Sharpness-Aware Minimization Improves Language Model Generalization
This work addresses the generalization challenge for language models, offering a computationally efficient method that is incremental but impactful for NLP applications.
The paper tackled the problem of improving language model generalization through better optimization, showing that Sharpness-Aware Minimization (SAM) substantially boosts performance on tasks like SuperGLUE and GLUE, with particularly large gains when training data is limited.
The allure of superhuman-level capabilities has led to considerable interest in language models like GPT-3 and T5, wherein the research has, by and large, revolved around new model architectures, training tasks, and loss objectives, along with substantial engineering efforts to scale up model capacity and dataset size. Comparatively little work has been done to improve the generalization of these models through better optimization. In this work, we show that Sharpness-Aware Minimization (SAM), a recently proposed optimization procedure that encourages convergence to flatter minima, can substantially improve the generalization of language models without much computational overhead. We show that SAM is able to boost performance on SuperGLUE, GLUE, Web Questions, Natural Questions, Trivia QA, and TyDiQA, with particularly large gains when training data for these tasks is limited.