Jeremy Kun

Since the research synthesis part is done, moving to more focused tickets like https://github.com/google/heir/issues/1734

I would guess this should be the very first pass, but I'm not sure. What we also need is a way to connect a single "input vector" to its representation...

Labeling this issue for the remaining TODO

TODO - infer max noise as best I can - test

Putting this on hold for a few more days while I work on other priorities. Notes to self: - Last thing I did was replace the "optional" version of the...

 I think the above formula (without the FFT modification) is the right baseline for the noise reset after a bootstrap

Also: they pointed me to the implementation of the concrete parameter selection optimizer https://github.com/zama-ai/concrete/tree/main/compilers/concrete-optimizer

These LUTs correspond to the following operations, where the input bits `a, b, c` are combined into a single value via concatenating their bits, or equivalently `input = 4*c +...

For fun, I tried asking GPT4 to interpret the third LUT, and it was unable to.

Combing the tfhe-rs codebase, I didn't find anything existing that handles these without a bootstrap. Moreover, the old TFHE API also seems to require a bootstrap for every boolean gate,...