Mathematical functions#
This layer contains tools for performing high-level operations such as matrix multiplication and polynomial evaluation. Central to this layer is the concept of a tile tensor: an innovative data structure that allows packing arbitrary tensors (multi-dimensional arrays widely used in AI) in a variety of ways. In addition, this layer contains a generic bootstrapping algorithm for the CKKS scheme. It can work with any CKKS implementation, though currently tested only with SEAL. The layer further contains innovative algorithms for computing functions not directly supported by the underlying HE libraries, such as sign(x), abs(x), and compare(x,y).
Main classes#
Class |
Description |
|---|---|
Some libraries and scheme support manipulation on individual bits of the encrypted numbers. This class contains API based on such manipulation: extract MSB, split to bits, and others. When using a library/scheme that does not support bit level manipulation, an exception will be thrown. The HeTraits class can be queried for the availability of this functionality. |
|
Efficient algorithms for evaluation of powers, polynomial evaluation, and some approximations for useful functions such as sigmoid, absolute, sign, and others. Note: If the underlying HE library supports some of these functions, and the library’s implementation is more efficient than the SDK’s, the calls will be delegated to the library’s functions. |
To learn more about Tile tensors see Tile Tensors