Trusted Compute

With each ML primitive, Neural focuses on meeting developers where they are comfortable and giving them optionality to build a product which meets their needs.

Certain machine learning applications like image generation and large language models require high computational power which breaches the practical limits of ZKML and OPML in their current state. Neural enables onchain access to these cutting edge models via Trusted ML (or Trusted Compute). Trusted ML empowers developers to seamlessly integrate the Neural Solidity SDK and offchain oracle node into existing machine learning workflows to build powerful and familiar apps for users.

Neural’s application of Trusted ML is best considered as a sliding scale, allowing the developer to determine the extent of trust, verification, redundancy, and IP protection. Within the Neural SDK developers have the option to customize the storage of model input and output to increase transparency and allow for public auditing, or increase redundancy and verification. Developers may also choose to allow multiple nodes to respond to each request which adds an extra layer of redundancy and verification which decreases the trust assumptions. Using Trusted ML, developers can protect their IP within Neural’s trusted environment.

A live example of Trusted ML can be found at animator.neuralexamples.com. For a more comprehensive understanding of the technical architecture behind this application, see how Animetor works under the hood here.