This research introduces CATS Net, a novel dual-module neural network designed to mimic how the human brain forms and communicates abstract concepts. The system utilizes a concept-abstraction module to compress complex sensory data into low-dimensional vectors, while a task-solving module applies these concepts to make visual judgments. Experiments demonstrate that these artificial concept spaces naturally align with human neurocognitive models and neural activity patterns in the ventral occipitotemporal cortex. Furthermore, the framework enables knowledge transfer between independent networks through a shared symbolic interface, mirroring human communication. By bridging the gap between raw sensorimotor experience and symbolic thought, the study provides a computational basis for understanding conceptual cognition. These findings suggest that both biological and artificial intelligences may converge toward similar semantic organizational principles when solving complex tasks.

References:

• Guo L, Chen H, Chen Y, et al. A neural network for modeling human concept formation, understanding and communication[J]. Nature Computational Science, 2026: 1-15.