Traditional Mathematics Can Simplify AI Training
The study conducted by researchers at the University of Jyvaskyla in Finland has shown that AI does not necessarily require deep learning, which involves establishing and simulating neural networks based on the mechanisms of the human brain.
The researchers showed that traditional mathematical optimization methods yield better results, the university stated in a press release.
Professor Tommi Karkkainen and Ph.D. researcher Jan Hanninen have used traditional mathematics to simplify the AI learning model.
Deep learning is particularly useful for teaching computers to tackle complex tasks such as generating new content, controlling cars and robots, or playing intricate strategy games. However, deep learning models are complex and difficult to comprehend, the researchers said.
"Our new neural network model is more expressive and can effectively summarize large datasets," said Karkkainen.
Meanwhile, Hanninen said, "Based on our results, applying neural networks to various tasks will become easier and more reliable."
A simpler network structure allows for easier implementation and better understanding. Since artificial intelligence is now part of almost all modern technology, it is crucial to comprehend its operations and functions.
Karkkainen and Hanninen's article was recently published in the prestigious series Neurocomputing.
"We are eager to see how the results will be received within the scientific community and among users of machine learning methods in the industry," Karkkainen said.
4155/v
