Walrus
Integration of Walrus Decentralized Storage in Our Project
Walrus was a foundational component of our decentralized application, ensuring data storage, retrieval, and overall system integrity in a fully decentralized manner. Here's how Walrus was integrated into the workflow:
Model Publishing and Retrieval:
Model Publishing:
When a Model Publisher uploads a machine learning model, the model is securely stored on Walrus.
This decentralized storage ensures that the model is accessible to all contributors without relying on a centralized server.
Model Retrieval by Contributors:
Contributors, upon deciding to participate and submitting their required stake, retrieve the published model directly from Walrus.
This decentralized retrieval ensures the system's transparency and autonomy, aligning with the principles of decentralized storage.
Local Machine Training and Verification:
Contributor Training:
Contributors perform local machine model training on the data or model retrieved from Walrus.
Verification of Training:
Once training is completed, the output is submitted alongside an encryption file based on either:
Pedersen Commitment Scheme or
Paillier Cryptosystem
These cryptographic schemes verify the integrity and correctness of the training process. This step is crucial to identify and penalize potentially malicious nodes.
Penalty for Malicious Nodes:
If verification reveals invalid training results, the contributor's stake is deducted, deterring malicious behavior and maintaining system trustworthiness.
Secure Cost Function with AES Encryption:
The verification encryption file also includes a cost function, which is encrypted using AES encryption for added security and confidentiality.
This ensures that critical computational details are protected from unauthorized access.
Dataset Upload and Usage:
User-Contributed Datasets:
Users can upload their datasets to the platform, encrypted with AES encryption, and accompanied by a description to facilitate selection by contributors.
Contributor Dataset Usage:
If a contributor finds a user’s dataset suitable for their task, they can request the decryption key from the user.
This decentralized and user-controlled mechanism ensures data privacy while enabling collaborative usage.
Walrus' Critical Role:
Walrus played a crucial role in ensuring that the entire project remained decentralized:
Data Decentralization: All models and datasets are stored on Walrus, eliminating reliance on centralized servers.
Access Control: Contributors and users interact with the stored content in a secure, transparent, and decentralized manner.
Enhanced Privacy: The integration of cryptographic schemes and AES encryption ensured that sensitive information, such as cost functions and datasets, remained secure.
By leveraging Walrus as the core storage backbone, we achieved a robust, decentralized application architecture that was transparent, secure, and efficient.
Last updated