A matrix challenge for breaking obfuscation (proposed by Colin)
Let , and let be secret matrices with each entry chosen uniformly at random from . Suppose we are given the polynomial as a list of coefficients. Help me recover , in the following sense – find matrices with entries in , bit-complexity not too high, such that there exist with for each .
Motivation
Solving this problem would make progress toward breaking a proposed obfuscation scheme for affine determinant programs (ADP) (the paper has more information on both ADP and the scheme).
For us, an affine determinant program is a collection of square matrices over ; to evaluate the program on an input , one simply computes the determinant
The main idea of the obfuscation scheme is as follows:
- Lift the matrices over to matrices over , with small integer coefficients. (In other words: think of as a matrix of zeros and ones, and add a randomly chosen small even integer to each entry.)
- Choose some very large odd number , and think of the matrices as matrices over .
- Choose two random matrices and of determinant over , and publish for each .
Given the matrices , anyone (evaluator or attacker) can compute the value of the integer determinant (at least in the range where that determinant is less than ). What can an attacker do with black-box access to the determinant? A solution to this problem would let the attacker recover the matrices , which would almost certainly render this obfuscation scheme useless.