MuBiNN: Multi-Level Binarized Recurrent Neural Network for EEG signal Classification
This work addresses efficiency issues for EEG classification on wearable devices, representing an incremental improvement over existing binarization methods.
The paper tackles the challenge of deploying complex RNNs on wearable devices for EEG classification by proposing a multi-level binarized LSTM that reduces computational delay by 47 times with less than 0.01% accuracy loss compared to full precision LSTMs.
Recurrent Neural Networks (RNN) are widely used for learning sequences in applications such as EEG classification. Complex RNNs could be hardly deployed on wearable devices due to their computation and memory-intensive processing patterns. Generally, reduction in precision leads much more efficiency and binarized RNNs are introduced as energy-efficient solutions. However, naive binarization methods lead to significant accuracy loss in EEG classification. In this paper, we propose a multi-level binarized LSTM, which significantly reduces computations whereas ensuring an accuracy pretty close to the full precision LSTM. Our method reduces the delay of the 3-bit LSTM cell operation 47* with less than 0.01% accuracy loss.