Using CKKS scheme, replace rotation with fast rotation, keep other codes unchanged, decrypt failed as following:
terminate called after throwing an instance of ‘lbcrypto::math_error’
what(): /home/xxx/tools/openfhe-development/src/pke/lib/encoding/ckkspackedencoding.cpp:520 The decryption failed because the approximation error is too high. Check the parameters.
Aborted
Why does this happen? And how to process this?
Please post the full example so we could recreate the issue and confirm that fast rotation is used correctly.
I have found the problem. The reason is i used the fast rotation with changed ciphertext.
Thank you all the same.
Hi, I have a similar problem. Is it because of automorphism before key switching that the noise is too high? But automorphism is the transformation of slot, why automorphism before key switching will cause too much noise, but automorphism after key switching can decrypt correctly?
If the problem is the same as reported by @Knightley, then it is not related to noise. Fast (hoisted) rotations are possible when you need multiple rotations for the same ciphertext (by design). In this case, an expensive digit decomposition step can be precomputed/reused for multiple rotations. If your problem is different, please create a new topic and describe there the problem in detail, including a minimal working example.