Proof of Inference

In the realm of AI, ensuring that the inference results from machine learning models are trustworthy and verifiable is critical. Incu AI introduces a robust Proof of Inference system to guarantee the integrity and reliability of AI model outputs. This system leverages a combination of advanced technologies and methodologies to provide secure and transparent inference proofs. Here’s a detailed explanation of how Proof of Inference works on Incu AI:

Proof of Inference Workflow

1. Submit Inference Task:

   User Submission: Users submit their input data to the Incu AI platform for processing by an AI model.

2. Run AI Model:

   Execution by ROFL: The Runtime Off-chain Logic (ROFL) executes the AI model on the provided input data to generate the inference results.

3. Generate Cryptographic Proof:

   Proof Creation: ROFL creates a cryptographic proof of the inference computation. Depending on the model's size and complexity, this proof can be generated using Trusted Execution Environments (TEEs) for larger models or zero-knowledge proofs (zkProofs) for smaller models. These proofs ensure that the computation was carried out correctly and securely.

4. Secure Transmission:

   Sending Proof and Results: ROFL securely transmits the inference results and the generated cryptographic proof to the Runtime On-chain Logic (RONL) using secure protocols such as EnclaveRPC.

5. Verification:

   Proof Verification: RONL verifies the cryptographic proof against predefined criteria and attestations to ensure the correctness and integrity of the inference computation.

6. Blockchain Recording:

   Recording on Blockchain: If the proof is validated successfully, RONL records the inference results and the proof on the blockchain. This step provides transparency and ensures that the results are immutable and auditable.

Technologies Used

Trusted Execution Environments (TEEs):

Security: TEEs provide a secure and isolated execution environment that ensures the integrity and confidentiality of the computations. They are particularly useful for handling larger models that require significant computational resources.

Verification: TEEs generate attestations that verify the secure execution of inference tasks, which can be trusted by the on-chain logic.

Zero-Knowledge Proofs (zkProofs):

• Privacy: zkProofs enable the verification of computation correctness without revealing the input data or the computational process itself. This is ideal for scenarios where privacy is paramount.

• Efficiency: These proofs are used for smaller models where computational efficiency and minimal overhead are crucial.

Adaptive Model Selection

Small Models:

• zkProofs: For smaller models, Incu AI uses zkProofs to provide efficient and private verification of inference computations.

Large Models:

• TEEs: For larger models, TEEs are utilized to ensure secure and reliable computation, leveraging hardware-based isolation for enhanced security.