1D Convolution

• Slide a kernel (filter) over a single dimension of the input data
• Typically used with sequential data (e.g., time series, audio signals)

▲ Convolution in one dimension for neural networks by Brandon Rohrer [video]

1 x 1 Convolution

• A special case of 2D convolution where the kernel size is 1x1
• Operates on each individual pixel across the channel dimension
• The network learns channel-wise interactions
• Reduces the number of channels while preserving spatial dimensions (reduce computation, serve as bottleneck layers)

▲ Network In Network by Andrew Ng