What architecture/methods are used to make a neural network which can get infinite big input and/or return infinite big output?
I have an idea how to make infinite big output. I just need extra input neurons and after the first calculation send output (or part of it) to input neurons.
But I have no clue how to make extensible input. Maybe use multiple iterations, and plug output to input, and change the rest of the input neurons accordingly to the next portion of input data?
Artificial intelligence is new for me, so it is possible that I'm asking something that I don't want or something impossible. Please provide simple answers.
The short answer is any RNN is capable of consuming, and producing, arbitrary length sequences. Depending on the structure of the data CNNs, Graph Nets etc. can also work with arbitrarily large inputs.