Compositional generalization is the new method for machine learning.

  Compositional generalization is the new method for machine learning.

Compositional generalization means the ability to create new entities using existent particles. The idea is like making music by using the same notes but changing words. In the same way, if the system knows the words "walking" and "two times" it's possible to connect those terms and make the sentence "walking two times". This is the idea of compositional generalization.

The compositional generalization can extend to physical works. The system can connect certain movements with certain objects and orders that users give to robots. But the key element in that kind of robot is language modeling. The robot must "understand" the things that people want it to do when they say something. The compositional generalization means that a robot can connect some words that it hears to certain movement series.

"Researchers have developed a technique called Meta-learning for Compositionality (MLC) that enhances the ability of artificial intelligence systems to make “compositional generalizations.” This ability, which allows humans to relate and combine concepts, has been a debated topic in the AI field for decades. Through a unique learning procedure, MLC showed performance comparable to, and at times surpassing, human capabilities in experiments. This breakthrough suggests that traditional neural networks can indeed be trained to mimic human-like systematic generalization." (ScitechDaily.com/The Future of Machine Learning: A New Breakthrough Technique)

In a physical model, the system can learn by looking at things. When a robot or learning machine learns to walk, it can see how people walk. In this example, a robot looks like a human. Then it must combine things that it sees as the right things. The system must know what limbs are and what parts it must use for moving things, like the left arm.

The servo engines that move joints must be equipped with microchips. The system must know what is on the left side. And then it must know where that servo engine is. When operators teach which side is left, they can raise the left hand of the robot. Then they must make the robot connect that moving hand to the word "left side".

The microchips that are controlling the left side might have small computers that know if it needs to move its servo engine. In that case, the computer "brain" of robots can send commands by using a common address or common transportation address. In that case, the computers that control servo engines filter out unnecessary messages from the command line that the system gives to those servo engines.

Because robots use networked systems where intelligent systems control the joints, another neural structure is simpler to make. And that also makes operations lighter for the system's brain, which is the computer. In networked systems, there are multiple secondary computers in the robot's body. And the robot group can form a whole where each member shares their information and computing capacity.

https://scitechdaily.com/the-future-of-machine-learning-a-new-breakthrough-technique/