Learning Programmatically Structured Representations with Perceptor Gradients
This work addresses the challenge of interpretable and transferable representation learning in AI, though it appears incremental as it builds on existing symbolic and neural network methods.
The paper tackles the problem of learning symbolic representations from raw observations by introducing the perceptor gradients algorithm, which decomposes an agent's policy into a perceptor network and a task encoding program, enabling efficient learning of transferable symbolic representations and generation of new observations based on semantic specifications.
We present the perceptor gradients algorithm -- a novel approach to learning symbolic representations based on the idea of decomposing an agent's policy into i) a perceptor network extracting symbols from raw observation data and ii) a task encoding program which maps the input symbols to output actions. We show that the proposed algorithm is able to learn representations that can be directly fed into a Linear-Quadratic Regulator (LQR) or a general purpose A* planner. Our experimental results confirm that the perceptor gradients algorithm is able to efficiently learn transferable symbolic representations as well as generate new observations according to a semantically meaningful specification.