Horizontal federated forest via secure aggregation
One example method includes constructing a machine learning model which, when completed, is operable to screen candidates, from a group of candidates, to define a candidate pool that has specified characteristics. The constructing includes: broadcasting, from a central node to edges of a federation, an indication that construction of a random forest, of the machine learning model, has started; performing a federated feature categorization, by the central node based on information received from the edges, of a feature to be included in respective decision trees of the edges; based on the categorizing, broadcasting a feature category to the edges; performing, by the central node using respective purity information received from the edges, a federated purity calculation; and based on the federated purity calculation, broadcasting, by the central node to the edges, a winning feature split for the feature.
1 . A method comprising:
constructing a machine learning model which, when completed, is operable to screen candidates, from a group of candidates, to define a candidate pool that has specified characteristics, wherein the constructing comprises:
broadcasting, from a central node to edge nodes of a federation, an indication that construction of a random forest of the machine learning model has started;
performing a federated feature categorization based on information received from the edge nodes, wherein the federated feature categorization includes categorizing a feature that is to be included in respective decision trees of the edge nodes;
based on the categorizing, broadcasting a feature category to the edge nodes;
performing, by the central node using respective purity information received from the edge nodes, a federated purity calculation; and
based on the federated purity calculation, broadcasting, by the central node to the edge nodes, a winning feature split for the feature,
wherein the federated purity calculation and the federated feature categorization are performed using a secure aggregation protocol that employs pairwise masking vectors that are structured in a manner so as to guarantee that zero information about any particular edge node can be obtained by an entity that has access to only one of the pairwise masking vectors.
2 . The method as recited in claim 1 , wherein the federated purity calculation is based on an information gain determined for the feature.
3 . The method as recited in claim 1 , wherein the federated feature categorization is performed without any exchange of data among the edge nodes.
4 . The method as recited in claim 1 , wherein the federated purity calculation is performed without any exchange of data among the edge nodes.
5 . The method as recited in claim 1 , wherein each of the edge nodes corresponds to a respective business entity to which the central node provides construction of the model as a service.
6 . The method as recited in claim 1 , wherein the feature is an attribute of one of the candidates.
7 . The method as recited in claim 1 , further comprising training, by the central node, the machine learning model using information received by the central node from the edge nodes.
8 . The method as recited in claim 7 , wherein the information from the edge nodes is aggregated to prevent an unauthorized actor from accessing any information of any of the edge nodes.
9 . The method as recited in claim 7 , wherein the pairwise masking vectors, which, when summed by the central node, cancel each other out.
10 . A non-transitory storage medium having stored therein instructions that are executable by one or more hardware processors to perform operations comprising:
constructing a machine learning model which, when completed, is operable to screen candidates, from a group of candidates, to define a candidate pool that has specified characteristics, and the constructing comprises:
broadcasting, from a central node to edges of a federation, an indication that construction of a random forest of the machine learning model has started;
performing a federated feature categorization of a feature to be included in respective decision trees of the edges, wherein said performing is performed by the central node and is performed based on information received from the edges;
based on the categorizing, broadcasting a feature category to the edges;
performing, by the central node using respective purity information received from the edges, a federated purity calculation; and
based on the federated purity calculation, broadcasting, by the central node to the edges, a winning feature split for the feature,
wherein the federated purity calculation and the federated feature categorization are performed using a secure aggregation protocol that employs pairwise masking vectors that are structured in a manner so as to guarantee that zero information about any particular edge can be obtained by an entity that has access to only one of the pairwise masking vectors.
11 . The non-transitory storage medium as recited in claim 10 , wherein the federated purity calculation is based on an information gain determined for the feature.
12 . The non-transitory storage medium as recited in claim 10 , wherein the federated feature categorization is performed without any exchange of data among the edges.
13 . The non-transitory storage medium as recited in claim 10 , wherein the federated purity calculation is performed without any exchange of data among the edges.
14 . The non-transitory storage medium as recited in claim 10 , wherein each of the edges corresponds to a respective business entity to which the central node provides construction of the model as a service.
15 . The non-transitory storage medium as recited in claim 10 , wherein the feature is an attribute of one of the candidates.
16 . The non-transitory storage medium as recited in claim 10 , further comprising training, by the central node, the machine learning model using information received by the central node from the edges.
17 . The non-transitory storage medium as recited in claim 16 , wherein the information from the edges is aggregated to prevent an unauthorized actor from accessing any information of any of the edges.
18 . The non-transitory storage medium as recited in claim 16 , wherein the pairwise masking vectors, when summed by the central node, cancel each other out.