What does an LSTM look for in classifying heartbeats?
share
Long short-term memory (LSTM) recurrent neural networks are renowned for being uninterpretable "black boxes". In the medical domain where LSTMs have shown promise, this is specifically concerning because it is imperative to understand the decisions made by machine learning models in such acute situations. This study employs techniques used in the convolutional neural network domain to elucidate the inputs that are important when LSTMs classify electrocardiogram signals. Of the various techniques available to determine input feature saliency, it was found that learning an occlusion mask is the most effective.
READ FULL TEXT