Can Programs Become Universal Concept Language?

The human brain comes into the forefront with the ability to learn complex concepts from limited sample. Understanding the underlying mechanism on the basis of this is a critical goal of artificial intelligence research. A new research suggests that the secret under this success can be programs and shows how artificial intelligence systems can advance concept learning by combining work in this field.

The study published in arXiv emphasizes several basic findings:

• Programs carry structural information: Human knowledge is expressed with more rich structural formats than simple data points; programs offer a strong way to capture this complexity.
• Progressive model research shows that the program-based concept representation opens new doors in machine learning.
• A universal representation language can be able to share concepts of different AI applications consistently.
• Program-based approaches for minimal data learning are the potential to mimic human-like generalization.

Research systematically examines the contributions of existing computational concept learning models. The application of findings can accelerate significant progress in areas such as natural language processing, visual recognition and robotic learning. In particular, developing systems that can gain new skills through a small number of examples is critical to approaching artificial intelligence technology to human.

The creation of a universal concept language is successful, artificial intelligence systems will approach a step further into thinking and learning like human.