Is it Time to Swish? Comparing Deep Learning Activation Functions Across NLP tasks
This work addresses the selection of activation functions for NLP practitioners, but it is incremental as it focuses on empirical comparison rather than introducing new methods.
The authors conducted a large-scale comparison of 21 activation functions across eight NLP tasks, finding that the penalized tanh function performed most stably and improved LSTM gates by 2 percentage points on a challenging task.
Activation functions play a crucial role in neural networks because they are the nonlinearities which have been attributed to the success story of deep learning. One of the currently most popular activation functions is ReLU, but several competitors have recently been proposed or 'discovered', including LReLU functions and swish. While most works compare newly proposed activation functions on few tasks (usually from image classification) and against few competitors (usually ReLU), we perform the first large-scale comparison of 21 activation functions across eight different NLP tasks. We find that a largely unknown activation function performs most stably across all tasks, the so-called penalized tanh function. We also show that it can successfully replace the sigmoid and tanh gates in LSTM cells, leading to a 2 percentage point (pp) improvement over the standard choices on a challenging NLP task.