NeurST: Neural Speech Translation Toolkit

NeurST is an open-source toolkit for neural speech translation developed by ByteDance AI Lab. The toolkit mainly focuses on end-to-end speech translation, which is easy to use, modify, and extend to advanced speech translation research and products. NeurST aims at facilitating the speech translation research for NLP researchers and provides a complete setup for speech translation benchmarks, including feature extraction, data preprocessing, distributed training, and evaluation. Moreover, The toolkit implements several major architectures for end-to-end speech translation. It shows experimental results for different benchmark datasets, which can be regarded as reliable baselines for future research. The toolkit is publicly available at https://github.com/bytedance/neurst.

Distance-wise Graph Contrastive Learning

Contrastive learning (CL) has proven highly effective in graph-based semi-supervised learning (SSL), since it can efficiently supplement the limited task information from the annotated nodes in graph. However, existing graph CL (GCL) studies ignore the uneven distribution of task information across graph caused by the graph topology and the selection of annotated nodes. They apply CL to the whole graph evenly, which results in an incongruous combination of CL and graph learning. To address this issue, we propose to apply CL in the graph learning adaptively by taking the received task information of each node into consideration. Firstly, we introduce Group PageRank to measure the node information gain from graph and find that CL mainly works for nodes that are topologically far away from the labeled nodes. We then propose our Distance-wise Graph Contrastive Learning (DwGCL) method from two views:(1) From the global view of the task information distribution across the graph, we enhance the CL effect on nodes that are topologically far away from labeled nodes; (2) From the personal view of each node's received information, we measure the relative distance between nodes and then we adapt the sampling strategy of GCL accordingly. Extensive experiments on five benchmark graph datasets show that DwGCL can bring a clear improvement over previous GCL methods. Our analysis on eight graph neural network with various types of architecture and three different annotation settings further demonstrates the generalizability of DwGCL.

Trustworthy Preference Completion in Social Choice

As from time to time it is impractical to ask agents to provide linear orders over all alternatives, for these partial rankings it is necessary to conduct preference completion. Specifically, the personalized preference of each agent over all the alternatives can be estimated with partial rankings from neighboring agents over subsets of alternatives. However, since the agents' rankings are nondeterministic, where they may provide rankings with noise, it is necessary and important to conduct the trustworthy preference completion. Hence, in this paper firstly, a trust-based anchor-kNN algorithm is proposed to find $k$-nearest trustworthy neighbors of the agent with trust-oriented Kendall-Tau distances, which will handle the cases when an agent exhibits irrational behaviors or provides only noisy rankings. Then, for alternative pairs, a bijection can be built from the ranking space to the preference space, and its certainty and conflict can be evaluated based on a well-built statistical measurement Probability-Certainty Density Function. Therefore, a certain common voting rule for the first $k$ trustworthy neighboring agents based on certainty and conflict can be taken to conduct the trustworthy preference completion. The properties of the proposed certainty and conflict have been studied empirically, and the proposed approach has been experimentally validated compared to state-of-arts approaches with several data sets.

Reciprocal Supervised Learning Improves Neural Machine Translation

Despite the recent success on image classification, self-training has only achieved limited gains on structured prediction tasks such as neural machine translation (NMT). This is mainly due to the compositionality of the target space, where the far-away prediction hypotheses lead to the notorious reinforced mistake problem. In this paper, we revisit the utilization of multiple diverse models and present a simple yet effective approach named Reciprocal-Supervised Learning (RSL). RSL first exploits individual models to generate pseudo parallel data, and then cooperatively trains each model on the combined synthetic corpus. RSL leverages the fact that different parameterized models have different inductive biases, and better predictions can be made by jointly exploiting the agreement among each other. Unlike the previous knowledge distillation methods built upon a much stronger teacher, RSL is capable of boosting the accuracy of one model by introducing other comparable or even weaker models. RSL can also be viewed as a more efficient alternative to ensemble. Extensive experiments demonstrate the superior performance of RSL on several benchmarks with significant margins.

Language Generation via Combinatorial Constraint Satisfaction: A Tree Search Enhanced Monte-Carlo Approach

Generating natural language under complex constraints is a principled formulation towards controllable text generation. We present a framework to allow specification of combinatorial constraints for sentence generation. We propose TSMH, an efficient method to generate high likelihood sentences with respect to a pre-trained language model while satisfying the constraints. Our approach is highly flexible, requires no task-specific training, and leverages efficient constraint satisfaction solving techniques. To better handle the combinatorial constraints, a tree search algorithm is embedded into the proposal process of the Markov chain Monte Carlo (MCMC) to explore candidates that satisfy more constraints. Compared to existing MCMC approaches, our sampling approach has a better mixing performance. Experiments show that TSMH achieves consistent and significant improvement on multiple language generation tasks.

