This decentralized spam detection demo showcases the implementation of a secure multi-party computation (MPC) protocol directly within your web browser by importing and running the Nada program simulator using PyScript and Pyodide.
The nltk library is used to tokenize text.
The spam identification Logistic Regression model, trained by its owner, distributes the encrypted model to the network. The data owner, without revealing the email content, tokenizes the message and privately sends it to the network for classification.
Model owner
1. The model owner starts by uploading the model to the browser.
2. Hit Mask & Classify to mask the model and send the secrets to the simulated network.
Data owner
3. Grab an email from your spam folder and paste it below. You can also randomly use some emails previously generated by chatGPT.
4. Tokenize the message first and then hit Mask & Classify.
Each node receives the masked model and tokenized email vector, and then computes its secret share of the overall spam prediction according to a logistic regression model.
Node 1
Registration Token
Authentication Token
Secret Share of Result
Node 2
Registration Token
Authentication Token
Secret Share of Result
Node 3
Registration Token
Authentication Token
Secret Share of Result
The data owner receives the secret shares of the result from each node and reconstructs the overall output.