Email
Password
Remember meForgot password?
    Log in with Twitter

article imageMagnets aid AI achieve efficiency of the human brain

By Tim Sandle     Apr 27, 2019 in Science
Scientists have developed a new way to improve the efficiency of artificial intelligence, edging it closer to the capabilities of the human brain. This development involves the use of magnets.
The research, which comes from Purdue University, has shown how new brain-like networks are possible and how these could assist robots approach human-like efficiency in terms of object recognition tasks. This rests on a different type of process from the conventional route taken for artificial intelligence development and this uses magnetics with brain-like networks in order to program and teach devices like personal robots, self-driving cars and drones, to view, understand, generalize, and respond in more effective ways about different objects.
The application of magnetism is a further step in an emergent field of research whereby neurocomputers that attempt to mimic the human brain by nanoelectronic components are being developed. The objective is to harness the efficiency of machines in recognition problems. The basis of the research is based on switching dynamics of a nano-magnet, since these are said to be very much like dynamics of neurons. This is called stochastic switching, and this is similar to switching within a neuron.
Discussing the new approach, lead researcher Professor Kaushik Roy states: "Our stochastic neural networks try to mimic certain activities of the human brain and compute through a connection of neurons and synapses."
The academic adds: "This allows the computer brain to not only store information but also to generalize well about objects and then make inferences to perform better at distinguishing between objects."
An example of how this type of technology might be applied is with training a robotic shopper, through machine learning, so that it can easily tell the difference between the thousands of products in the store. This application could be used within warehouses or even within the retail store of the future.
The research was presented to the April 2019 annual German Physical Sciences Conference.
READ MORE: Walmart unveils its 'store of the future' to test out new tech
The research is published in the journal Frontiers in Neuroscience, with the peer reviewed study titled "ReStoCNet: Residual Stochastic Binary Convolutional Spiking Neural Network for Memory-Efficient Neuromorphic Computing."
More about Artificial intelligence, Human brain, machine learning
More news from
Latest News
Top News