arXiv subjects

Bei Li

Publications and source records attributed to Bei Li.

80 records · Page 5Linked to original sources

arXiv2021

Weight Distillation: Transferring the Knowledge in Neural Network Parameters

Knowledge distillation has been proven to be effective in model acceleration and compression. It allows a small network to learn to generalize in the same way as a large network. Recent successes in pre-training suggest the effectiveness of transferring model parameters. Inspired by this, we investigate methods of model acceleration and compression in another line of research. We propose Weight Distillation to transfer the knowledge in the large network parameters through a parameter generator. Our experiments on WMT16 En-Ro, NIST12 Zh-En, and WMT14 En-De machine translation tasks show that weight distillation can train a small network that is 1.88~2.94x faster than the large network but with competitive performance. With the same sized small network, weight distillation can outperform knowledge distillation by 0.51~1.82 BLEU points.

cs.CL↗

arXiv2021

ODE Transformer: An Ordinary Differential Equation-Inspired Model for Neural Machine Translation

It has been found that residual networks are an Euler discretization of solutions to Ordinary Differential Equations (ODEs). In this paper, we explore a deeper relationship between Transformer and numerical methods of ODEs. We show that a residual block of layers in Transformer can be described as a higher-order solution to ODEs. This leads us to design a new architecture (call it ODE Transformer) analogous to the Runge-Kutta method that is well motivated in ODEs. As a natural extension to Transformer, ODE Transformer is easy to implement and parameter efficient. Our experiments on three WMT tasks demonstrate the genericity of this model, and large improvements in performance over several strong baselines. It achieves 30.76 and 44.11 BLEU scores on the WMT'14 En-De and En-Fr test data. This sets a new state-of-the-art on the WMT'14 En-Fr task.

cs.CL↗

arXiv2020

Learning Light-Weight Translation Models from Deep Transformer

Recently, deep models have shown tremendous improvements in neural machine translation (NMT). However, systems of this kind are computationally expensive and memory intensive. In this paper, we take a natural step towards learning strong but light-weight NMT systems. We proposed a novel group-permutation based knowledge distillation approach to compressing the deep Transformer model into a shallow model. The experimental results on several benchmarks validate the effectiveness of our method. Our compressed model is 8X shallower than the deep model, with almost no loss in BLEU. To further enhance the teacher model, we present a Skipping Sub-Layer method to randomly omit sub-layers to introduce perturbation into training, which achieves a BLEU score of 30.63 on English-German newstest2014. The code is publicly available at https://github.com/libeineu/GPKD.

cs.CL↗

arXiv2020

Shallow-to-Deep Training for Neural Machine Translation

Deep encoders have been proven to be effective in improving neural machine translation (NMT) systems, but training an extremely deep encoder is time consuming. Moreover, why deep models help NMT is an open question. In this paper, we investigate the behavior of a well-tuned deep Transformer system. We find that stacking layers is helpful in improving the representation ability of NMT models and adjacent layers perform similarly. This inspires us to develop a shallow-to-deep training method that learns deep models by stacking shallow models. In this way, we successfully train a Transformer system with a 54-layer encoder. Experimental results on WMT'16 English-German and WMT'14 English-French translation tasks show that it is $1.4$ $\times$ faster than training from scratch, and achieves a BLEU score of $30.33$ and $43.29$ on two tasks. The code is publicly available at https://github.com/libeineu/SDT-Training/.

