With recent advances in distantly supervised (DS) relation extraction (RE), considerable attention is attracted to leverage multi-instance learning (MIL) to distill high-quality supervision from the noisy DS. Here, we go beyond label noise and identify the key bottleneck of DS-MIL to be its low data utilization: as high-quality supervision being refined by MIL, MIL abandons a large amount of training instances, which leads to a low data utilization and hinders model training from having abundant supervision. In this paper, we propose collaborative adversarial training to improve the data utilization, which coordinates virtual adversarial training (VAT) and adversarial training (AT) at different levels. Specifically, since VAT is label-free, we employ the instance-level VAT to recycle instances abandoned by MIL. Besides, we deploy AT at the bag-level to unleash the full potential of the high-quality supervision got by MIL. Our proposed method brings consistent improvements (~ 5 absolute AUC score) to the previous state of the art, which verifies the importance of the data utilization issue and the effectiveness of our method.
Multi-head attention plays a crucial role in the recent success of Transformer models, which leads to consistent performance improvements over conventional attention in various applications. The popular belief is that this effectiveness stems from the ability of jointly attending multiple positions. In this paper, we first demonstrate that jointly attending multiple positions is not a unique feature of multi-head attention, as multi-layer single-head attention also attends multiple positions and is more effective. Then, we suggest the main advantage of the multi-head attention is the training stability, since it has less number of layers than the single-head attention, when attending the same number of positions. For example, 24-layer 16-head Transformer (BERT-large) and 384-layer single-head Transformer has the same total attention head number and roughly the same model size, while the multi-head one is significantly shallower. Meanwhile, we show that, with recent advances in deep learning, we can successfully stabilize the training of the 384-layer Transformer. As the training difficulty is no longer a bottleneck, substantially deeper single-head Transformer achieves consistent performance improvements without tuning hyper-parameters.
Identifying and understanding quality phrases from context is a fundamental task in text mining. The most challenging part of this task arguably lies in uncommon, emerging, and domain-specific phrases. The infrequent nature of these phrases significantly hurts the performance of phrase mining methods that rely on sufficient phrase occurrences in the input corpus. Context-aware tagging models, though not restricted by frequency, heavily rely on domain experts for either massive sentence-level gold labels or handcrafted gazetteers. In this work, we propose UCPhrase, a novel unsupervised context-aware quality phrase tagger. Specifically, we induce high-quality phrase spans as silver labels from consistently co-occurring word sequences within each document. Compared with typical context-agnostic distant supervision based on existing knowledge bases (KBs), our silver labels root deeply in the input domain and context, thus having unique advantages in preserving contextual completeness and capturing emerging, out-of-KB phrases. Training a conventional neural tagger based on silver labels usually faces the risk of overfitting phrase surface names. Alternatively, we observe that the contextualized attention maps generated from a transformer-based neural language model effectively reveal the connections between words in a surface-agnostic way. Therefore, we pair such attention maps with the silver labels to train a lightweight span prediction model, which can be applied to new input to recognize (unseen) quality phrases regardless of their surface names or frequency. Thorough experiments on various tasks and datasets, including corpus-level phrase ranking, document-level keyphrase extraction, and sentence-level phrase tagging, demonstrate the superiority of our design over state-of-the-art pre-trained, unsupervised, and distantly supervised methods.
Multiple intriguing problems hover in adversarial training, including robustness-accuracy trade-off, robust overfitting, and gradient masking, posing great challenges to both reliable evaluation and practical deployment. Here, we show that these problems share one common cause -- low quality samples in the dataset. We first identify an intrinsic property of the data called problematic score and then design controlled experiments to investigate its connections with these problems. Specifically, we find that when problematic data is removed, robust overfitting and gradient masking can be largely alleviated; and robustness-accuracy trade-off is more prominent for a dataset containing highly problematic data. These observations not only verify our intuition about data quality but also open new opportunities to advance adversarial training. Remarkably, simply removing problematic data from adversarial training, while making the training set smaller, yields better robustness consistently with different adversary settings, training methods, and neural architectures.
As the excessive pre-training cost arouses the need to improve efficiency, considerable efforts have been made to train BERT progressively--start from an inferior but low-cost model and gradually increase the computational complexity. Our objective is to help advance the understanding of such Transformer growth and discover principles that guide progressive training. First, we find that similar to network architecture selection, Transformer growth also favors compound scaling. Specifically, while existing methods only conduct network growth in a single dimension, we observe that it is beneficial to use compound growth operators and balance multiple dimensions (e.g., depth, width, and input length of the model). Moreover, we explore alternative growth operators in each dimension via controlled comparison to give practical guidance for operator selection. In light of our analyses, the proposed method CompoundGrow speeds up BERT pre-training by 73.6% and 82.2% for the base and large models respectively while achieving comparable performances. Code will be released for reproduction and future studies.
