Implement Relinearization operation of BFV Scheme #3977
Conversation
rav1kantsingh
added
the
Type: New Feature ➕
|
Check out this pull request on Review Jupyter notebook visual diffs & provide feedback on notebooks. Powered by ReviewNB |
Codecov Report
@@ Coverage Diff @@
## master #3977 +/- ##
==========================================
- Coverage 94.88% 94.76% -0.13%
==========================================
Files 202 203 +1
Lines 20853 21038 +185
==========================================
+ Hits 19786 19936 +150
- Misses 1067 1102 +35
|
| temp_poly_ptr, encrypted_ptr[j], key_mod[j], self.poly_modulus | ||
| ) | ||
|
|
||
| return CipherText(ct[:-1], param_id) |
There was a problem hiding this comment.
I think the current algorithm only apply for length 3 ciphertexts, right? So we can explicitly specify ct[0:2] to make it clear here
| def _get_sk_power_2(self, sk): | ||
| sk_power_2 = [] | ||
| for i in range(len(self._coeff_modulus)): | ||
| sk_power_2.append( | ||
| poly_mul_mod(sk[i], sk[i], self._coeff_modulus[i], self._poly_modulus) | ||
| ) | ||
| return sk_power_2 |
There was a problem hiding this comment.
I think I've seen this method in quite some places, can we factorize it and provide it in utils maybe?
There was a problem hiding this comment.
Yes, it is used in decrypter and here but I modified it from the other one. it only gives the power of 2 and the decryptors calculate any power given to it. I can get this done later because merging both to the same function will require changing indexes at various positions.😅
test/torch/tensors/test_fv.py
Outdated
| @pytest.mark.parametrize( | ||
| "val", [(0), (1), (-1), (100), (1000), (-1000), (-99)], | ||
| ) | ||
| def test_fv_relin(val): | ||
| ctx = Context(EncryptionParams(128, CoeffModulus().create(128, [40, 40]), 64)) | ||
| keygenerator = KeyGenerator(ctx) | ||
| keys = keygenerator.keygen() | ||
| relin_key = keygenerator.get_relin_keys() | ||
| encoder = IntegerEncoder(ctx) | ||
| encryptor = Encryptor(ctx, keys[1]) # keys[1] = public_key | ||
| decryptor = Decryptor(ctx, keys[0]) # keys[0] = secret_key | ||
| evaluator = Evaluator(ctx) | ||
|
|
||
| op = encryptor.encrypt(encoder.encode(val)) | ||
| temp_prod = evaluator.mul(op, op) | ||
| relin_prod = evaluator.relin(temp_prod, relin_key) | ||
| assert len(temp_prod.data) - 1 == len(relin_prod.data) |
There was a problem hiding this comment.
Can we add more tests, ones that do multiple mul and relin, and others that check the failure for len 2 and >3 ciphertexts?
| with pytest.raises(Warning): | ||
| evaluator.relin(op1, relin_key) # Ciphertext size 2 | ||
|
|
||
| with pytest.raises(Exception): | ||
| evaluator.relin(evaluator.mul(temp_prod, val1), relin_key) # Ciphertext size 4 |
There was a problem hiding this comment.
We doesn't need to run this on every values, I think you can separate this into a different test with a single value
rav1kantsingh
force-pushed
the
fv-relinearization-operation
branch
from
415bd7f
to
3b1aa9a
Compare
Description
Closes #3662 #3791
Implementing Relin Key and Relinearization operation for the FV scheme. So that we can decrement the size of ciphertexts if needed. Mainly required after multiplication operation.
Affected Dependencies
None
How has this been tested?
Added tests for the relinearization operation.
Checklist