Internal representation dynamics and geometry in recurrent neural networks
This work addresses the interpretability gap in RNNs for researchers, though it is incremental as it focuses on a simple task without broad practical applications.
The study investigated how vanilla recurrent neural networks (RNNs) implement classification tasks by analyzing internal dynamics and hidden state geometry, finding that early representations reflect true labels but are not directly accessible to the output, and that dynamics and sequence length are crucial for accuracy even without extra task-relevant information.
The efficiency of recurrent neural networks (RNNs) in dealing with sequential data has long been established. However, unlike deep, and convolution networks where we can attribute the recognition of a certain feature to every layer, it is unclear what "sub-task" a single recurrent step or layer accomplishes. Our work seeks to shed light onto how a vanilla RNN implements a simple classification task by analysing the dynamics of the network and the geometric properties of its hidden states. We find that early internal representations are evocative of the real labels of the data but this information is not directly accessible to the output layer. Furthermore the network's dynamics and the sequence length are both critical to correct classifications even when there is no additional task relevant information provided.