Sparse R-CNN: End-to-End Object Detection with Learnable Proposals

We present Sparse R-CNN, a purely sparse method for object detection in images. Existing works on object detection heavily rely on dense object candidates, such as $k$ anchor boxes pre-defined on all grids of image feature map of size $H\times W$. In our method, however, a fixed sparse set of learned object proposals, total length of $N$, are provided to object recognition head to perform classification and location. By eliminating $HWk$ (up to hundreds of thousands) hand-designed object candidates to $N$ (e.g. 100) learnable proposals, Sparse R-CNN completely avoids all efforts related to object candidates design and many-to-one label assignment. More importantly, final predictions are directly output without non-maximum suppression post-procedure. Sparse R-CNN demonstrates accuracy, run-time and training convergence performance on par with the well-established detector baselines on the challenging COCO dataset, e.g., achieving 44.5 AP in standard $3\times$ training schedule and running at 22 fps using ResNet-50 FPN model. We hope our work could inspire re-thinking the convention of dense prior in object detectors. The code is available at: https://github.com/PeizeSun/SparseR-CNN.

DmifNet:3D Shape Reconstruction Based on Dynamic Multi-Branch Information Fusion

3D object reconstruction from a single-view image is a long-standing challenging problem. Previous work was difficult to accurately reconstruct 3D shapes with a complex topology which has rich details at the edges and corners. Moreover, previous works used synthetic data to train their network, but domain adaptation problems occurred when tested on real data. In this paper, we propose a Dynamic Multi-branch Information Fusion Network (DmifNet) which can recover a high-fidelity 3D shape of arbitrary topology from a 2D image. Specifically, we design several side branches from the intermediate layers to make the network produce more diverse representations to improve the generalization ability of network. In addition, we utilize DoG (Difference of Gaussians) to extract edge geometry and corners information from input images. Then, we use a separate side branch network to process the extracted data to better capture edge geometry and corners feature information. Finally, we dynamically fuse the information of all branches to gain final predicted probability. Extensive qualitative and quantitative experiments on a large-scale publicly available dataset demonstrate the validity and efficiency of our method. Code and models are publicly available at https://github.com/leilimaster/DmifNet.

Dense Contrastive Learning for Self-Supervised Visual Pre-Training

To date, most existing self-supervised learning methods are designed and optimized for image classification. These pre-trained models can be sub-optimal for dense prediction tasks due to the discrepancy between image-level prediction and pixel-level prediction. To fill this gap, we aim to design an effective, dense self-supervised learning method that directly works at the level of pixels (or local features) by taking into account the correspondence between local features. We present dense contrastive learning, which implements self-supervised learning by optimizing a pairwise contrastive (dis)similarity loss at the pixel level between two views of input images. Compared to the baseline method MoCo-v2, our method introduces negligible computation overhead (only <1% slower), but demonstrates consistently superior performance when transferring to downstream dense prediction tasks including object detection, semantic segmentation and instance segmentation; and outperforms the state-of-the-art methods by a large margin. Specifically, over the strong MoCo-v2 baseline, our method achieves significant improvements of 2.0% AP on PASCAL VOC object detection, 1.1% AP on COCO object detection, 0.9% AP on COCO instance segmentation, 3.0% mIoU on PASCAL VOC semantic segmentation and 1.8% mIoU on Cityscapes semantic segmentation. Code is available at: https://git.io/AdelaiDet

On the Sentence Embeddings from Pre-trained Language Models

Pre-trained contextual representations like BERT have achieved great success in natural language processing. However, the sentence embeddings from the pre-trained language models without fine-tuning have been found to poorly capture semantic meaning of sentences. In this paper, we argue that the semantic information in the BERT embeddings is not fully exploited. We first reveal the theoretical connection between the masked language model pre-training objective and the semantic similarity task theoretically, and then analyze the BERT sentence embeddings empirically. We find that BERT always induces a non-smooth anisotropic semantic space of sentences, which harms its performance of semantic similarity. To address this issue, we propose to transform the anisotropic sentence embedding distribution to a smooth and isotropic Gaussian distribution through normalizing flows that are learned with an unsupervised objective. Experimental results show that our proposed BERT-flow method obtains significant performance gains over the state-of-the-art sentence embeddings on a variety of semantic textual similarity tasks. The code is available at https://github.com/bohanli/BERT-flow.

Brain Tumor Segmentation Network Using Attention-based Fusion and Spatial Relationship Constraint

Delineating the brain tumor from magnetic resonance (MR) images is critical for the treatment of gliomas. However, automatic delineation is challenging due to the complex appearance and ambiguous outlines of tumors. Considering that multi-modal MR images can reflect different tumor biological properties, we develop a novel multi-modal tumor segmentation network (MMTSN) to robustly segment brain tumors based on multi-modal MR images. The MMTSN is composed of three sub-branches and a main branch. Specifically, the sub-branches are used to capture different tumor features from multi-modal images, while in the main branch, we design a spatial-channel fusion block (SCFB) to effectively aggregate multi-modal features. Additionally, inspired by the fact that the spatial relationship between sub-regions of tumor is relatively fixed, e.g., the enhancing tumor is always in the tumor core, we propose a spatial loss to constrain the relationship between different sub-regions of tumor. We evaluate our method on the test set of multi-modal brain tumor segmentation challenge 2020 (BraTs2020). The method achieves 0.8764, 0.8243 and 0.773 dice score for whole tumor, tumor core and enhancing tumor, respectively.