Our goal is to understand why the robustness drops after conducting adversarial training for too long. Although this phenomenon is commonly explained as overfitting, our analysis suggest that its primary cause is perturbation underfitting. We observe that after training for too long, FGSM-generated perturbations deteriorate into random noise. Intuitively, since no parameter updates are made to strengthen the perturbation generator, once this process collapses, it could be trapped in such local optima. Also, sophisticating this process could mostly avoid the robustness drop, which supports that this phenomenon is caused by underfitting instead of overfitting. In the light of our analyses, we propose APART, an adaptive adversarial training framework, which parameterizes perturbation generation and progressively strengthens them. Shielding perturbations from underfitting unleashes the potential of our framework. In our experiments, APART provides comparable or even better robustness than PGD-10, with only about 1/4 of its computational cost.
We explore the application of very deep Transformer models for Neural Machine Translation (NMT). Using a simple yet effective initialization technique that stabilizes training, we show that it is feasible to build standard Transformer-based models with up to 60 encoder layers and 12 decoder layers. These deep models outperform their baseline 6-layer counterparts by as much as 2.5 BLEU, and achieve new state-of-the-art benchmark results on WMT14 English-French (43.8 BLEU) and WMT14 English-German (30.1 BLEU).The code and trained models will be publicly available at: https://github.com/namisan/exdeep-nmt.
While typical named entity recognition (NER) models require the training set to be annotated with all target types, each available datasets may only cover a part of them. Instead of relying on fully-typed NER datasets, many efforts have been made to leverage multiple partially-typed ones for training and allow the resulting model to cover a full type set. However, there is neither guarantee on the quality of integrated datasets, nor guidance on the design of training algorithms. Here, we conduct a systematic analysis and comparison between partially-typed NER datasets and fully-typed ones, in both theoretical and empirical manner. Firstly, we derive a bound to establish that models trained with partially-typed annotations can reach a similar performance with the ones trained with fully-typed annotations, which also provides guidance on the algorithm design. Moreover, we conduct controlled experiments, which shows partially-typed datasets leads to similar performance with the model trained with the same amount of fully-typed annotations
Transformers have been proved effective for many deep learning tasks. Training transformers, however, requires non-trivial efforts regarding carefully designing learning rate schedulers and cutting-edge optimizers (the standard SGD fails to train Transformers effectively). In this paper, we study Transformer training from both theoretical and empirical perspectives. Our analysis reveals that unbalanced gradients are not the root cause of the instability of training. Instead, we identify an amplification effect that substantially influences training. Specifically, we observe that for each layer in a multi-layer Transformer model, heavy dependency on its residual branch makes training unstable since it amplifies small parameter perturbations (e.g., parameter updates) and result in significant disturbances in the model output, yet a light dependency limits the potential of model training and can lead to an inferior trained model. Inspired by our analysis, we propose Admin ($\mathbf{Ad}$aptive $\mathbf{m}$odel $\mathbf{in}$itialization) to stabilize the training in the early stage and unleash its full potential in the late stage. Extensive experiments show that Admin is more stable, converges faster, and leads to better performance.
Infrared small target detection is a key technique in infrared search and tracking (IRST) systems. Although deep learning has been widely used in the vision tasks of visible light images recently, it is rarely used in infrared small target detection due to the difficulty in learning small target features. In this paper, we propose a novel lightweight convolutional neural network TBC-Net for infrared small target detection. The TBCNet consists of a target extraction module (TEM) and a semantic constraint module (SCM), which are used to extract small targets from infrared images and to classify the extracted target images during the training, respectively. Meanwhile, we propose a joint loss function and a training method. The SCM imposes a semantic constraint on TEM by combining the high-level classification task and solve the problem of the difficulty to learn features caused by class imbalance problem. During the training, the targets are extracted from the input image and then be classified by SCM. During the inference, only the TEM is used to detect the small targets. We also propose a data synthesis method to generate training data. The experimental results show that compared with the traditional methods, TBC-Net can better reduce the false alarm caused by complicated background, the proposed network structure and joint loss have a significant improvement on small target feature learning. Besides, TBC-Net can achieve real-time detection on the NVIDIA Jetson AGX Xavier development board, which is suitable for applications such as field research with drones equipped with infrared sensors.