Non-Interactive Private Decision Tree Evaluation
AssigneeSap SE
InventorsAnselme Kemgne Tueno, Yordan Boev, Florian Kerschbaum
Life of this patent
Estimating the term…
Abstract
A non-interactive protocol is provided for evaluating machine learning models such as decision trees. A client can delegate the evaluation of a machine learning model such as a decision tree to a server by sending an encrypted input and receiving only the encryption of the result. The inputs can be encoded using their binary representation. Efficient data representations are then combined with different algorithmic optimizations to keep the computational overhead and the communication cost low. Related apparatus, systems, techniques and articles are also described.
Claims20
A computerized method comprising:
- homomorphically encrypting a plaintext data set bitwise;
- generating a plurality of random strings;
- forming an encrypted input comprising the plurality of random strings and the homomorphically encrypted data set;
- transmitting the encrypted input to a server executing a decision tree for evaluation;
- and
- receiving, from the server, data characterizing the evaluation by the server;
- and
- decrypting the evaluation.
A computerized method for securely classifying encrypted data transmitted over a network to a remotely executed machine learning model, the method comprising:
- receiving, by a server from a client, an encrypted input encapsulating a data set;
- generating, by the server using the data set and without decryption, a classification using a computer-implemented machine learning model comprising a decision tree;
- and
- providing, by the server to the client, the generated classification to enable the client to decrypt the classification.
A computerized method comprising:
- receiving, by a server from a client via a network, an encrypted input encapsulating a data set;
- homomorphically evaluating, by the server without decryption, the encrypted input using a decision tree;
- and
- transmitting, by the server to the client, a result of the evaluation over the network to enable the client to decrypt the evaluation.
Dependencies are inferred from the claim text.