Science News Hubb
Advertisement
  • Home
  • Science News
  • Technology
  • Contact us
No Result
View All Result
  • Home
  • Science News
  • Technology
  • Contact us
No Result
View All Result
Science News Hubb
No Result
View All Result
Home Science News

AI uses artificial sleep to learn new task without forgetting the last

admin by admin
November 10, 2022
in Science News


Many AIs can only become good at one task, forgetting everything they know if they learn another. A form of artificial sleep could help stop this from happening



Technology



10 November 2022

By Jeremy Hsu

AIs may need to sleep too

Shutterstock/Ground Picture

Artificial intelligence can learn and remember how to do multiple tasks by mimicking the way sleep helps us cement what we learned during waking hours.

“There is a huge trend now to bring ideas from neuroscience and biology to improve existing machine learning – and sleep is one of them” says Maxim Bazhenov at the University of California, San Diego.

Many AIs can only master one set of well-defined tasks –  they can’t acquire additional knowledge later on without losing everything they had previously learned. “The issue pops up if you want to develop systems which are capable of so-called lifelong learning,” says Pavel Sanda at the Czech Academy of Sciences in the Czech Republic. Lifelong learning is how humans accumulate  knowledge to adapt to and solve future challenges.

Bazhenov, Sanda and their colleagues trained a spiking neural network – a connected grid of artificial neurons resembling the human brain’s structure – to learn two different tasks without overwriting connections learned from the first task. They accomplished this by interspersing focused training periods with sleep-like periods.

The researchers simulated sleep in the neural network by activating the network’s artificial neurons in a noisy pattern. They also ensured that the sleep-inspired noise roughly matched the pattern of neuron firing during the training sessions – a way of replaying and strengthening the connections learned from both tasks.

The team first tried training the neural network on the first task, followed by the second task, and then finally adding a sleep period at the end. But they quickly realised that this sequence still erased the neural network connections learned from the first task.

Instead, follow-up experiments showed that it was important to “have rapidly alternating sessions of training and sleep” while the AI was learning the second task, says Erik Delanois at the University of California, San Diego. This helped consolidate the connections from the first task that would have otherwise been forgotten.

Experiments showed how a spiking neural network trained in this way could enable an AI agent to learn two different foraging patterns in searching for simulated food particles while avoiding poisonous particles.

“Such a network will have the ability to combine consecutively learned knowledge in smart ways, and apply this learning to novel situations – just like animals and humans do,” says Hava Siegelmann at the University of Massachusetts Amherst.

Spiking neural networks, with their complex, biologically-inspired design, haven’t yet proven practical for widespread use because it’s difficult to train them, says Siegelmann. The next big steps for showing this method’s usefulness would require demonstrations with more complex tasks on the artificial neural networks commonly used by tech companies.

One advantage for spiking neural networks is that they are more energy-efficient than other neural networks. “I think over the next decade or so there will be kind of a big impetus for a transition to more spiking network technology instead,” says Ryan Golden at the University of California, San Diego. “It’s good to figure those things out early on.”

Journal reference: PLOS Computational Biology, DOI: 10.1371/journal.pcbi.1010628

More on these topics:



Source link

Tags: artificial intelligencebrainsleep
Previous Post

‘Fuzzy’ dark matter might make stars form in giant ‘pancakes’

Next Post

Some populations of large ocean fish are recovering, but not sharks

Next Post

Some populations of large ocean fish are recovering, but not sharks

Recommended

Fast, Deep Cuts in Emissions Are Needed to Avoid ‘Climate Time Bomb’

March 21, 2023

Research Summaries Written by AI Fool Scientists

January 14, 2023

Don't miss it

Science News

Microplastics are in our bodies. Here’s why we don’t know the health risks

March 25, 2023
Science News

Electronic wound dressing releases drugs to help injuries heal

March 25, 2023
Science News

See the moon turn blood red in this stunning lunar eclipse video

March 25, 2023
Technology

IPCC Report Will Likely Shake Up U.N. Climate Talks

March 25, 2023
Technology

Your Guide to Staying Safe While Visiting National Parks

March 25, 2023
Technology

Mimicking Tissue Mechanics for Better Wound Healing Models

March 25, 2023

© 2022 Science News Hubb All rights reserved.

Use of these names, logos, and brands does not imply endorsement unless specified. By using this site, you agree to the Privacy Policy and Terms & Conditions.

Navigate Site

  • Home
  • Science News
  • Technology
  • Contact us

Newsletter Sign Up

No Result
View All Result
  • Home
  • Science News
  • Technology
  • Contact us

© 2022 Science News Hubb All rights reserved.