Turn abandoned patents
into your own filing
into your own filing
This invention describes a method for one computer to efficiently share information about a neural network model with another computer. Instead of sending the entire model, it communicates specific details about its structure or parameters. Crucially, the model is broken down into parts unique to it and parts shared with other models, and the information sent distinguishes between these dedicated and common layers. The receiving computer then uses this partial information to understand or reconstruct the model.
Why it matters: Since 2023, the scale and prevalence of large neural networks and foundation models have significantly increased. This invention's focus on efficiently communicating shared and dedicated network layers is now even more relevant for deploying and updating these complex, often modular, AI systems.
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
