Private deep neural network training
Systems and methods for private deep neural network training are disclosed. Method includes storing first private values at first machine and second private values at second machine; providing, to third machine, first share of first private values and first share of second private values; providing, to fourth machine, second share of first private values and second share of second private values; computing, at third machine, third machine-value based on first share of first private values and first share of second private values; computing, at fourth machine, fourth machine-value based on second share of first private values and second share of second private values; providing, to first machine and second machine, third machine-value and fourth machine-value; and computing, at first machine, a mathematical function of first private values and second private values, mathematical function being computed based on first private values stored at first machine, third machine-value, and fourth machine-value.
1. A method comprising:
storing first private values at a first machine and second private values at a second machine;
providing, to a third machine, a first share of the first private values and a first share of the second private values;
providing, to a fourth machine, a second share of the first private values and a second share of the second private values;
computing, at the third machine, a third machine value based on the first share of the first private values and the first share of the second private values;
computing, at the fourth machine, a fourth machine value based on the second share of the first private values and the second share of the second private values;
providing, to the first machine and the second machine, the third machine value and the fourth machine value;
computing, at the first machine, a mathematical function of the first private values and the second private values, the mathematical function being computed based on the first private values stored at the first machine, the third machine value, and the fourth machine value; and
providing an output of the computed mathematical function.
2. The method of claim 1 , further comprising:
computing, at the second machine, the mathematical function of the first private values and the second private values, the mathematical function being computed based on the second private values stored at the second machine, the third machine value, and the fourth machine value.
3. The method of claim 2 , wherein the second machine lacks access to the first private values stored at the first machine.
4. The method of claim 1 , wherein the first machine lacks access to the second private values stored at the second machine.
5. The method of claim 1 , wherein the first private values are constructible based on the first share of the first private values and the second share of the first private values, and wherein the second private values are constructible based on the first share of the second private values and the second share of the second private values.
6. The method of claim 5 , wherein the first private values comprise an additional mathematical function of the first share of the first private values and the second share of the first private values, and wherein the second private values comprise the additional mathematical function of the first share of the second private values and the second share of the second private values.
7. The method of claim 6 , wherein the additional mathematical function comprises a sum.
8. One or more non-transitory machine-readable media storing instructions which, when executed by processing circuitry of one or more machines, cause the processing circuitry to perform operations comprising:
storing first private values at a first machine and second private values at a second machine;
providing, to a third machine, a first share of the first private values and a first share of the second private values;
providing, to a fourth machine, a second share of the first private values and a second share of the second private values;
computing, at the third machine, a third machine value based on the first share of the first private values and the first share of the second private values;
computing, at the fourth machine, a fourth machine value based on the second share of the first private values and the second share of the second private values;
providing, to the first machine and the second machine, the third machine value and the fourth machine value;
computing, at the first machine, a mathematical function of the first private values and the second private values, the mathematical function being computed based on the first private values stored at the first machine, the third machine value, and the fourth machine value; and
providing an output of the computed mathematical function.
9. The one or more machine-readable media of claim 8 , the operations further comprising:
computing, at the second machine, the mathematical function of the first private values and the second private values, the mathematical function being computed based on the second private values stored at the second machine, the third machine value, and the fourth machine value.
10. The one or more machine-readable media of claim 9 , wherein the second machine lacks access to the first private values stored at the first machine.
11. The one or more machine-readable media of claim 8 , wherein the first machine lacks access to the second private values stored at the second machine.
12. The one or more machine-readable media of claim 8 , wherein the first private values are constructible based on the first share of the first private values and the second share of the first private values, and wherein the second private values are constructible based on the first share of the second private values and the second share of the second private values.
13. The one or more machine-readable media of claim 12 , wherein the first private values comprise an additional mathematical function of the first share of the first private values and the second share of the first private values, and wherein the second private values comprise the additional mathematical function of the first share of the second private values and the second share of the second private values.
14. The one or more machine-readable media of claim 13 , wherein the additional mathematical function comprises a sum.