Find a Question:
Discovery larger brain capacity helps research neural networks
Researchers think they can improve neural networks thanks to a discovery that brains have a much greater potential capacity than previously thought. The discovery could help fuel efficient, but to develop specific neural networks.
The discovery that brains than thought until now ten times more capacity comes from a study of synaptic connections in the hippocampus of rat brains. That area plays an important role in, inter alia, to translate the short-term to long-term memory. Investigators used a virtual 3D reconstruction of a few cubic microns, and analyzed the sub-structures of the neurons. They discovered that there is much more variation in the sizes of synapsverbindingen is than previously thought.
The dimensions of the synaptic connection which neurons communicate with one another, together with their number a measure of signal strength: larger synapses give a stronger signal and synaptic connections become stronger through successful communication. In this way, the “storage capacity” of a neuron and of brain structures are estimated on the basis of the number of synapses and their relative sizes or signal strength. Researchers found in their 3D reconstructions no synapsafmetingen handful, but at least 26 categories. Thus, the number of corresponding potential “bits” of information from one or two to approximately 4.7 bits. The potential memory capacity is thus according to researchers at the Salk institute an order of magnitude more extensive than imagined. Extrapolating to people brains about a petabyte capacity would amount to ten times more than previously thought.
With the discovery, the researchers can better understand how brains work and how they can thrive on so little energy. The precision would be due to the averaging of signals over time and varying the size of the synapses under the influence of signals. Thus, a single abnormal signal can have a significant impact and be successful signals getting stronger. Such force could be applied in computers in neural networks in order to achieve better deep learning. Neural networks could then promote desired signals and ignore unwanted signals, so learning gives better results. In addition, the technique could produce more fuel-efficient neural networks.Viewing:-127
Answer this Question
You must be Logged In to post an Answer.
Not a member yet? Sign Up Now »
Star Points Scale
Earn points for Asking and Answering Questions!