Michael Shoemate

Results 284 issues of Michael Shoemate

A component that may only be approved under a reduced privacy definition. The evaluated output is substituted directly into the release in the same way as the Literal component. The...

We should eliminate the class of queries on unsized datasets where the definition of neighboring is based on the substitution distance. The set of neighboring datasets excludes all datasets with...

First, rename "batch" in proto::Component to "submission", because one submission may contain multiple batches. When compute_epsilon is called: 1. Derive batch ids from dependency structure on graph. 2. Subpartition derived...

There are some remaining assumptions we make of the data-- - ~~the number of columns is known~~ - ~~the number of columns is nonzero~~ - the number of rows is...

Design Phase

There are multiple potential implementations. Here is how a covariance implementation could look. We need the additional components `Inverse` and `Matmul`. ## Component: Inverse ### Arguments - data ## Component:...

Some utility components that manipulate properties. ToContinuous takes in categorical integers, and outputs continuous integers. Pseudocode: ``` Nature::Continuous { lower: min(categories), upper: max(categories) } ``` ToCategorical takes in continuous integers,...

Effort 1 - Small :coffee:
Priority 3: Low

Randomized components have a different value every time they are run. At this point, the only randomized private component is the even partition. Unfortunately, you can extend partitioned portions of...

If a node ends up using less privacy than the upper bound provided to it, this lower usage is stored in the release. But nodes introduced by expansions are typically...

When data passes through mechanisms, the mechanism switches the `releasable` property to `true`. Currently the runtime releases anything with `releasable` set to `true`. Going forward, there could be an extra...

new algorithm
Priority 2: Medium

Partial implementation is up in the `groupby-aggregate` branch: https://github.com/opendifferentialprivacy/whitenoise-core/tree/groupby-aggregate What is desired: A unified system to split data, release aggregations of splits, and release aggregations of aggregations of splits. Components...

new algorithm
Priority 2: Medium