Abstract: A federated query engine system and method for multiple datasets is enhanced with privacy preserving features. It may, for example, limit the movement of data from one or more of the datasets being accessed. It may use cryptographic long-term keys, enabling fuzzy table joins that do not require a comparison of the plaintext column values. The query plan may leverage the particular infrastructure of the storage system that houses each of the datasets. The query engine receives a standard SQL query, translates the query into a logical plan for performing the query across the multiple datasets, converts the logical plan into physical plans that are specific to the implementational architecture of the multiple datasets, and sends these physical plans to SQL workers located near the data warehouses housing each dataset.

Abstract: A machine-learning system includes worker nodes communicating with a single server node. Worker nodes are independent neural networks initialized locally on separate data silos. The server node receives the last layer output (“smashed data”) from each worker node during training, aggregates the result, and feeds into its own server neural network. The server then calculates an error and instructs the worker nodes to update their model parameters using gradients to reduce the observed error. A parameterized level of noise is applied to the worker nodes between each training iteration for differential privacy. Each worker node separately parameterizes the amount of noise applied to its local neural network module in accordance with its independent privacy requirements.

Abstract: The present invention is directed to a differential privacy platform in which the privacy risk of a computation can be objectively and quantitatively calculated. This measurement is performed by simulating a sophisticated privacy attack on the system for various measures of privacy cost or epsilon, and measuring the level of success of the attack. In certain embodiments, a linear program reconstruction-type attack is used. By calculating the loss of privacy resulting from sufficient attacks at a particular epsilon, the platform may calculate a level of risk for a particular use of data. The privacy budget for the use of the data may thereby be set and controlled by the platform to remain below a desired risk threshold.

Abstract: A data analytics platform provides secure access to federated data for advanced analytics and machine learning. No raw data is exposed or moved outside its original location, thereby providing data privacy. A coordinator located in the provider cloud communicates with runners in each client data silo. The runners ensure that no raw private data is ever exposed to the coordinator. Silo managers are implemented in the client data silo in order to manage and maintain the client cloud components of the platform remotely. In some embodiments, the platform can anonymize verified models for privacy and compliance, and users can export and deploy secure models outside the original data location.