published in the Journal of Physics D: Applied Physics.
Connectomics: Mapping the Neural Network Governing Male Roundworm Mating
In a study published today online in Science, researchers at Albert Einstein College of Medicine of Yeshiva University have determined the complete wiring diagram for the part of the nervous system controlling mating in the male roundworm Caenorhabditis elegans, an animal model intensively studied by scientists worldwide.
The study represents a major contribution to the new field of connectomics – the effort to map the myriad neural connections in a brain, brain region or nervous system to find the specific nerve connections responsible for particular behaviors. A long-term goal of connectomics is to map the human “connectome” – all the nerve connections within the human brain.
Because C. elegans is such a tiny animal – adults are one millimeter long and consist of just 959 cells – its simple nervous system totaling 302 neurons make it one of the best animal models for understanding the millions-of-times-more-complex human brain.
The Einstein scientists solved the structure of the male worm’s neural mating circuits by developing software that they used to analyze serial electron micrographs that other scientists had taken of the region. They found that male mating requires 144 neurons – nearly half the worm’s total number – and their paper describes the connections between those 144 neurons and 64 muscles involving some 8,000 synapses. A synapse is the junction at which one neuron (nerve cell) passes an electrical or chemical signal to another neuron.
"Establishing the complete structure of the synaptic network governing mating behavior in the male roundworm has been highly revealing," said Scott Emmons, Ph.D., senior author of the paper and professor in the department of genetics and in the Dominick P. Purpura Department of Neuroscience at Einstein. "We can see that the structure of this network has spatial characteristics that help explain how it exerts neural control over the multi-step decision-making process involved in mating."
In addition to determining how the neurons and muscles are connected, Dr. Emmons and his colleagues for the first time accurately measured the weights of those connections, i.e., an estimate of the strength with which one neuron or muscle communicates with another.
Triangles guide the way for live neural circuits in a dish
July 19, 2012
Korean scientists have used tiny stars, squares and triangles as a toolkit to create live neural circuits in a dish.
They hope the shapes can be used to create a reproducible neural circuit model that could be used for learning and memory studies as well as drug screening applications; the shapes could also be integrated into the latest neural tissue scaffolds to aid the regeneration of neurons at injured sites in the body, such as the spinal cord.
Published today in the Journal of Neural Engineering, the study, by researchers at the Korea Advanced Institute of Science and Technology (KAIST), found that triangles were the most effective shape for helping to facilitate the growth of axons and guide them onto specific paths to form a complete circuit.
Co-author of the study, Professor Yoonkey Nam, said: “Eventually, we want to know if we can design a neural tissue model that biologically mimics some neural circuits in our brain.”
A neuron is an electrically excitable cell that processes and transmits information around the body. The neuron is composed of three main parts: a cell body, or soma, dendrites and an axon, which extends from the soma and links to other cells, creating a network.
When axons grow they are usually guided by proteins. Many researchers have been trying to re-create this key process in a dish by manipulating nerve cells from rat brains.
As nerve cells are usually just a few tens of micrometres in size, the challenge associated with creating a live neural network is firstly positioning cells in desired locations and, secondly, making connections between these cells by guiding the axons in designated directions.
The researchers investigated whether two star shapes, five regular shapes (square, circle, triangle, pentagon and hexagon) and three different sizes of isosceles triangles could guide axons in designated directions. Each shape was the size of a single cell and was replicated to form an array which was printed onto a glass surface.
Each of the arrays had an overall size of 1cm-by-1cm with a gap of 10 micrometres between each shape. Hippocampal neurons were taken from rats and plated onto the patterned surfaces. The neurons were fluorescently labelled with dyes so that images could be taken of their growth.
The researchers found that triangles were the most efficient shape to encourage the growth and guidance of an axon. The key to this was the angles at the points where two of the triangle’s lines meet, also known as the vertices. It was shown that the smaller the vertices, the higher chance the triangle had of inducing growth.
"Based on our results, we are suggesting a new design principle for guiding axons in a dish. We can control the axonal growth in a certain direction by putting a sharp triangle pointing to a certain direction. Then, a neuron that adhered to the triangle will have an axon in the sharp vertex direction.
"Overall, we integrated microtechnology with neurobiology to find a new engineering solution" continued Professor Nam.
Provided by Institute of Physics
Source: medicalxpress.com
Roke Manor Research Ltd (Roke), a Chemring Group company, has developed the world’s first threat monitoring system for autonomous vehicles that emulates a mammal’s conditioned fear-response mechanism.
The STARTLE system uses a combination of artificial neural network and diagnostic expert systems to continually monitor and assess potential threats.
“Startle delivers local autonomy to a vehicle by providing a mechanism for machine situation awareness to efficiently detect and assess potential threats. This allows vehicle sensing and processing resources to be devoted to the assigned task, but if a threat is detected it will cue the other systems to deal with it swiftly before continuing its mission. These vital seconds could be the difference between mission failure and success.”
Source: Neuroscience News