How to implement a seclist of seclist/list

[1191000814]

I want to create a seclist L which contains elements for seclist/list, because I want to access L with secure index, but it may be impossible because it make error when I get item of the seclist with sec_int index. So how can I do in other ways?

[lschoe]

Indeed, the elements of a seclist are currently required to be a secure number (secure type mpc.SecFld(), mpc.SecInt(), or mpc.SecFxp()).

If you want a nested secure list, say a 2D list, where the inner lists are all of the same length n, you may try to convert 2D indices (i,j) to n*i+j to flatten the secure list of lists.

[1191000814]

I use the 2d array for a adjacency matrix of graph. But sometimes I want to compute the degree of a node, that is to calculate the sum of a row(the row num is a secure int) of a 2d matrix, how can I get it?

[lschoe]

Well, I would first compute the row sums of the adjacency matrix, and then obliviously select the row sum that you want. Something like

 d = seclist([sum(A[n*i:n*(i+1)]) for i in range(n)])[r]

where r is a secure integer holding the secret row number, and A is the (flat) seclist representing the graph.

[1191000814]

Hello, What does the attribute l (bit_length) of SecInt represent? Why can it still represent a large number when I set l=1? In addition, can setting a smaller l value reduce decoding time? Because I want to make my program computing with a large dataset faster.

[lschoe]

For mpc.Secint(l), the parameter l represent the bit length for secure (signed) integers. This allows for secure integer arithmetic with correct results provided the secret-shared integer values remain between $-2^{l-1}$ and $2^{l -1}$. If the secret-shared values get outside this range, some programs will still work because of the so-called headroom used with secure integers. The headroom is needed for secure comparisons and other operations: these operations will potentially give wrong results if the values are outside the above range.

By using smaller bit lengths l the performance of your program will improve. You can test to see at which l things start to fail. And then try to analyze why your program should still work around l+1, say.

[1191000814]

Hi, if I use ls = sec_int.array(np.array([1, 2, 3, 4])) to create a secure array, How can I get a value with a secure index like ls[sec_int(1)](I know this causes a error), So How to get it ?

[lschoe]

Secure indexing is only supported for seclists, not for secure Numpy arrays. Using np_tolist()and np_fromlist() you can convert between secure Numpy arrays and list of secure integers, and then also convert to seclists.

[1191000814]

Hi again! If I want to use mpyc on multiple machines, rather than multiple terminals on one machine, how can I do it?

[lschoe]

Hi, for that check the documentation, e.g., at https://mpyc.readthedocs.io/en/latest/demos.html and search for "-P" to find examples on how to do this.