How robot baby learns meaningful representations

摘要

The representations we provide to machines are meaningful but the meanings are exogenous, they live in our heads, not in the minds of machines. The Robot Baby project develops algorithms for robots to learn representations that are meaningful to them. As meanings must ground out in the time-series data provided to the robot from its sensors, the talk will ask what kinds of meanings can be learned by a robot and whether they are sufficient for learning human language.