Computing with Neural Networks/VadumSecure

Vadum will develop and test Cryptography Hidden In Plain Sight (CHIPS) – a novel technique for secure computing with neural networks used in computer vision applications. The technique transforms any deep convolutional neural network (CNN) operating on data in the plaintext domain into a cryptographically secure equivalent making it resilient to white-box and black-box attacks. Homomorphic encryption schemes are theoretically capable of a similar level of security, but practical implementations introduce noise, latency and complexity preventing widespread adoption. CHIPS can encrypt a CNN of any depth and complexity with secure and straightforward network and data transformations, with minimal impact on speed of inference.