Conv1d

1D Convolutional Layer

Applies 1D convolution over an input signal (sequences, time series, audio).

Parameters:

• in_channels: Number of channels in input
• out_channels: Number of channels produced by convolution
• kernel_size: Size of the convolving kernel
• stride: Stride of the convolution (default: 1)
• padding: Zero-padding added to both sides (default: 0)
• dilation: Spacing between kernel elements (default: 1)
• groups: Number of blocked connections (default: 1)
• bias: If true, adds learnable bias (default: true)

Shape Contract:

• Input: [batch, in_channels, length]
• Output: [batch, out_channels, out_length]

Notes:

• Output length: (length + 2padding - dilation(kernel_size-1) - 1) / stride + 1
• Used for sequence modeling, audio processing, time series
• WaveNet, TCN, and other temporal architectures

Signature

neuron Conv1d(in_channels, out_channels, kernel_size, stride=Int(1), padding=Int(0), dilation=Int(1), groups=Int(1), bias=Bool(true))

Ports

Inputs:

• default: [batch, in_channels, length]

Outputs:

• default: [batch, out_channels, *]

Implementation

Source { source: "core", path: "convolutions/Conv1d" }