Automated and adaptive design and training of neural networks

This invention describes a method for training computer models, like those used for AI, more effectively. It involves a preliminary learning phase where the model's learning speed is varied, which then helps determine the optimal duration for the main training. The claims specifically narrow this to a computer system that follows a pre-defined schedule, adjusting both the learning speed and how much the model remembers from previous steps across at least two distinct training phases.

Why it matters: Filed before the widespread adoption of large, complex AI models. The need for efficient, adaptive training methods for such models, especially with heterogeneous data, has become paramount since 2022.

Status: Not in forceListed as no longer active. The specific reason is not in the record we hold.
How hard to build: SpecializedSoftware logic for machine learning algorithms.

3 AI reinvention directions for this patent

See all three

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

Where this one came from

Every morning we go through millions of abandoned patents and pick out 15 worth your time. This was one of today’s.

This one

+10