pTensor is a wrapper around the PALISADE library, which is a general-purpose lattice cryptography library. pTensor composes relevant functions from PALISADE to create a numpy-like interface for use in encrypted machine learning.

• Creation of pTensors from either ciphertexts or plaintext messages

  • empty tensor
  • from cipher and cipher's transpose
  • cipher

• Encryption

  • we also take the encrypted transpose to potentially save us from the expensive operation

• Decryption

• Addition

• Subtraction

• Multiplication

• Dot product

  • Supported between Matrix-vector and vector-vector

• Sum

  • all reduce or reducing across specified axes

• Transpose

  • expensive encrypted transpose. We attempt to circumvent this by precomputing the transpose in plaintext if possible

• plainT

  • plaintext transpose

• Identity matrix

• Random tensor

  • uniform
  • normal

• shuffling

  • this is comprised of 2 parts: shuffling n number of (user specified) folds if given messages, then having an adapter that feeds the data.

• hstack

  • re: expensive transpose, we do a vstack on the transpose to get the resulting transpose without actually doing the entire thing

• vstack

  • re: expensive transpose, we do a hstack on the transpose to get the resulting transpose without actually doing the entire thing

This library is pronounced Tensor as the "p" is silent.