I came across a paper that describes its model architecture in the following way.

Our TRIL network is a two-channel network jointly trained to predict the expert’s action given state and the system’s next state transition given state and expert action. The training procedure of TRIL is similar to that of a multi- channel supervised classifier with regularization. Let $\theta_{π_0}$ be the parameters of TRIL and $L_{ce}$ be the cross entropy loss for predicting expert’s action and $L_{mse}$ be the mean squared error loss on predicting next state given current state and the expert’s action

The loss function is given in the following manner $$L(\theta_{\pi_0}) = L_{ce}(a, \pi_0(s)) + \lambda L_{mse}(T_{\pi_0}(s,a),s')$$

TRIL is a dual- channel network that shares certain hidden layers and jointly predicts expert action(a) and state transitions(s’)

I am not sure what a dual channel network means and what does it mean when it is able to jointly predict two outputs ? It seems something similar to a multi-task learning since there is shared hidden layers and different "task" prediction but i am not too sure of that either.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.