2 ms·
That is a good way to think about it. I am really encourage by new non-sequential architectures, multiple input and output channels, etc., and how these archite
by mark_l_watson 9y ago
That is a good way to think about it. I am really encourage by new non-sequential architectures, multiple input and output channels, etc., and how these architectures can be expressed in functional Keras. It seems like building with something like functional composition is another paradigm shift. BTW, thanks for your writing, especially about RNNs: really useful when the ordering of sequential data is important. I use what I have learned from reading your blogs and papers literally every day at work.