Anaphora AI

1. Semantic Matchmaking Phase

The client’s request contains requested asset specifications for data, algorithms and infrastructure. Since we aim to build a use case agnostic system, we must support a wide and dynamic range of application domains. To this end, we rely on an ontology-based resource retrieval and allocation system. Based on the asset ontology, each provider can analyze the asset specifications published by the client and know if it owns a matching asset, in which case it will take part in the auction phase through solution/service bids.

2. Auction Phase

The auction phase is the sequence of actions after the client’s request is made public. It involves AI providers interacting with the auction smart contract by placing service bids. Along with its solution specifications, the client locks an ANAPH token (the token used in the Anaphora AI ecosystem) amount, the reserve price, that will be used to pay AI providers for their solution. The auction ends when the client adjudicates the auction contract or when a predetermined time elapses.

3. Secure Learning Phase

Once all the providers are identified, the immaterial assets from the data provider and the Algorithm provider need to be transmitted onto the infrastructure provider's infrastructure for machine learning computation. To prevent any data or intellectual property leaks from a malicious or compromised infrastructure provider, a secured learning environment is required. For dataset security, trusted execution environments (TEE), such as Intel SGX enclaves, are used to perform both machine learning and model creation in a secure way.

4. Restitution Phase

Once the learning phase is over, the model is provided through the TEE secure channel to the client. The client then assesses the result of the model off-chain and publishes an acknowledgment in the auction SLA (Service Level Agreement) contract to close the process, unlock the payments to the providers and retrieve the potential unspent resources from its reserve price.