Continuation KD: Improved Knowledge Distillation through the Lens of Continuation Optimization

TL;DR

Continuation-KD introduces a continuation optimization approach to knowledge distillation, progressively refining the training process to handle capacity gaps and noisy teacher outputs, leading to improved performance in NLP and vision tasks.

Contribution

It proposes a novel continuation optimization-based training procedure for KD that enhances effectiveness by smoothing the objective and gradually increasing complexity.

Findings

Achieves state-of-the-art results on GLUE benchmark

Outperforms previous KD methods on CIFAR-10 and CIFAR-100

Effectively mitigates teacher noise and capacity gap issues

Abstract

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).