In the United States, researchers have found a new way to make the neural networks used by artificial intelligence more efficient. By integrating working memory across the network, synapses can be updated in real time, greatly reducing the number of data processing operations.
Artificial intelligence (AI) consumes a lot of energy, especially during the training phase. This is a problem of this magnitudeElon MuskElon Musk Predict that the world will run out of electricity for artificial intelligence within a year. But researchers from Cold Spring Harbor Laboratory Perhaps in the United States they found the solution, inspired by it brainbrain Humans.
“Impressive as it is ChatGPTChatGPT And all current artificial intelligence technologies in terms of interaction with the world physicalphysicalIt is still very limited. Even in areas like mathematical problem solving and essay writing, they need to practice billions and billions of examples before they can solve them well. “, explained Kyle Daruwalla, lead author of the article published in the journal Frontiers in computational neuroscience.
Working memory allows synapses to be updated
For him, the problem is data movement. Artificial neural networks consist of billions of connections, and the data must pass through entire circuits before the model is updated. Using this new approach, each artificial neuron receives feedback through an auxiliary memory network and can be modified in real time.
This creates a direct connection between Working memoryWorking memory and synaptic updates, providing new evidence for an unproven theory in neuroscience linking working memory to learning and academic performance. This new approach can therefore significantly reduce computing power and thusenergyenergy Necessary for artificial intelligence to work.
“Music guru. Incurable web practitioner. Thinker. Lifelong zombie junkie. Tv buff. Typical organizer. Evil beer scholar.”
More Stories
After the discovery of norovirus, these berries should not be eaten.
Mechanics Strike | WestJet Cancels Nearly 700 Flights, Affects Nearly 100,000 Passengers
Three 'basic' Airbnb listings: Owner shares how he easily skirted the rules