TCN
A sequence modeling network architecture based on dilated causal convolutions. It captures long-range temporal dependencies without introducing future information, offering better parallelism and gradient stability than LSTM-based architectures.
Timeline (last 90 days)
The core design of Temporal Convolutional Networks (TCN) is causal dilated convolution, through which the receptive field grows exponentially with the number of layers via dilation.
Stacking multiple layers of causal convolutions or using dilated convolutions can exponentially expand the receptive field to capture long-range dependencies, a strategy validated by classic architectures such as WaveNet and TCN.
All Facts (2)
The core design of Temporal Convolutional Networks (TCN) is causal dilated convolution, through which the receptive field grows exponentially with the number of layers via dilation.
50%UnverifiedStacking multiple layers of causal convolutions or using dilated convolutions can exponentially expand the receptive field to capture long-range dependencies, a strategy validated by classic architectures such as WaveNet and TCN.
50%