Testing the limits of bundling capacity (2)
How many records fit in a 2,048-dimensional hyperspace before unrelated records start to look alike? Our experiments suggest that the answer depends on how many comparisons we make.
Posts by Prashanth Rao
Showing 1–10 of 22 posts
Reviewing TrainableHD: What changes when the encoder learns?
How TrainableHD learns its encoder as well as its classes. We follow a mistake through training, work through an incident example, and examine the costs of deployment.
Testing the limits of bundling capacity (1)
What happens as we pack more and more facts into a bundle? We study the limits of bundling capacity with a 50-item questionnaire.
What do cleanup and crosstalk mean in HDC?
A look back at how the HDC literature describes the terms "cleanup" and "crosstalk", and how we can view both terms through a modern lens using vector search in LanceDB.
Durable memory, transient prototypes
An HDC experiment that separates stored examples from temporary class summaries, and compares rebuilding with caching.
How HDC connects predictions to clinical evidence
Explore how HDC combines medical imaging and clinical language to classify retinal disease and support review of AI predictions.
How context can change a retrieval ranking
Why shared bipolar binding preserves a ranking, how weighted similarity can change it, and what remains to be tested for useful retrieval.
Semantic compression inside wide hypervectors
How a compact abstraction can use thousands of coordinates, and what the extra width buys an HDC system.
Turn graph data into hypervectors for associative search
Learn how to encode node properties and edge direction with HDC, compose a partial query, and resolve similarity matches back to exact graph records.
Your data is connected. Can you find a pattern like this one?
A delayed sales order shows how searching connected patterns could help teams find useful past cases, then check the facts in a graph.
Contact
Have a representation problem in mind?
If you're working with connected data, retrieval, agent memory or online learning, we'd love to hear what you're building and where the current representation is falling short.
Get in touch
We usually reply within two business days.