Vadym Doroshenko

Public partitions as a cartesian product of dimensions

2

# Context **Note**: [here](https://pipelinedp.io/key-definitions/) is more about terminology. **Definitions** (from terminology page) _A partition_ is a subset of the data corresponding to a given value of the aggregation criterion. Usually...

Vector summation DP aggregation

7

# Context [DPEngine.aggregate](https://github.com/OpenMined/PipelineDP/blob/41b70a3c7e19b82024e2d0f44842aaab570440bd/pipeline_dp/dp_engine.py#L52) performs DP aggregations of scalar values (sum, count, mean etc). A set of computed metrics is controlled with [metrics](https://github.com/OpenMined/PipelineDP/blob/41b70a3c7e19b82024e2d0f44842aaab570440bd/pipeline_dp/aggregate_params.py#L76) field of `aggregate_params` argument. The result of this...

Find nice datasets and use cases for anonymizating with PipelineDP

6

This issue for tracking ideas of datasets and usecases of using PipelineDP. Having datasets/use cases would be helpful for - showing how to work with PipelineDP (maybe in Colab) -...

Laplace/Gaussian partition selection strategies.

3

# Context From [key definitions](https://pipelinedp.io/key-definitions/) (see there for more details): _A partition_ is a subset of the data corresponding to a given value of the aggregation criterion. Usually we want...

Introduce per partition contribution bounding for sum

2

# Context **Note:** Terminology might be found [here](https://pipelinedp.io/key-definitions/). In order to protect data of each user, it's required to make contribution bounding of each user data. There are 2 types...

Auto-tunning contribution bounding parameters

2

# Context [DPEngine.aggregate](https://github.com/OpenMined/PipelineDP/blob/41b70a3c7e19b82024e2d0f44842aaab570440bd/pipeline_dp/dp_engine.py#L52) is API function that performs DP aggregation. [Parameters](https://github.com/OpenMined/PipelineDP/blob/41b70a3c7e19b82024e2d0f44842aaab570440bd/pipeline_dp/aggregate_params.py#L55) of this function include parameters related to contribution bounding (`max_partitions_contributed`, `max_contributions_per_partition`). For `max_partitions_contributed`, `max_contributions_per_partition` if a privacy unit (definition...

# Context On the [issue](https://github.com/OpenMined/PipelineDP/issues/10) a framework for Explain calculation reports ([code](https://github.com/OpenMined/PipelineDP/blob/main/pipeline_dp/report_generator.py)) was implemented. Currently, when the most computations has been implemented, we can add information about these computation in...

Advance composition in PipelineDP

4

# Short DP references **Definition:** Basic (or naive) composition of differential privacy mechanisms with parameters `(eps1, delta1)` and `(eps2, delta2)` is a mechanism with parameter `(eps1+eps2, delta1+delta2)`. There are many...

# Context [PyDP](https://github.com/OpenMined/PyDP) library wraps [Google C++ building block library](https://github.com/google/differential-privacy/tree/main/cc/algorithms). Algorithms are inherited from [MetaAlgorithm](https://github.com/OpenMined/PyDP/blob/a88ee73053aa2bdc1be327a77109dd5907ab41d6/src/pydp/algorithms/_algorithm.py#L10) class. For now the most interesting for PipelineDP is [Pecentile](https://github.com/OpenMined/PyDP/blob/a88ee73053aa2bdc1be327a77109dd5907ab41d6/src/pydp/algorithms/laplacian/_percentile.py#L8) algorithm, but in future support...

DataFrame style API

2

# Context Now **PipelineDP** supports 3 execution modes - with Apache Spark, Apache Beam, w/o frameworks ([here](https://github.com/OpenMined/PipelineDP/blob/main/examples/movie_view_ratings/run_all_frameworks.py) is an example how to run on different frameworks). Basically the current API...