Hyperdimensional Computing Labs develops software and methodologies that apply hyperdimensional computing (HDC) to encode connected, heterogeneous data as high-dimensional hypervectors for associative search and online learning. The company helps organizations build inspectable, adaptable AI systems that learn from small numbers of examples and preserve explicit data structure. Their approach is demonstrated through practical use cases like finding similar past sales orders or enabling AI memory for warehouse robots.
Funding
Funding not disclosed
Founders
Product
Problem
Today's AI systems are powerful, but their learned representations can be difficult to inspect and expensive to adapt. This limits their usefulness in applications that require transparent reasoning, rapid updates, or the ability to learn from small amounts of new data.
Solution
Hyperdimensional Computing Labs provides a different approach to representing and working with data by encoding information as very high-dimensional hypervectors, typically 10,000 dimensions, that can be composed, compared, and updated using simple algebra. Instead of starting with a learned embedding for each whole item, the company builds representations from the structure already present in the data, including roles, values, and relationships. These representations share one hypervector space, enabling associative search, online learning from a small number of examples, and preservation of explicit structure. The company benchmarks its HDC representations against conventional retrieval and machine learning methods on real data to understand where HDC helps and where it doesn't.
Target Audience
Primary customers are organizations and engineering teams working with connected or heterogeneous data, associative search, agent memory, and systems that need to learn as new examples arrive, including those in data-intensive industries seeking more transparent and adaptable AI solutions.
Features
- Encodes strings, numbers, and lists into hypervectors using bind, bundle, and permute operations
- Supports associative search by composing partial queries and resolving similarity matches back to exact records
- Enables online learning where prototypes are updated incrementally as new samples arrive
- Preserves explicit data structure including node properties and edge direction in graph data
- Provides reproducible experiments and research on hypervector properties, such as similarity structure preservation between HRR and MAP approaches
- Includes practical use case demonstrations, such as finding connected patterns in delayed sales orders and AI memory for warehouse robots