Reproduce “Optimal quadratic binding for relational reasoning in vector symbolic neural architectures” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Naoki Hiratani, Haim Sompolinsky

The authors study octonion binding to retrieve object–place relations from a vector. They sum codes for each pair, then query the sum to recover one side. For standard Gaussian inputs, one-pair amplitude-preserving normalized mean-squared error is 1/4, versus 1+2/N for HRR. The advantage is strongest at low loads; at high loads, random binding can perform about as well.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.