Researchers at University of Technology Sydney (UTS) have developed a new artificial intelligence (AI) algorithm named Torque Clustering and claimed that is much closer to natural intelligence than current methods.
This method allows AI to learn and find patterns in data without human help. Torque Clustering can analyze huge amounts of data in fields like biology, chemistry, and finance, revealing new insights about diseases, fraud, or behavior.
Many AI systems use supervised learning, which needs human-labeled data. But Torque Clustering uses unsupervised learning that learns from the environment like animals do. Supervised learning is expensive and slow because it requires data labeling, but unsupervised learning can handle complex tasks without it.
Potential for robotics and AGI
The researchers have described the Torque Clustering method in a paper published in IEEE Transactions on Pattern Analysis and Machine Intelligence. What makes it unique is its use of the physics concept of torque to find data clusters automatically. Clusters are groups of data points that are more similar to each other than to those in other groups. This approach allows the Torque Clustering method to adapt to various data types.
The researchers say that Torque Clustering outperforms other unsupervised methods. It's autonomous, doesn't need settings adjusted, and is highly efficient with large datasets. Tests show it scores 97.7% on average for clustering, far better than the 80% from other methods.
Torque Clustering is inspired by the torque balance in gravitational interactions when galaxies merge. The researchers suggest that like last year's Nobel Prize in physics for neural networks, this unsupervised learning method could be groundbreaking.
According to the researchers, the Torque Clustering method could advance AI toward Artificial General Intelligence (AGI), especially in robotics, by improving movement and decision-making. The researchers think Torque Clustering could change how we approach unsupervised learning, moving towards more autonomous AI. The code for this method is now available for researchers to use.