Optical disorder generates high-dimensional vectors for computing and inference
Hypervectors from Optical Disorder: Programmable Encoding and Optical Inference for Hyperdimensional Computing
Emerging Technologies
Summary
Hyperdimensional computing uses very large random-looking vectors to represent and process information efficiently. The authors show a way to create these vectors physically using a special medium that scatters light unpredictably, combined with a silicon chip. This approach generates complex, high-dimensional vectors from only a few inputs quickly and without storing large amounts of random data. They also demonstrate how these vectors can be used for making decisions by comparing them optically, potentially speeding up certain computations.
What this means in practice
- •For photonics engineers: Create high-dimensional vectors directly in optical hardware for fast, energy-efficient computing and memory tasks without heavy digital processing.
- •For embedded systems designers: Implement optical inference modules that use physically generated vectors to reduce memory and computation overhead in resource-limited devices.
Authors
Takuya Iwata, Namthip Srisuthep, Satoshi Sunada
Abstract
Hyperdimensional computing (HDC) is a computing framework that represents information as high-dimensional pseudorandom vectors called hypervectors (HVs), enabling learning and inference through simple algebraic operations. The HV dimensionality provides computational capacity and error robustness, but the required high-dimensional randomness must be stored or regenerated, imposing a memory--computation tradeoff. Here, we address this tradeoff by physically embodying the randomness required for HV generation in the static disorder of a scattering medium. Specifically, a silicon photonic circuit combined with the scattering medium generates high-dimensional HVs from a small number of input values in a single optical shot. Detector-side encoding programs HV correlations for continuous and categorical representations. The resulting HVs reproduce key statistical and compositional properties of ideal i.i.d.\ random HVs, including pairwise similarity statistics, associative-memory capacity, and factorization capacity. We further demonstrate an optically assisted inference path using the generated HVs and optical similarity evaluation.