Dreamcoat: Cosine Similarity using Bit Values (Part 1)
7 min readMar 28, 2024
Complex floating point vectors can be (I hope to show) accurately represented by binary values converting each floating point number down to a single binary digit. This allows a single integer operation to simultaneously operate on 32 dimensions (4 bytes). Even with an additional step in calculation, this 32X increase in computational speed is further increased by being integer based operations.