chainer.functions.simplified_dropconnect(x, W, b=None, ratio=0.5, train=True, mask=None, use_batchwise_mask=True)[source]

Linear unit regularized by simplified dropconnect.

Simplified dropconnect drops weight matrix elements randomly with probability ratio and scales the remaining elements by factor 1 / (1 - ratio). It accepts two or three arguments: an input minibatch x, a weight matrix W, and optionally a bias vector b. It computes \(Y = xW^\top + b\).

In testing mode, zero will be used as simplified dropconnect ratio instead of ratio.

Notice: This implementation cannot be used for reproduction of the paper. There is a difference between the current implementation and the original one. The original version uses sampling with gaussian distribution before passing activation function, whereas the current implementation averages before activation.

  • x (Variable or N-dimensional array) – Input variable. Its first dimension n is assumed to be the minibatch dimension. The other dimensions are treated as concatenated one dimension whose size must be N.
  • W (Variable or N-dimensional array) – Weight variable of shape (M, N).
  • b (Variable or N-dimensional array) – Bias variable (optional) of shape (M,).
  • ratio (float) – Dropconnect ratio.
  • train (bool) – If True, executes simplified dropconnect. Otherwise, simplified dropconnect function works as a linear function.
  • mask (None or chainer.Variable or numpy.ndarray or cupy.ndarray) – If None, randomized dropconnect mask is generated. Otherwise, The mask must be (n, M, N) or (M, N) shaped array, and use_batchwise_mask is ignored. Main purpose of this option is debugging. mask array will be used as a dropconnect mask.
  • use_batchwise_mask (bool) – If True, dropped connections depend on each sample in mini-batch.

Output variable.

Return type:


See also


See also

Li, W., Matthew Z., Sixin Z., Yann L., Rob F. (2013). Regularization of Neural Network using DropConnect. International Conference on Machine Learning. URL