qianlou / she Goto Github PK

I just find the shiftDotProduction, ReLU, and max pooling operations code, but where is the source code of NN models of MNIST，CIFAR-10 ?

When I clone the repository and follow the execution instructions, the program runs correctly.

However, whenever I try to execute the program after making some changes in SHE.cpp file and following the execution instruction again, the program terminates abruptly by giving a segmentation fault.

For example, I get the following output when running the program after some changes:

######## 1. shiftDot(A[0:input_size-1], Be[0:input_size-1]) Verification#######
A=[ 0 1 2 3 4 5 6 7]
Be=[ 1 1 1 1 1 1 1 1]
Segmentation fault (core dumped)

I could not understand why this is happening. Can someone point me in the right direction?

Hi,

From SHE paper, the dot product is mentioned as summation(input_i×2^WeightQi).
It seems the weight is assumed to be always positive.

How do you handle weights which are negative (e.g. -0.25 = -(2^-2))?

Thanks.

the function uses (deleted function) =operator
maybe update

*result = *a;

to

result = a;

Hi,

When i edit line 119 in https://github.com/qianlou/SHE/blob/master/src/SHE.cpp to "0", the value of "Max_plain_result" variable is still -4. Any idea?

Thanks.

I cannot find code regard of leveled TFHE that mentioned in your paper.
It seems like your code used bootstrapping gates such as bootsAND.
Moreover, it seems like TFHE library does not support Leveled TFHE.
Thus, would you give any information of LTFHE?
Thanks.

Hi,

When i edit line 93 of https://github.com/qianlou/SHE/blob/master/src/SHE.cpp to -1, below is the difference in result between plain and decrypted:
Decrypted Result:16
Result:0

Any idea?

Thanks.