Abstract: An encoding process performed by a computing device (e.g., a user's private device) can include obtaining private data that includes a private value. According to an aspect of the present disclosure, the computing device can produce a plurality of messages that respectively comprise a plurality of message values, where a total sum of the plurality of message values approximates the private value, and where at least one of the plurality of message values is randomly selected. The device can provide the plurality of messages for aggregation with a plurality of additional messages respectively generated for a plurality of additional private values. For example, the messages can be transmitted to a shuffler model configured to shuffle the plurality of messages with the plurality of additional messages.

Abstract: An encoding process performed by a computing device (e.g., a user's private device) can include obtaining private data that includes a private value. According to an aspect of the present disclosure, the computing device can produce a plurality of messages that respectively comprise a plurality of message values, where a total sum of the plurality of message values approximates the private value, and where at least one of the plurality of message values is randomly selected. The device can provide the plurality of messages for aggregation with a plurality of additional messages respectively generated for a plurality of additional private values. For example, the messages can be transmitted to a shuffler model configured to shuffle the plurality of messages with the plurality of additional messages.

Abstract: An encoding method for enabling privacy-preserving aggregation of private data can include obtaining private data including a private value, determining a probabilistic status defining one of a first condition and a second condition, producing a multiset including a plurality of multiset values, and providing the multiset for aggregation with a plurality of additional multisets respectively generated for a plurality of additional private values. In response to the probabilistic status having the first condition, the plurality of multiset values is based at least in part on the private value, and in response to the probabilistic status having the second condition, the plurality of multiset values is a noise message. The noise message is produced based at least in part on a noise distribution that comprises a discretization of a continuous unimodal distribution supported on a range from zero to a number of multiset values included in the plurality of multiset values.

Abstract: A computer-implemented method of compressing floating point data of a machine-learned model into a compressed representation of the floating point data can include obtaining floating point data including a plurality of machine-learned model parameters encoded as a tensor, determining a sign vector including a sign bit of each of the plurality of machine-learned model parameters, determining a normalization exponent based on the floating point data, determining a plurality of offsets descriptive of a difference between an exponent of the machine-learned model parameters and the normalization exponent, determining a bitmap including a unary representation of the plurality of offsets, determining a plurality of adjusted mantissas based at least in part on the plurality of offsets, and storing a compressed representation of the floating point data, the compressed representation including the sign vector, the normalization exponent, the bitmap, and one or more bits of each of the plurality of adjusted mantissas.

Abstract: The present disclosure provides practical communication-efficient and low-error algorithms for aggregation of private data. For example, the proposed algorithms can be implemented in the shuffled DP model. Specific example operations that can be performed using the proposed algorithms include summation (e.g., binary summation, integer summation) and histograms over a moderate number of buckets. The proposed algorithms achieve accuracy that is arbitrarily close to that of central DP algorithms with an expected communication per user essentially matching what is needed without any privacy constraints.

Abstract: An encoding method for enabling privacy-preserving aggregation of private data can include obtaining private data including a private value, determining a probabilistic status defining one of a first condition and a second condition, producing a multiset including a plurality of multiset values, and providing the multiset for aggregation with a plurality of additional multisets respectively generated for a plurality of additional private values. In response to the probabilistic status having the first condition, the plurality of multiset values is based at least in part on the private value, and in response to the probabilistic status having the second condition, the plurality of multiset values is a noise message. The noise message is produced based at least in part on a noise distribution that comprises a discretization of a continuous unimodal distribution supported on a range from zero to a number of multiset values included in the plurality of multiset values.

Abstract: An encoding method for enabling privacy-preserving aggregation of private data can include obtaining private data including a private value, determining a probabilistic status defining one of a first condition and a second condition, producing a multiset including a plurality of multiset values, and providing the multiset for aggregation with a plurality of additional multisets respectively generated for a plurality of additional private values. In response to the probabilistic status having the first condition, the plurality of multiset values is based at least in part on the private value, and in response to the probabilistic status having the second condition, the plurality of multiset values is a noise message. The noise message is produced based at least in part on a noise distribution that comprises a discretization of a continuous unimodal distribution supported on a range from zero to a number of multiset values included in the plurality of multiset values.

Abstract: An encoding process performed by a computing device (e.g., a user's private device) can include obtaining private data that includes a private value. According to an aspect of the present disclosure, the computing device can produce a plurality of messages that respectively comprise a plurality of message values, where a total sum of the plurality of message values approximates the private value, and where at least one of the plurality of message values is randomly selected. The device can provide the plurality of messages for aggregation with a plurality of additional messages respectively generated for a plurality of additional private values. For example, the messages can be transmitted to a shuffler model configured to shuffle the plurality of messages with the plurality of additional messages.

Abstract: A computer-implemented method of compressing floating point data of a machine-learned model into a compressed representation of the floating point data can include obtaining floating point data including a plurality of machine-learned model parameters encoded as a tensor, determining a sign vector including a sign bit of each of the plurality of machine-learned model parameters, determining a normalization exponent based on the floating point data, determining a plurality of offsets descriptive of a difference between an exponent of the machine-learned model parameters and the normalization exponent, determining a bitmap including a unary representation of the plurality of offsets, determining a plurality of adjusted mantissas based at least in part on the plurality of offsets, and storing a compressed representation of the floating point data, the compressed representation including the sign vector, the normalization exponent, the bitmap, and one or more bits of each of the plurality of adjusted mantissas.

Abstract: An encoding method for enabling privacy-preserving aggregation of private data can include obtaining private data including a private value, determining a probabilistic status defining one of a first condition and a second condition, producing a multiset including a plurality of multiset values, and providing the multiset for aggregation with a plurality of additional multisets respectively generated for a plurality of additional private values. In response to the probabilistic status having the first condition, the plurality of multiset values is based at least in part on the private value, and in response to the probabilistic status having the second condition, the plurality of multiset values is a noise message. The noise message is produced based at least in part on a noise distribution that comprises a discretization of a continuous unimodal distribution supported on a range from zero to a number of multiset values included in the plurality of multiset values.

Abstract: A computer-implemented method of processing a query vector and a data vector), comprising: generating a set of masks and a first set of multiple signatures and a second set of multiple signatures by applying the set of masks to the query vector and the data vector, respectively, and generating candidate pairs, of a first signature and a second signature, by identifying matches of a first signature and a second signature. The set of masks comprises a configuration of the elements that is a Hadamard code; a permutation of a Hadamard code; or a code that deviates from a Hadamard code or a permutation of a Hadamard code in less than 40% of its elements.

Abstract: A computer-implemented method of processing a query vector and a data vector), comprising: generating a set of masks and a first set of multiple signatures and a second set of multiple signatures by applying the set of masks to the query vector and the data vector, respectively, and generating candidate pairs, of a first signature and a second signature, by identifying matches of a first signature and a second signature. The set of masks comprises a configuration of the elements that is a Hadamard code; a permutation of a Hadamard code; or a code that deviates from a Hadamard code or a permutation of a Hadamard code in less than 40% of its elements.

Abstract: The disclosure relates to a device for removing filtered substances from combustion processes from a filter under laboratory conditions. In some implementations, this comprises a mount for the filter and a rinsing means for rinsing out the filter in its mount by means of a liquid. It may also relate to a method for removing filtered substances from combustion processes from a mounted filter under laboratory conditions in order to analyse the filtrate, wherein the filter is rinsed out in its mount with a liquid.