I'm working with the CKKS scheme to perform a calculation involving plaintexts and ciphertexts. It appears that there is a larger accuracy difference to the same cleartext calculations when using 32k vs 16k cyclotomic polynomial.
What could be the reason for it?
The encryption parameters are: scale=2^60, coefficient modulus: {60, 60, 60, 60, 60}.
I tested the calculation using 16k and 32k cyclotomic polynomial separately.
My SEAL version is 4.1.1.
The cleartext vs encrypted comparison results are:
Hi,
I'm working with the CKKS scheme to perform a calculation involving plaintexts and ciphertexts. It appears that there is a larger accuracy difference to the same cleartext calculations when using 32k vs 16k cyclotomic polynomial. What could be the reason for it?
The encryption parameters are: scale=2^60, coefficient modulus: {60, 60, 60, 60, 60}. I tested the calculation using 16k and 32k cyclotomic polynomial separately. My SEAL version is 4.1.1.
The cleartext vs encrypted comparison results are:
testing poly modulus: 16384 enc result: 274895208699.0369873046875 cleartext result: 274895208699 diff: 0.0369873046875
testing poly modulus: 32768 enc result: 274895208700.9400634765625 cleartext result: 274895208699 diff: 1.9400634765625
The code: