Recently, knowledge distillation (KD) has shown great success in BERT compression. Instead of only learning from the teacher’s soft label as in conventional KD, researchers find that the rich information contained in the hidden layers of BERT is conducive to the student’s performance. To better exploit the hidden knowledge, a common practice is to force the student to deeply mimic the teacher’s hidden states of all the tokens in a layer-wise manner. In this paper, however, we observe that although distilling the teacher’s hidden state knowledge (HSK) is helpful, the performance gain (marginal utility) diminishes quickly as more HSK is distilled. To understand this effect, we conduct a series of analysis. Specifically, we divide the HSK of BERT into three dimensions, namely depth, length and width. We first investigate a variety of strategies to extract crucial knowledge for each single dimension and then jointly compress the three dimensions. In this way, we show that 1) the student’s performance can be improved by extracting and distilling the crucial HSK, and 2) using a tiny fraction of HSK can achieve the same performance as extensive HSK distillation. Based on the second finding, we further propose an efficient KD paradigm to compress BERT, which does not require loading the teacher during the training of student. For two kinds of student models and computing devices, the proposed KD paradigm gives rise to training speedup of 2.7x 3.4x.

Residual has been widely applied to build deep neural networks with enhanced feature propagation and improved accuracy. In the literature, multiple variants of residual structure are proposed. However, most of them are manually designed for particular tasks and datasets and the combination of existing residual structures has not been well studied. In this work, we propose the Self-Adaptive Scaling (SAS) approach that automatically learns the design of residual structure from data. The proposed approach makes the best of various residual structures, resulting in a general architecture covering several existing ones. In this manner, we construct a learnable residual structure which can be easily integrated into a wide range of residual-based models. We evaluate our approach on various tasks concerning different modalities, including machine translation (IWSLT-2015 EN-VI and WMT-2014 EN-DE, EN-FR), image classification (CIFAR-10 and CIFAR-100), and image captioning (MSCOCO). Empirical results show that the proposed approach consistently improves the residual-based models and exhibits desirable generalization ability. In particular, by incorporating the proposed approach to the Transformer model, we establish new state-of-the-arts on the IWSLT-2015 EN-VI low-resource machine translation dataset.

Open-domain question answering (OpenQA) aims to answer questions based on a number of unlabeled paragraphs. Existing approaches always follow the distantly supervised setup where some of the paragraphs are wrong-labeled (noisy), and mainly utilize the paragraph-question relevance to denoise. However, the paragraph-paragraph relevance, which may aggregate the evidence among relevant paragraphs, can also be utilized to discover more useful paragraphs. Moreover, current approaches mainly focus on the positive paragraphs which are known to contain the answer during training. This will affect the generalization ability of the model and make it be disturbed by the similar but irrelevant (distracting) paragraphs during testing. In this paper, we first introduce a ranking model leveraging the paragraph-question and the paragraph-paragraph relevance to compute a confidence score for each paragraph. Furthermore, based on the scores, we design a modified weighted sampling strategy for training to mitigate the influence of the noisy and distracting paragraphs. Experiments on three public datasets (Quasar-T, SearchQA and TriviaQA) show that our model advances the state of the art.

The encode-decoder framework has shown recent success in image captioning. Visual attention, which is good at detailedness, and semantic attention, which is good at comprehensiveness, have been separately proposed to ground the caption on the image. In this paper, we propose the Stepwise Image-Topic Merging Network (simNet) that makes use of the two kinds of attention at the same time. At each time step when generating the caption, the decoder adaptively merges the attentive information in the extracted topics and the image according to the generated context, so that the visual information and the semantic information can be effectively combined. The proposed approach is evaluated on two benchmark datasets and reaches the state-of-the-art performances.