Tweeted By @nicola_decao
How do Decisions Emerge across Layers in Neural Models? Interpretation with Differentiable Masking https://t.co/EEhXnHUu6N
— Nicola De Cao (@nicola_decao) May 1, 2020
Learning to mask superfluous input and hidden states reveals the internal processing of DNNs (like BERT and RNNs)! With @iatitov @wilkeraziz @michael_sejr pic.twitter.com/GHlkK9dYpe