Jeremy Kun comments

Results 329 comments of


                                            Jeremy Kun

trafficstars

Port remaining HECO evaluation artifacts and ensure they produce equivalently optimal results.

Roberts cross had a bug in the input MLIR that is fixed in https://github.com/google/heir/pull/676

Port remaining HECO evaluation artifacts and ensure they produce equivalently optimal results.

> > So remaining work is: > > > > * Improve box_blur > > I think this is the last open item in this issue, right? According to the...

Port remaining HECO evaluation artifacts and ensure they produce equivalently optimal results.

Closing and removing remaining work to https://github.com/google/heir/issues/758

Support encoding short tensors into BGV ciphertexts of restricted dimension

The solution may also be related to https://github.com/google/heir/issues/646, since in that ticket a single scalar value can be encoded by repetition.

Generate a helper that configures OpenFHE crypto context based on the IR content

I don't necessarily think you should use IntRangeAnalysis directly for this. If I were you I would write a new analysis that uses the same base framework (I'm mobile so...

Add initial poly approximation for "ReLU via tosa.clamp"

Note to self: since ReLU is recognized via clamp here, maybe instead try clamp(MIN, VAL, MAX) = max(MIN, min(VAL, MAX)) since lolremez supports max and min.

Add initial poly approximation for "ReLU via tosa.clamp"

Yeah this is way old now

Implement initial ReLU/sign function via polynomial approximation

Also cf. https://openreview.net/pdf?id=Hq16Jk2bVlp and https://eprint.iacr.org/2021/1688 which use these approximations.

Implement initial ReLU/sign function via polynomial approximation

The paper linked above 2020/834 does not release source code, but there is a reference implementation in LattiGo: https://github.com/tuneinsight/lattigo/blob/4cce9a48c1daaa2dd122921822f5ad70cd444156/he/hefloat/minimax_composite_polynomial.go#L124

Implement initial ReLU/sign function via polynomial approximation

The paper https://eprint.iacr.org/2019/1234 is a precursor to https://eprint.iacr.org/2020/834, but also seems to explain more of the motivation behind the composite polynomials.