Convolutional Neural networks are used in supervised learning meaning models are always "set in stone" after training (architecture and paramters) so this might not even be possible, but is there any research done on playing around with the data paths, model size (number of layers) and architecture during runtime, i.e after training is done, for instance creating a model that can be modified online to use less or more recources, skip layers or use portions of the model.

There is some work done recently on creating flexible frameworks for training and designing networks, but that's always "offline". There is also an interesting paper on training two models, "big" and "little" for the same application and using an accuracy/power trade-off policy to deploy one of the two.


1 Answer 1


See this work on a network which uses more or less layers depending to trade off between accuracy and computation time, so that easy instances of a task can be computed quickly and hard instances of a task will make use of more computation.

Also see this work on composing specialized network "modules", and then reconfiguring the modules on the fly whenever a new question is asked. Modules can be arranged in any arbitrary tree configuration and not all modules are necessarily used for every task.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.