Blockchain-enabled crowdsourcing computing system

Blockchain-enabled crowdsourcing computing system

Short answer questions

Briefly describe how the blockchain-enabled crowdsourcing computing system works.

Explain the roles and responsibilities of application initiators, computing contributors, and verification contributors in this system.

What role does smart contract play in this system?

Explain the use of homomorphic encryption in this system.

Why is decentralized storage needed in this system?

Describe two strategies for merging private models from different computing contributors.

Explain the difference between "private model" and "verified model".

In this system, how to ensure the security of private data and models?

Why in this system, verification contributors need to test encrypted private models instead of decrypted models?

How does this system incentivize participants (application initiators, computing contributors, and verification contributors)?

Answers

The blockchain-enabled crowdsourcing computing system develops machine learning models by leveraging the computing power of a distributed network. The application initiator initiates a smart contract on the blockchain platform to define the task, the computing contributor trains the private model with local data and encrypts it before uploading it, and the verification contributor tests the encrypted model and verifies its performance.

The application initiator defines the task and provides data or initial model, the computing contributor trains and shares the encrypted private model, and the verification contributor evaluates the performance of the encrypted model.

Smart contracts define the task objectives, verification criteria, participant reward mechanism and other execution rules.

Homomorphic encryption allows calculations on encrypted data, thereby protecting private models from being leaked during the verification process.

Decentralized storage is used to securely store and share data, models and other information, and ensure the resilience and accessibility of the system.

Strategy 1: Directly fuse the hidden layer features of all models and train the weight matrix. Strategy 2: Gradually fuse models to learn the correlation between different models while retaining the uniqueness of each model.

Private models are models trained by computing contributors, while verified models are private models that have been verified by verification contributors to meet the task criteria.

The privacy and security of data and models are protected by homomorphic encryption, decentralized storage and random selection of verification contributors.

Verification contributors test encrypted models to avoid leaking private models during the verification process and protect the intellectual property rights of computing contributors.

The system provides rewards to participants in the form of digital currency through the reward mechanism defined in the smart contract to encourage them to contribute computing power and data.

Professional terms

Term definitionBlockchainA decentralized, distributed database that records transactions and keeps data secure. Smart contractsA self-executing contract that runs on a blockchain, with its terms written directly into the code. Crowdsourced computingA distributed computing pattern that distributes computing tasks to a large number of individual users or devices. Machine learningA form of artificial intelligence that enables computers to learn from data and improve their performance without being explicitly programmed. Homomorphic encryptionA form of encryption that allows computation to be performed on encrypted data without first decrypting it. Decentralized storageA form of data storage where data is not stored in a single location but distributed across multiple locations. Application initiatorAn entity that defines a machine learning task and provides data or an initial model. Compute contributorAn entity that trains a private model using local data and shares it to the system. Verification contributorAn entity that evaluates the performance of an encrypted private model and determines whether it meets the criteria for the task. Private modelA machine learning model trained by a compute contributor that is encrypted before sharing to protect its privacy. Verified modelA private model that is verified by a verification contributor to meet the criteria for the task. Model fusionThe process of combining multiple machine learning models into a single model with better performance. Decentralized Application (DApp) An application running on a decentralized network (such as a blockchain) that is not controlled by any single entity. Software Defined Network (SDN) A network architecture that allows network traffic to be programmed and dynamically managed by a centralized controller, allowing for greater flexibility and controllability. Peer-to-Peer Network (P2P) A decentralized network architecture in which each node can act as a client and server, communicating directly with other nodes without routing through a central server.