Binary Space Partitioning as Intrinsic Reward
This addresses data dimensionality reduction for autonomous robots, but appears incremental as it builds on existing concepts like intrinsic reward and hierarchical feature representation.
The paper tackles the problem of high-dimensional sensory data in autonomous humanoid robots by proposing an unsupervised method using binary space partitioning to extract features and compute information gain as intrinsic reward for reinforcement learning.
An autonomous agent embodied in a humanoid robot, in order to learn from the overwhelming flow of raw and noisy sensory, has to effectively reduce the high spatial-temporal data dimensionality. In this paper we propose a novel method of unsupervised feature extraction and selection with binary space partitioning, followed by a computation of information gain that is interpreted as intrinsic reward, then applied as immediate-reward signal for the reinforcement-learning. The space partitioning is executed by tiny codelets running on a simulated Turing Machine. The features are represented by concept nodes arranged in a hierarchy, in which those of a lower level become the input vectors of a higher level.