Holographic Reduced Representations

Holographic Reduced Representations (HRRs) are a compressed binding technique developed by Tony Plate that approximates the full tensor product of vectors while maintaining fixed dimensionality. They are a key component of tensor-product representations in cognitive science, enabling the encoding of complex symbolic structures in distributed vectors without the exponential dimensionality growth of exact tensor products.

HRRs use circular convolution — equivalent to element-wise multiplication in the frequency domain — to bind vectors together, and approximate inverse operations to unbind them. The technique sacrifices exact structural fidelity for computational tractability, a trade-off characteristic of systems operating under resource constraints. HRRs have been applied to modeling human memory, language processing, and analogical reasoning, and they represent an important bridge between symbolic and connectionist approaches in cognitive science.