cs.CL↗

arXiv2020

A multilayer interstitial fluid flow along vascular adventitia

Objective: Interstitial fluid flow through vascular adventitia has been disclosed recently. However, its kinetic pattern was unclear. Methods and Results: We used histological and topographical identifications to observe ISF flow along venous vessels in rabbits. By MRI in alive subjects, the inherent ISF flow pathways in legs, abdomen and thorax were enhanced by paramagnetic contrast from ankle dermis. By fluorescence stereomicroscopy and layer-by-layer dissection after the rabbits were sacrificed, the perivascular and adventitial connective tissues (PACT) along the saphenous veins and inferior vena cava were found to be stained by sodium fluorescein from ankle dermis, which coincided with the findings by MRI. By confocal microscopy and histological analysis, the stained PACT pathways were verified to be the fibrous connective tissues and consisted of longitudinally assembled fibers. By usages of nanoparticles and surfactants, a PACT pathway was found to be accessible for a nanoparticle under 100nm and contain two parts: a tunica channel part and an absorptive part. In real-time observations, the calculated velocity of a continuous ISF flow along fibers of a PACT pathway was 3.6-15.6 mm/sec. Conclusion: These data further revealed more kinetic features of a continuous ISF flow along vascular vessel. A multiscale, multilayer, and multiform interstitial/interfacial fluid flow throughout perivascular and adventitial connective tissues was suggested as one of kinetic and dynamic mechanisms for ISF flow, which might be another principal fluid dynamic pattern besides convective/vascular and diffusive transport in biological system.

q-bio.TO↗

arXiv2020

Does Multi-Encoder Help? A Case Study on Context-Aware Neural Machine Translation

In encoder-decoder neural models, multiple encoders are in general used to represent the contextual information in addition to the individual sentence. In this paper, we investigate multi-encoder approaches in documentlevel neural machine translation (NMT). Surprisingly, we find that the context encoder does not only encode the surrounding sentences but also behaves as a noise generator. This makes us rethink the real benefits of multi-encoder in context-aware translation - some of the improvements come from robust training. We compare several methods that introduce noise and/or well-tuned dropout setup into the training of these encoders. Experimental results show that noisy training plays an important role in multi-encoder-based NMT, especially when the training data is small. Also, we establish a new state-of-the-art on IWSLT Fr-En task by careful use of noise generation and dropout methods.

cs.CL↗

arXiv2019

Learning Deep Transformer Models for Machine Translation

Transformer is the state-of-the-art model in recent machine translation evaluations. Two strands of research are promising to improve models of this kind: the first uses wide networks (a.k.a. Transformer-Big) and has been the de facto standard for the development of the Transformer system, and the other uses deeper language representation but faces the difficulty arising from learning deep networks. Here, we continue the line of research on the latter. We claim that a truly deep Transformer model can surpass the Transformer-Big counterpart by 1) proper use of layer normalization and 2) a novel way of passing the combination of previous layers to the next. On WMT'16 English- German, NIST OpenMT'12 Chinese-English and larger WMT'18 Chinese-English tasks, our deep system (30/25-layer encoder) outperforms the shallow Transformer-Big/Base baseline (6-layer encoder) by 0.4-2.4 BLEU points. As another bonus, the deep model is 1.6X smaller in size and 3X faster in training than Transformer-Big.

cs.CL↗

arXiv2018

Analyzing the Weighted Nuclear Norm Minimization and Nuclear Norm Minimization based on Group Sparse Representation

Rank minimization methods have attracted considerable interest in various areas, such as computer vision and machine learning. The most representative work is nuclear norm minimization (NNM), which can recover the matrix rank exactly under some restricted and theoretical guarantee conditions. However, for many real applications, NNM is not able to approximate the matrix rank accurately, since it often tends to over-shrink the rank components. To rectify the weakness of NNM, recent advances have shown that weighted nuclear norm minimization (WNNM) can achieve a better matrix rank approximation than NNM, which heuristically set the weight being inverse to the singular values. However, it still lacks a sound mathematical explanation on why WNNM is more feasible than NNM. In this paper, we propose a scheme to analyze WNNM and NNM from the perspective of the group sparse representation. Specifically, we design an adaptive dictionary to bridge the gap between the group sparse representation and the rank minimization models. Based on this scheme, we provide a mathematical derivation to explain why WNNM is more feasible than NNM. Moreover, due to the heuristical set of the weight, WNNM sometimes pops out error in the operation of SVD, and thus we present an adaptive weight setting scheme to avoid this error. We then employ the proposed scheme on two low-level vision tasks including image denoising and image inpainting. Experimental results demonstrate that WNNM is more feasible than NNM and the proposed scheme outperforms many current state-of-the-art methods.

cs.CV↗