decrypt(circshift(c, gk)) does not work with Julia 1.5.0-beta1

[sloede]

I've tried to run the introductory examples on https://juliacomputing.com/blog/2019/11/22/encrypted-machine-learning.html with Julia 1.5.0-beta1. When I reached the circshift test, Julia threw the following error:

julia> decrypt(circshift(c, gk))
ERROR: MethodError: no method matching decrypt(::CipherText{CKKSEncoding{FixedRational{1099511627776,T} where T},CKKSParams,ToyFHE.NTT.RingElement{ℤ₁₃₂₉₂₂₇₉₉₇₅₆₈₀₈₁₄₅₇₄₀₂₇₀₁₂₀₇₁₀₄₂₄₈₂₅₇/(x¹⁶ + 1),ToyFHE.CRTEncoded{3,Tuple{𝔽₁₀₉₉₅₁₁₆₂₇₈₇₃,𝔽₁₀₉₉₅₁₁₆₂₈₁₆₁,𝔽₁₀₉₉₅₁₁₆₂₈₇₆₉}},StructArrays.StructArray{ToyFHE.CRTEncoded{3,Tuple{𝔽₁₀₉₉₅₁₁₆₂₇₈₇₃,𝔽₁₀₉₉₅₁₁₆₂₈₁₆₁,𝔽₁₀₉₉₅₁₁₆₂₈₇₆₉}},1,Tuple{Array{𝔽₁₀₉₉₅₁₁₆₂₇₈₇₃,1},Array{𝔽₁₀₉₉₅₁₁₆₂₈₁₆₁,1},Array{𝔽₁₀₉₉₅₁₁₆₂₈₇₆₉,1}},Int64}},2})
Closest candidates are:
  decrypt(::PrivKey, ::CipherText{T,P,T1,N} where N where T1 where P<:ToyFHE.SHEShemeParams) where T at /home/mschlott/gdrive/work/code/CryptoSim.jl/ToyFHE.jl/src/rlwe_she.jl:199
  decrypt(::KeyPair, ::Any) at /home/whatever/ToyFHE.jl/src/rlwe_she.jl:217
Stacktrace:
 [1] top-level scope at REPL[21]:1

Here's an MWE:

julia> using ToyFHE

julia> N = 8;

julia> ℛ = NegacyclicRing(2N, (40, 40, 40))
ℤ₁₃₂₉₂₂₇₉₉₇₅₆₈₀₈₁₄₅₇₄₀₂₇₀₁₂₀₇₁₀₄₂₄₈₂₅₇/(x¹⁶ + 1)

julia> params = CKKSParams(ℛ)
CKKS parameters

julia> Tscale = FixedRational{2^40}
FixedRational{1099511627776,T} where T

julia> plain = CKKSEncoding{Tscale}(zero(ℛ))
8-element CKKSEncoding{FixedRational{1099511627776,T} where T} with indices 0:7:
 0.0 + 0.0im
 0.0 + 0.0im
 0.0 + 0.0im
 0.0 + 0.0im
 0.0 + 0.0im
 0.0 + 0.0im
 0.0 + 0.0im
 0.0 + 0.0im

julia> kp = keygen(params)
CKKS key pair

julia> kp.priv
CKKS private key

julia> kp.pub
CKKS public key

julia> foreach(i->plain[i] = i+1, 0:7); plain
8-element CKKSEncoding{FixedRational{1099511627776,T} where T} with indices 0:7:
 1.0 + 0.0im
 2.0 + 0.0im
 3.0 + 0.0im
 4.0 + 0.0im
 5.0 + 0.0im
 6.0 + 0.0im
 7.0 + 0.0im
 8.0 + 0.0im

julia> c = encrypt(kp.pub, plain)
CKKS ciphertext (length 2, encoding CKKSEncoding{FixedRational{1099511627776,T} where T})

julia> gk = keygen(GaloisKey, kp.priv; steps=2)
CKKS galois key (element 25)

julia> decrypt(circshift(c, gk))

Any idea what the problem could be?

[sloede]

By the way, kudos to implementing this package in pure Julia, this has been a great help to kick around a few ideas 👍

[Keno]

Whoops, there's a typo in the blog post, obviously, you need to pass the private key to decrypt it. In your MWE:

julia> decrypt(kp.priv, circshift(c, gk))
8-element CKKSEncoding{FixedRational{1099511627776,T} where T} with indices 0:7:
 -18.506166149368628 + 7.60456192425713im
   40.80057950134403 + 8.450631478608264im
 -20.786099653978496 - 5.443269584828531im
  -16.50816701251562 - 0.3042282041572637im
  -19.25951679089231 - 29.497772481885757im
  21.089374991774747 - 16.74889104138221im
  2.5942574658725626 + 9.605495431441932im
   9.309505110847219 - 15.199594797590228im

I'll fix the blog post. Sorry for missing this issue. In the future, please feel free to just ping me directly on issues here.

[Keno]

Although now that I actually look at the output, the noise obviously took over here. It's been a while since I thought about FHE noise growth, so I'm not quite sure what the issue is.

[Keno]

Hmm works fine with the parameter choices in the test suite: https://github.com/JuliaCrypto/ToyFHE.jl/blob/master/test/ckks_rotate.jl. Not sure whether there's an actual issue here. Unfortunately, if there is, it'd be something subtle about the noise behavior, which is beyond the level of time commitment I have available to look into issues here. That said, hopefully you should be able to use the parameters in the test suite for experimentation if you want.