🎯Encrypted AI Inference Solution
The Why:
AI requires security & control. Pretrained Large Language Models (LLMs) have been shown to memorize training examples and have been misused to recover private information. Without protection, AI models could potentially be manipulated to cause unpredictable impact to human society.
The How:
Register and create your own Data Cocoon (A secure data mat for only you)
Write/Upload: The user's data (Alice) is encrypted on the client side, then written or uploaded to Mind Network with encryption.
Predict/Score: The encrypted input from the data owner (Alice) could be used as input for AI models to predict and score results. The AI models could also be encrypted so that no others could modify the parameters or manipulate the results. Encrypted prediction is completed and reverted back to the data owner to ensure end-to-end protection for the whole AI pipeline.
The What (A Live Example: https://mindnetwork.xyz/):
Step 1: draw a figure, encrypt with a generated private key, then upload to Mind Network
Step 2: run an AI model to classify the encrypted figure with Mind Network
Step 3: decrypt the classification result with the private key
Last updated
