TCN: Receptive Field
Intermediate
The problem, from first principles
Temporal convolutional networks (Bai et al., 2018) use causal convolutions with exponentially growing dilations. The receptive field grows exponentially with depth, so long histories are covered without recurrence.
From theory to code
Implement tcn_receptive_field(kernel, levels), the receptive field of the stack.
Constraints
- Dilation at level i is two to the i.
Hints
Hint 1
Sum (kernel minus one) times the dilation over the levels, which is a geometric series, then add one.
Keywords
research-paperssequence-modelsconvolutionsequence
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.