March 17, 2021
Large-scale language models have recently demonstrated impressive empirical performance. Nevertheless, the improved results are attained at the price of bigger models, more power consumption, and slower inference, which hinder their applicability to low-resource (both memory and computation) platforms. Knowledge distillation (KD) has been demonstrated as an effective framework for compressing such big models. However, large-scale neural network systems are prone to memorize training instances, and thus tend to make inconsistent predictions when the data distribution is altered slightly. Moreover, the student model has few opportunities to request useful information from the teacher model when there is limited task-specific data available. To address these issues, we propose MixKD, a data-agnostic distillation framework that leverages mixup, a simple yet efficient data augmentation approach, to endow the resulting model with stronger generalization ability. Concretely, in addition to the original training examples, the student model is encouraged to mimic the teacher's behavior on the linear interpolation of example pairs as well. We prove from a theoretical perspective that under reasonable conditions MixKD gives rise to a smaller gap between the generalization error and the empirical error. To verify its effectiveness, we conduct experiments on the GLUE benchmark, where MixKD consistently leads to significant gains over the standard KD training, and outperforms several competitive baselines. Experiments under a limited-data setting and ablation studies further demonstrate the advantages of the proposed approach.
Written by
Kevin J Liang
Weituo Hao
Dinghan Shen
Yufan Zhou
Weizhu Chen
Changyou Chen
Lawrence Carin
Publisher
ICLR 2021
Research Topics
November 16, 2022
Kushal Tirumala, Aram H. Markosyan, Armen Aghajanyan, Luke Zettlemoyer
November 16, 2022
October 31, 2022
Fabio Petroni, Giuseppe Ottaviano, Michele Bevilacqua, Patrick Lewis, Scott Yih, Sebastian Riedel
October 31, 2022
December 06, 2020
Michael Lewis, Armen Aghajanyan, Gargi Ghosh, Luke Zettlemoyer, Marjan Ghazvininejad, Sida Wang
December 06, 2020
November 30, 2020
Dhruv Batra, Devi Parikh, Meera Hahn, Jacob Krantz, James Rehg, Peter Anderson, Stefan Lee
November 30, 2020
April 30, 2018
Yedid Hoshen, Lior Wolf
April 30, 2018
November 01, 2018
Yedid Hoshen, Lior Wolf
November 01, 2018
December 02, 2018
Sagie Benaim, Lior Wolf
December 02, 2018
June 30, 2019
Geng Ji, Dehua Cheng, Huazhong Ning, Changhe Yuan, Hanning Zhou, Liang Xiong, Erik B. Sudderth
June 30, 2019
Foundational models
Latest news
Foundational models