Modular Networks: Learning to Decompose Neural Computation
This addresses the challenge of increasing model size without proportional resource growth for deep learning practitioners, representing a novel method rather than an incremental improvement.
The paper tackles the problem of scaling model capacity efficiently by proposing modular networks that learn to decompose neural computation, achieving superior performance in image recognition and language modeling tasks compared to baselines.
Scaling model capacity has been vital in the success of deep learning. For a typical network, necessary compute resources and training time grow dramatically with model size. Conditional computation is a promising way to increase the number of parameters with a relatively small increase in resources. We propose a training algorithm that flexibly chooses neural modules based on the data to be processed. Both the decomposition and modules are learned end-to-end. In contrast to existing approaches, training does not rely on regularization to enforce diversity in module use. We apply modular networks both to image recognition and language modeling tasks, where we achieve superior performance compared to several baselines. Introspection reveals that modules specialize in interpretable contexts.