The latest articles on DEV Community by Jonathan West (@unrealjon). https://dev.to/unrealjon https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3760487%2F91a28204-6100-4bf2-b5d5-2266628fcca4.jpg https://dev.to/unrealjon en Jonathan West Fri, 13 Feb 2026 13:36:54 +0000 https://dev.to/unrealjon/distributed-transform-domain-representation-for-persistent-storage-of-semantically-encoded-28d2 https://dev.to/unrealjon/distributed-transform-domain-representation-for-persistent-storage-of-semantically-encoded-28d2 <p>I’ve been experimenting with a representation I call DTDR (Distributed Transform-Domain Representation).The idea is to store vectors or model parameters as quantised coefficients of a structured orthogonal transform and treat that representation as the persistent form rather than a preprocessing step.<br> Some interesting behaviours seem to emerge:</p> <ul> <li>- similarity relationships are often still usable directly in coefficient space</li> <li>- approximate nearest-neighbour search sometimes improves due to coefficient dilution</li> <li>- quantised representations can be reconstructed into standard FP tensors for normal inference</li> <li>The stored form often remains further losslessly compressible</li> </ul> <p>I’m trying to understand whether this is just another way of viewing existing quantisation techniques, or whether it has genuinely different implications for storage/search pipelines.</p> <p>Constructive critique very welcome — especially pointers to prior work I may have missed.<br> Code and demos: <a href="https://github.com/UnrealJon/" rel="noopener noreferrer">https://github.com/UnrealJon/</a> </p> ai machinelearning similarity vectordatabase