☰
Credit Card Fraud Demonstration
Evaluating Model Accuracy
A model has been pretrained on existing credit card fraud data. The model will have its accuracy evaluated by comparing the true outcomes to the predicted results.
Data
This is a sample of the data as it appears in the trusted environment. Data is encrypted before being sent off for processing in an untrusted environment:
V1
V2
...
V28
Amount
Encrypted Data
This is a buffered sample of the data as it appears in the untrusted environment. Data is received in an encrypted form and then prepared in the FHE context for processing:
Accuracy
Time Taken
Prediction took: 0s
Real Time Fraud Detection
By creating a data set with a single transaction we can evaluate whether or not this transaction is fraudulent by encrypting it and sending it to a third party to be evaluated using an encrypted logistic regression model.
Data
This is an example of a fraudulent transaction as it appears in the trusted environment. This transaction can be encrypted just before evaluation to get real time secure predictions:
V1
V2
...
V28
Amount
Encrypted Data
This is a buffered sample of the data as it appears in the untrusted environment. Data is received in an encrypted form and then prepared in the FHE context for processing:
Predicted Outcome
Transaction is not fraudulent.
Time Taken
Prediction took: 0s
Patient Action Recommendation
In this demonstration a group of patients are evaluated against a pre-trained neural network. With the patient data being protected and highly sensitive information sending this data to a third party would be a breach of confidentiality but by encrypting it before having it evaluated we can circumvent these restrictions and avoid a breach.
Data
This is a sample of the codified data as found in the 16 samples before analysis. The data is an extracted and altered form of the dataset found here. The data is encrypted before being sent off for processing in an untrusted environment:
Predictions
These are the results produced by the analysis. The true patient outcome is the action the patient should have taken in retrospect. The predicted patient outcome is the action we predict the patient should take analysing the encrypted data without knowledge of the true outcome.
True Patient Outcome
Predicted Patient Outcome
Accuracy
Time Spent
Time taken for all predictions: 0s
Time taken for each individual prediction: 0s
Search an Encrypted Database
A database can be searched by encrypting both the dataset and the query and running a search for the encrypted query. This means that organisations can keep their data encrypted while preserving its usefulness without having to unencrypt it at the time of query thus exposing it to security risks.
Data
This is a sample of the data as it appears in the trusted environment. The dataset has only two columns:
Country
Capital
Encrypted Data
This is a buffered sample of the data as it appears in the untrusted environment. Data is received in an encrypted form and then prepared in the FHE context for processing. The data is encrypted into a CTile format:
Results
The of is
Time Spent
Time taken to encrypt the database: 0s
Time taken to encrypt the query: 0s
Time taken to perform the search: 0s
Time taken to decrypt the result: 0s
Searching an Encrypted Database
A database can be searched by encrypting both the dataset and the query and running a search for the encrypted query. This means that organisations can keep their data encrypted while preserving its usefulness without having to unencrypt it at the time of query thus exposing it to security risks.
Data
This is a sample of the data as it appears in the trusted environment. Two columns have been separated from the dataset and encrypted:
Country