/industry-wired/media/agency_attachments/2024/12/04/2024-12-04t130344212z-iw-new.png)
/industry-wired/media/agency_attachments/2024/12/04/2024-12-04t130332454z-iw-new.jpg)
/industry-wired/media/post_attachments/wp-content/uploads/2022/02/Scientists-are-Using-Pseudo-Rehearsal-to-Improve-the-Learning-Capacity-of-AI.jpg)
Scientists across the globe use Pseudo Rehearsal to solve the Catastrophic forgetting problem.
As we all know Artificial Intelligence machines try to mimic the natural cognitive process. However, it is still not easy to behave like the human brain. Do you think that these technological advancements will make machines remember everything? Eventually, no. Machines do forget things and it even has the name Catastrophic Forgetting Problem. To get over it, scientists are using Pseudo Rehearsal Technique.
According to Science Daily, it is said that computer chips will precisely rewire themselves to take in new data like the human brain does which will help AI to keep learning over time because companies use more data to improve AI as it recognizes images, languages, and other complex tasks. It is also said that when artificial intelligence learns something new it tends to forget the data it already learned.
Shriram Ramanathan professor at Purdue University School of Materials Engineering created an artificial platform for machines to learn throughout their lifetime. Usually, the circuits used by a machine for ages and the circuits made for the machine in a factory aren't different.
To make AI more portable for the sake of autonomous vehicles, robots making which make decisions, forgetting already learned data is a huge complication. The machines would operate more effectively and efficiently if AI were embedded into hardware.
Building A Brain Chip
Perovskite Nickelate is a device that has more hydrogen, serves as a synapse connecting between neurons, that the brain uses to store memory in complex neural circuits. The neural networks use reservoir computing stated by Purdue's Team which explains how different parts of a brain communicate and transfer information. The researchers stated that a dynamic network can pick and choose which circuits are best fit for addressing those problems.
Ramanathan believes that semiconductor-compatible techniques can be adopted by the semiconductor industry. To build a brain-inspired computer, researchers are working to demonstrate these concepts on large-scale test chips.
The catastrophic forgetting problem can be resolved by providing samples from previous data. The data will be used by the machine to retain the knowledge it has gained and this technique is known as “Pseudo rehearsal”.