Quaternion Recurrent Neural Networks
This work addresses the challenge of handling internal dependencies in multi-dimensional features for tasks like speech recognition, offering a more compact and efficient model, though it is incremental as it builds on existing RNN and LSTM architectures.
The authors tackled the problem of modeling multi-dimensional sequential data with internal dependencies by proposing quaternion recurrent neural networks (QRNN) and quaternion LSTMs (QLSTM), which achieved better performance in automatic speech recognition and reduced the number of free parameters by up to 3.3x compared to real-valued RNNs and LSTMs.
Recurrent neural networks (RNNs) are powerful architectures to model sequential data, due to their capability to learn short and long-term dependencies between the basic elements of a sequence. Nonetheless, popular tasks such as speech or images recognition, involve multi-dimensional input features that are characterized by strong internal dependencies between the dimensions of the input vector. We propose a novel quaternion recurrent neural network (QRNN), alongside with a quaternion long-short term memory neural network (QLSTM), that take into account both the external relations and these internal structural dependencies with the quaternion algebra. Similarly to capsules, quaternions allow the QRNN to code internal dependencies by composing and processing multidimensional features as single entities, while the recurrent operation reveals correlations between the elements composing the sequence. We show that both QRNN and QLSTM achieve better performances than RNN and LSTM in a realistic application of automatic speech recognition. Finally, we show that QRNN and QLSTM reduce by a maximum factor of 3.3x the number of free parameters needed, compared to real-valued RNNs and LSTMs to reach better results, leading to a more compact representation of the relevant information.