"Vector Symbolic Architectures In Clojure"
summary
Demonstrates building Clojure data structures as single hyperdimensional vectors using MAP (Multiply-Add-Permute) algebra, with vectors sized 10,000–1,000,000 dimensions and initialized ±1. Uses elementwise bundling (clipped add), binding (elementwise multiply as its own inverse), permutation (rotate) and a cleanup-memory atom to recover keys and values. Implements get, assoc and indexed sequence semantics, runs in the REPL on a laptop via d-type.next, and shows noise-tolerant fuzzy queries and working examples of maps, sequences and compound queries.
AI-written summary · about
chapters (5)
- 0:00Start
- 3:39Blessing of High Dimensionality
- 5:46Hyperdimensional Vector Avantages
- 7:04Vector Symbolic Architectures
- 14:04VSA Example
related talks
43mCapacity Analysis of Vector Symbolic Architectures
38mScientific Clojure, a bird's eye view
39mWhat's in a Vector Database?
34mFunctional Vectors, Maps, and Sets in Julia
01h 23mVector Databases Demystified: Discover how they work
42mClojure is Not Afraid of the GPU
35mImpact of Vector Search
46mA fun and absurd introduction to Vector Databases