Machine Perception and Cognitive Robotics Laboratory

Hyperdimensional Computing

Hyperdimensional Computing encodes information as 10,000-dimensional binary vectors for agent navigation tasks. Unlike neural networks, it requires no gradient descent — learning happens through vector binding and bundling operations inspired by how the brain represents concepts.

Model Card

10,000-dimensional binary vectors. Operations: binding (XOR), bundling (majority vote), permutation. No gradient descent — learning through vector algebra. Applied to agent navigation.
Spatial navigation environments with state-action pairs encoded as hypervectors.
Agents navigate environments using HD computing without any gradient-based training. Comparable accuracy to small neural networks with faster inference.
Dimensionality must be tuned per task. Complex sequential reasoning remains challenging in the HD framework.

Team

William Edward Hahn, Ph.D.

Founder and Co-Principal Investigator (Co-PI)
Associate Professor

[email protected]

Machine Perception and Cognitive Robotics Laboratory