The human mind may be quite complex at times. That’s when we resort to meditation, taking deep breaths to distract ourselves from the chaos and confusion. It gives us a sense of satisfaction by making us more understanding and empathetic, and sometimes just by making us do nothing.
What if robotic systems could meditate or do nothing for a day or two in order to learn more effectively?
As silly as it may sound, a group of academics has uncovered how artificial neural networks may imitate human brain sleep patterns, increasing their applicability across a wide range of study topics. Furthermore, this may aid networks in reducing the risk of catastrophic forgetfulness later.
Allow machines to sleep for a few hours!
The human body requires at least 7 to 13 hours of sleep every day on average. While sleeping, the heart rate, metabolism, respiration, and hormone levels all fluctuate while the body remains in a relaxed condition. However, the brain stays nearly unchanged.
“The brain is quite busy while we sleep, repeating what we have learned throughout the day,” explains Maxim Bazhenov, professor of medicine and sleep researcher at the University of California San Diego School of Medicine. He goes on to say that sleep helps reorganize memories in order to display them in the most effective way possible.
Bazhenov and his colleagues previously showed how sleep improves the ability to recall indirect or arbitrary relationships between items or people, builds logical memory, and protects against forgetting old memories. Artificial neural networks have used the design of the human brain to improve several technology systems ranging from fundamental research to economics and health.
Many neural networks have attained superhuman capability, the most recent example being the recent increase in processing speed. However, they fall short in one crucial area. When new information is learned sequentially, artificial neural networks tend to overwrite prior knowledge. The behavior is sometimes referred to as catastrophic forgetting. According to Bazhenov, the human brain learns continuously, assimilating new information into current knowledge. He says that when fresh instruction is mixed with intervals of sleep, the brain learns the best for memory formation.
Instead of continually communicating information, the scientists employed increasing neural networks to simulate genuine brain systems, which are sent as discrete events at certain time points. The scientists observed that the network of spiking brain systems minimized catastrophic amnesia when it was taught with periodic off-line intervals that mimicked sleep. ‘Sleeping,’ like the human brain, enabled the networks to repeat previous memories without utilizing any old training data.
Augmenting sleep pattern
The human brain has thousands of memories, which are represented by synaptic weight patterns. Each pattern demonstrates the strength of a connection between two neurons. The study shows that while learning new information, these neurons fire in a certain order, which increases the number of synapses between them. The spiking patterns generated during a human mind’s waking state are reproduced numerous times in a spontaneous way when asleep, which is sometimes referred to as ‘replay’ or ‘reactivation.’
“Synaptic plasticity, the ability to be altered or molded, is still present during sleep and it can further increase synaptic weight patterns that indicate memory, helping to avoid forgetting or to promote the transfer of information from old to new activities,” adds Bazhenov.
When the scientists applied this method to artificial neural networks, they discovered that it helped the networks avoid catastrophic forgetting. That is, the networks might learn in the same way that people or animals do. The study was a fresh advance in understanding how the brain processes specific information during sleep, just to aid boost human memory. Moreover, improving sleep patterns can lead to improved memory. The team has also created computer models to generate ideal techniques for applying stimulation to sleep components such as sensory tones, which increase learning and improve sleep patterns. They also say that the initiative becomes critical when patients suffer from non-optimal memory in illnesses such as Alzheimer’s.
Moreover, Google DeepMind launched a framework to enable the building of AI bots capable of understanding human commands to do certain behaviors.
Several existing AI frameworks have been chastised for failing to consider the situational understanding of how humans use language. DALL.E 2, which is a generative AI comments on how it failed to comprehend the grammar of the text questions. For an input text of ‘a spoon on a cup,’ the results would include photos from the dataset including a cup and a spoon—without understanding the connection between the items supplied in the text prompt.
Author- Toshank Bhardwaj, AI Content Creator
