IRASL Panel on Interpretability and Robustness in Audio, Speech, and Language
Ideas
Pros and cons
Votes
End-to-End deep learning systems are gaining popularity, but are making neural networks even more "black boxes". Do you think that is possible to build interpretable end-to-end neural networks?
Do you think that in order to create interpretable models (for hypotesis testing and discovery on science research) we should look more in the direction of causal inference?
by Pablo
2
Vote
Ehsan
Hamid
Do you think that in order to create robust models we should look more in the direction of causal inference?
by Olga
2
Vote
Pablo
Sheng
Feature learning (through neural networks) has led to improved performance (and robustness) while sacrificing interpretability. Are there examples in which these two goals are aligned, where improved robustness has been achieved simultaneous with improved interpretability? Or are these necessari...more
2
Vote
Pablo
Titouan P.
Do you think that possible advances in the theoretical analysis of deep learning could be the key to build interpretable machines?
by Mirco
1
Vote
Bhuvana
In which practical application do you think it is more important the development of interpretable machines? Do you think that interpretability is an important element in audio, speech, and language applications? Why?
by Mirco
1
Vote
Anonymous
If you are deploying a system to a production setting there is often a tradeoff between performance, robustness, and interpretability. Given a choice between systems (one that is highly performant, one that is highly robust, and one that is highly interpretable) which would you chose to deploy an...more
by Phillip
0
Vote
Adding new ideas is temporarily disabled by this poll's creator.