How can I build a convolutional layer using the BFV scheme?

[reaneling]

Question

Hello!

I am researching convolutional neural networks on various homomorphic encryption schemes. I noticed that you have support for the BFV scheme. However, I cannot build a convolutional layer and I can find a similar command "im2col_encoding" for BFV too.
Is it possible in principle to build CNN-BFV using your library and how to do it?

Further Information

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In [36], [line 2](vscode-notebook-cell:?execution_count=36&line=2)
      [1](vscode-notebook-cell:?execution_count=36&line=1) enc_model = EncConvNet(model)
----> [2](vscode-notebook-cell:?execution_count=36&line=2) enc_test(context, enc_model, test_loader, criterion, kernel_shape, stride)

Cell In [32], [line 49](vscode-notebook-cell:?execution_count=32&line=49)
     [45](vscode-notebook-cell:?execution_count=32&line=45) class_total = list(0. for i in range(10))
     [47](vscode-notebook-cell:?execution_count=32&line=47) for data, target in test_loader:
     [48](vscode-notebook-cell:?execution_count=32&line=48)     # Encoding and encryption
---> [49](vscode-notebook-cell:?execution_count=32&line=49)     x_enc, windows_nb = ts.im2col_encoding(
     [50](vscode-notebook-cell:?execution_count=32&line=50)         context, data.view(28, 28).tolist(), kernel_shape[0],
     [51](vscode-notebook-cell:?execution_count=32&line=51)         kernel_shape[1], stride
     [52](vscode-notebook-cell:?execution_count=32&line=52)     )
     [53](vscode-notebook-cell:?execution_count=32&line=53)     # Encrypted evaluation
     [54](vscode-notebook-cell:?execution_count=32&line=54)     enc_output = enc_model(x_enc, windows_nb)

File ~/src/tenseal/tenseal/__init__.py:37, in im2col_encoding(context, tensor, kernel_n_rows, kernel_n_cols, stride)
     [34](https://file+.vscode-resource.vscode-cdn.net/Users/reaneling/BFV/~/src/tenseal/tenseal/__init__.py:34)         raise ValueError("tensor must be a matrix")
     [35](https://file+.vscode-resource.vscode-cdn.net/Users/reaneling/BFV/~/src/tenseal/tenseal/__init__.py:35)     matrix = tensor.tolist()
---> [37](https://file+.vscode-resource.vscode-cdn.net/Users/reaneling/BFV/~/src/tenseal/tenseal/__init__.py:37) ckks_vec, windows_nb = _ts_cpp.im2col_encoding(
     [38](https://file+.vscode-resource.vscode-cdn.net/Users/reaneling/BFV/~/src/tenseal/tenseal/__init__.py:38)     context.data, matrix, kernel_n_cols, kernel_n_rows, stride
     [39](https://file+.vscode-resource.vscode-cdn.net/Users/reaneling/BFV/~/src/tenseal/tenseal/__init__.py:39) )
     [40](https://file+.vscode-resource.vscode-cdn.net/Users/reaneling/BFV/~/src/tenseal/tenseal/__init__.py:40) return CKKSVector._wrap(ckks_vec), windows_nb

ValueError: no global scale

System Information

• OS: MacOS
• OS Version: 15.0
• Language Version: Python 3.9.13
• Package Manager Version: pip 24.2