Home AI News Groundbreaking Neural Network Learns and Remembers ‘On the Fly’, Mimicking the Brain

Groundbreaking Neural Network Learns and Remembers ‘On the Fly’, Mimicking the Brain

Groundbreaking Neural Network Learns and Remembers ‘On the Fly’, Mimicking the Brain

A Breakthrough in Neural Network Learning

How Nanowire Networks Work

Advantages of Online Learning

Researchers at the University of Sydney and University of California at Los Angeles have achieved a groundbreaking development in the field of artificial intelligence (AI). In a study published in Nature Communications, a physical neural network was shown to learn and remember information in a way similar to the brain’s neurons.

Nanowire networks, consisting of tiny wires arranged in patterns similar to neural networks, were used to perform memory and learning tasks. These networks functioned by responding to changes in electronic resistance at junctions where the nanowires overlapped. The electrical inputs caused changes in conductivity, much like synapses in the human brain.

Efficient and Low-Energy Machine Intelligence

The significance of this breakthrough lies in its potential to develop efficient and low-energy machine intelligence for complex real-world tasks. Lead author Ruomin Zhu, a PhD student, explained that the findings demonstrate how brain-inspired learning can be applied to process dynamic data.

Unleashing the Potential of Online Learning

One of the advantages of this new approach is its ability to process information online, avoiding heavy memory and energy usage. Traditional methods store data in memory and then train a machine learning model using that stored information. However, the new nanowire neural network can learn and remember online, sample by sample, extracting data in real-time.

This online learning capability is crucial for scenarios where data is continuously changing, such as streaming sensor data. The ability to adapt in real-time sets this neural network apart from artificial neural networks currently in use.

Successful Image Recognition and Memory Retrieval

To evaluate the network’s capabilities, it was tested on an image recognition task using the MNIST database of handwritten digits. The network displayed an impressive 93.4% accuracy in correctly identifying test images. Additionally, it was able to remember sequences of up to eight digits, simulating the memory task of recalling a phone number.

The success of this research paves the way for advancements in machine learning and memory functions. With further development and optimization, this neural network can revolutionize AI technology and lead to more efficient and low-energy solutions for real-world learning and memory tasks.

Source link


Please enter your comment!
Please enter your name here