Turn abandoned patents
into your own filing
into your own filing
This invention describes a way to make neural networks run more efficiently on specialized computer chips. It works by taking a key piece of information (a variable) from the network, making multiple copies of it, and sending those copies to different parts of the network to be processed at the same time. The results from these parallel operations are then combined to update the neural network, specifically for parallel training with different data. The claims focus on a method implemented on reconfigurable hardware.
Why it matters: The exponential growth in size and complexity of neural networks, especially large language models, since 2019 has made efficient hardware acceleration and parallel training techniques, like those described, far more critical for practical deployment. Tools and expertise for designing such systems on reconfigurable hardware have also significantly advanced.
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
