1. Emergence of machine language: towards symbolic intelligence with neural networks.
- Author
-
Wang, Yuqi, Zhang, Xu-Yao, Liu, Cheng-Lin, Tan, Tieniu, and Zhang, Zhaoxiang
- Subjects
- *
PROGRAMMING languages , *DEEP learning , *ARTIFICIAL neural networks , *CONVOLUTIONAL neural networks - Abstract
The article discusses the emergence of "machine language," a new form of representation that combines deep neural networks with symbolic intelligence. The authors propose three essential properties that machine language should possess: spontaneity, flexibility, and semantics. They simulate the process of language emergence through a collaborative game between two agents, where one describes an image using machine language and the other interprets and draws the image based on the description. The article also compares machine language with discrete language and continuous features in terms of interpretability, generalization, and robustness. The authors argue that machine language, which emphasizes visual information, can lead to significant progress in AI capability. [Extracted from the article]
- Published
- 2024
- Full Text
- View/download PDF