The new AI technique, which is based on a structure dating back to the 18th century, is more transparent and easier to ethically use, particularly in medical applications where complex deep learning techniques can risk patient safety. The researchers believe their findings, which challenge current perceptions of deep learning, will make the use of neural networks in various applications more reliable.
Key takeaways:
- Researchers from the University of Jyväskylä have simplified deep learning, a popular artificial intelligence technique, using 18th-century mathematics and found that traditional training algorithms work better than recent ones.
- The new model is more expressive, easier to maintain, and can reliably reduce large datasets while preserving necessary information, making the use of neural networks in various applications easier and more reliable.
- The simpler AI models are more transparent, easier to ethically use, and are more environmentally friendly as they save computational resources and use significantly less energy, contributing to the development of green IT.
- Despite the resistance from the scientific and business community due to the prominent role of deep learning, the researchers believe their new AI has a range of applications in nanotechnology, digital learning environments, and medical technology.