Skip to main content

Introduction

We invite data scientists to participate in an exciting ML competition! The goal is to develop a machine learning model that can accurately predict the probability of default for our consumer contracts at [address]. The winner of this challenge will have the opportunity to become an ML service provider for a consumer contract located at [address], earning fees for each inference performed.

Competition Details

Data:​

Input data:

  • On-chain metadata: Accessible from the contract pool at [address]
  • Off-chain metadata: Retrieved from the trusted source API at [api-url], corresponding to target token ID

Training data: [link to ipfs]

Test data for validators: [api]

API to access test result data (validators only): [URL]

Note: Validators must sign messages with their registered Credio private key before sending them to the API. The competition creator will verify the signature off-chain and return the result to the validator. Additionally, the timestamp may be checked before a response is provided.

Model Development:​

  • Model input: [10 off-chain fields, 5 on-chain fields] → [PD] with scale parameters set to 10 (using the Credio CLI to generate the settings file)
  • Model verification: Use the aggregate feature in the CLI to create 3 contract verifiers: single proof, aggregate 2 proofs, and aggregate 5 proofs
  • Model development tool: Credio CLI

Timeline:​

Model submission deadline: 30 days from the creation of the contract. For each model submission, the modeler must provide the following:

  • Commit model verification key: This includes keys for both off-chain verification and the on-chain verification contract.
  • API for zkML model inference: This API allows the validator to retrieve results generated by your model using zkML.
  • API for pure ML model inference: This API allows users (including the validator) to retrieve results generated by your model without zk-proofs (pure machine learning).

Evaluation:​

Evaluation criteria: Modeler with the highest accuracy in predicting default probability on the test data set Winner's prize: Opportunity to become an ML service provider for the consumer contract at [address], earning 1 USD per inference for predicting the default probability of a single asset (5 USD for aggregate 5 proof)

Staking:​

Participating modelers must stake tokens. If a modeler wins the challenge, they will upload proofs to the on-chain system and claim rewards from consumers. The challenge creator will also stake tokens to prevent spam challenges and support a future reporting mechanism.