This invention describes a two-stage automated process for designing neural networks, which are computer systems inspired by the human brain. It first builds a basic network from pre-defined building blocks and refines it, then further optimizes these blocks to construct a final, high-performing network. The method specifically organizes potential network operations into clusters of similar types and selects from these clusters to build the network.
Why it matters: Filed before the explosion of complex AI models. This method's structured search and two-stage optimization could offer efficiency for designing specialized networks.
AI gives you a few directions you could take this. Pick one, and we check whether your version is different enough to patent, then write the filing.
Reinvent this with AI