Learning concepts with energy functions
OpenAI Blog
Researchers created an energy-based model that learns spatial concepts like "near" and "between" from minimal examples. The model achieved concept learning with only five demonstrations per concept and successfully transferred learned concepts from 2D particle simulations to 3D robot control tasks. This approach enables quick adaptation of conceptual understanding across different environments without retraining from scratch.
Why it matters
We’ve developed an energy-based model that can quickly learn to identify and generate instances of concepts, such as near, above, between, closest, and furthest, expressed as sets of 2d points. Our model learns these concepts after only five demonstrations. We also show cross-domain transfer: we use concepts learned in a 2d particle environment to solve tasks on a 3-dimensional physics-based robot.
