Sap Se

Benjamin Weggenmann

Martin Haerterich

Florian Kerschbaum

Valentin Rublack

Florian Knoerzer

Various examples are directed to systems and methods for obscuring natural language data. A programmed computer may access natural language data comprising a first plurality of sequenced tokens. The programmed computer may apply an encoder model to the first plurality of sequenced tokens to generate a latent space representation of the first plurality of sequenced tokens, the latent space representation comprising a first content latent vector describing a content of the natural language data and a first author latent vector describing an author of the natural language data. The programmed computer may modify the first author latent vector to generate an obscured author latent vector. The programmed computer may apply a decoder model to the first content latent vector and the obscured author latent vector to generate obscured natural language data, the obscured natural language data comprising a second plurality of tokens.

Data obscuring for natural language text

Various examples are directed to systems and methods for obscuring directional data to improve privacy. An example system may access a first unit of directional data. The example system may select a sampled value from an angular cumulative distribution function (CDF) of a random distribution. The example system may use the selected sampled value to generate a random sample from the random distribution and apply the random sample to the first unit of directional data to generate a first obscured unit of directional data.

Privacy arrangement for directional data

Various examples are directed to systems and methods for obscuring private information in input data. A system may apply an encoder model to an input data unit to generate a latent space representation of the input data unit. The system may apply multi-dimensional noise to the latent space representation of the input data unit, the multi-dimensional noise having a first value in a first latent space dimension and a second value different than the first value in a second latent space dimension. The system may apply a decoder model to the latent space representation of the input data unit to generate an obscured data unit.

Data obscuring for privacy-enhancement

Techniques for implementing a differentially private variational autoencoder for data obfuscation are disclosed. In some embodiments, a computer system performs operations comprising: encoding input data into a latent space representation of the input data, the encoding of the input data comprising: inferring latent space parameters of a latent space distribution based on the input data, the latent space parameters comprising a mean and a standard deviation, the inferring of the latent space parameters comprising bounding the mean within a finite space and using a global value for the standard deviation, the global value being independent of the input data; and sampling data from the latent space distribution; and decoding the sampled data of the latent space representation into output data.

Differentially private variational autoencoders for data obfuscation

Directional data often conveys particularly sensitive information, such as user location. To protect user privacy, directional data is replaced with modified directional data that is selected based on the actual directional data, a privacy parameter, and a probability distribution on an n-sphere. In this way, the modified directional data value is useful when aggregated with other modified directional data values, but does not infringe the privacy of the directional data of the user.

Privacy for directional data

A method of producing an anonymized vector for a text mining task in lieu of a feature vector is disclosed. A vocabulary is created from a corpus of documents, each of the corpus of documents having a context that is similar to a set of target documents. The set of target documents is received. The feature vector is generated from a first document of the set of target documents. The feature vector is transformed into a composition vector. A synthetic vector is constructed based on the composition vector. The synthetic vector is shared as the anonymized vector in lieu of the feature vector.

Growing community of inventors

Karlsruhe, Germany

Benjamin Weggenmann