Continuation KD: Improved Knowledge Distillation through the Lens of Continuation Optimization
This addresses limitations in knowledge distillation for improving small models in natural language understanding and computer vision, representing an incremental advancement.
The paper tackles the problems of capacity gap and teacher noise in knowledge distillation by proposing Continuation-KD, a method inspired by continuation optimization that smooths the training objective. It achieves state-of-the-art performance on NLU and computer vision benchmarks, with improvements such as a 1.2% average gain on GLUE.
Knowledge Distillation (KD) has been extensively used for natural language understanding (NLU) tasks to improve a small model's (a student) generalization by transferring the knowledge from a larger model (a teacher). Although KD methods achieve state-of-the-art performance in numerous settings, they suffer from several problems limiting their performance. It is shown in the literature that the capacity gap between the teacher and the student networks can make KD ineffective. Additionally, existing KD techniques do not mitigate the noise in the teacher's output: modeling the noisy behaviour of the teacher can distract the student from learning more useful features. We propose a new KD method that addresses these problems and facilitates the training compared to previous techniques. Inspired by continuation optimization, we design a training procedure that optimizes the highly non-convex KD objective by starting with the smoothed version of this objective and making it more complex as the training proceeds. Our method (Continuation-KD) achieves state-of-the-art performance across various compact architectures on NLU (GLUE benchmark) and computer vision tasks (CIFAR-10 and CIFAR-100).