Bag of Tricks for Effective Language Model Pretraining and Downstream Adaptation: A Case Study on GLUE

This technical report briefly describes our JDExplore d-team's submission Vega v1 on the General Language Understanding Evaluation (GLUE) leaderboard, where GLUE is a collection of nine natural language understanding tasks, including question answering, linguistic acceptability, sentiment analysis, text similarity, paraphrase detection, and natural language inference. [Method] We investigate several effective strategies and choose their best combination setting as the training recipes. As for model structure, we employ the vanilla Transformer with disentangled attention as the basic block encoder. For self-supervised training, we employ the representative denoising objective (i.e., replaced token detection) in phase 1 and combine the contrastive objective (i.e., sentence embedding contrastive learning) with it in phase 2. During fine-tuning, several advanced techniques such as transductive fine-tuning, self-calibrated fine-tuning, and adversarial fine-tuning are adopted. [Results] According to our submission record (Jan. 2022), with our optimized pretraining and fine-tuning strategies, our 1.3 billion model sets new state-of-the-art on 4/9 tasks, achieving the best average score of 91.3. Encouragingly, our Vega v1 is the first to exceed powerful human performance on the two challenging tasks, i.e., SST-2 and WNLI. We believe our empirically successful recipe with a bag of tricks could shed new light on developing efficient discriminative large language models.

Toward Efficient Language Model Pretraining and Downstream Adaptation via Self-Evolution: A Case Study on SuperGLUE

This technical report briefly describes our JDExplore d-team's Vega v2 submission on the SuperGLUE leaderboard. SuperGLUE is more challenging than the widely used general language understanding evaluation (GLUE) benchmark, containing eight difficult language understanding tasks, including question answering, natural language inference, word sense disambiguation, coreference resolution, and reasoning. [Method] Instead of arbitrarily increasing the size of a pretrained language model (PLM), our aim is to 1) fully extract knowledge from the input pretraining data given a certain parameter budget, e.g., 6B, and 2) effectively transfer this knowledge to downstream tasks. To achieve goal 1), we propose self-evolution learning for PLMs to wisely predict the informative tokens that should be masked, and supervise the masked language modeling (MLM) process with rectified smooth labels. For goal 2), we leverage the prompt transfer technique to improve the low-resource tasks by transferring the knowledge from the foundation model and related downstream tasks to the target task. [Results] According to our submission record (Oct. 2022), with our optimized pretraining and fine-tuning strategies, our 6B Vega method achieved new state-of-the-art performance on 4/8 tasks, sitting atop the SuperGLUE leaderboard on Oct. 8, 2022, with an average score of 91.3.

Improving Sharpness-Aware Minimization with Fisher Mask for Better Generalization on Language Models

Fine-tuning large pretrained language models on a limited training corpus usually suffers from poor generalization. Prior works show that the recently-proposed sharpness-aware minimization (SAM) optimization method can improve the model generalization. However, SAM adds a perturbation to each model parameter equally (but not all parameters contribute equally to the optimization of training), which we argue is sub-optimal and will lead to excessive computation. In this paper, we propose a novel optimization procedure, namely FSAM, which introduces a Fisher mask to improve the efficiency and performance of SAM. In short, instead of adding perturbation to all parameters, FSAM uses the Fisher information to identity the important parameters and formulates a Fisher mask to obtain the sparse perturbation, i.e., making the optimizer focus on these important parameters. Experiments on various tasks in GLUE and SuperGLUE benchmarks show that FSAM consistently outperforms the vanilla SAM by 0.67~1.98 average score among four different pretrained models. We also empirically show that FSAM works well in other complex scenarios, e.g., fine-tuning on generation tasks or limited training data. Encouragingly, when training data is limited, FSAM improves the SAM by a large margin, i.e., up to 15.1.

PANDA: Prompt Transfer Meets Knowledge Distillation for Efficient Model Adaptation

Prompt-tuning, which freezes pretrained language models (PLMs) and only fine-tunes few parameters of additional soft prompt, shows competitive performance against full-parameter fine-tuning (i.e.model-tuning) when the PLM has billions of parameters, but still performs poorly in the case of smaller PLMs. Hence, prompt transfer (PoT), which initializes the target prompt with the trained prompt of similar source tasks, is recently proposed to improve over prompt-tuning. However, such a vanilla PoT approach usually achieves sub-optimal performance, as (i) the PoT is sensitive to the similarity of source-target pair and (ii) directly fine-tuning the prompt initialized with source prompt on target task might lead to catastrophic forgetting of source knowledge. In response to these problems, we propose a new metric to accurately predict the prompt transferability (regarding (i)), and a novel PoT approach (namely PANDA) that leverages the knowledge distillation technique to transfer the "knowledge" from the source prompt to the target prompt in a subtle manner and alleviate the catastrophic forgetting effectively (regarding (ii)). Furthermore, to achieve adaptive prompt transfer for each source-target pair, we use our metric to control the knowledge transfer in our PANDA approach. Extensive and systematic experiments on 189 combinations of 21 source and 9 target datasets across 5 scales of PLMs demonstrate that: 1) our proposed metric works well to predict the prompt transferability; 2) our PANDA consistently outperforms the vanilla PoT approach by 2.3% average score (up to 24.1%) among all tasks and model sizes; 3) with our PANDA approach, prompt-tuning can achieve competitive and even better performance than model-tuning in various PLM scales scenarios. Code and models will be released upon acceptance.

E2S2: Encoding-Enhanced Sequence-to-Sequence Pretraining for Language Understanding and Generation

Sequence-to-sequence (seq2seq) learning has become a popular trend for pretraining language models, due to its succinct and universal framework. However, the prior seq2seq pretraining models generally focus on reconstructive objectives on the decoder side and neglect the effect of encoder-side supervisions, which may lead to sub-optimal performance. To this end, we propose an encoding-enhanced seq2seq pretraining strategy, namely E2S2, which improves the seq2seq models via integrating more efficient self-supervised information into the encoders. Specifically, E2S2 contains two self-supervised objectives upon the encoder, which are from two perspectives: 1) denoising the corrupted sentence (denoising objective); 2) learning robust sentence representations (contrastive objective). With these two objectives, the encoder can effectively distinguish the noise tokens and capture more syntactic and semantic knowledge, thus strengthening the ability of seq2seq model to comprehend the input sentence and conditionally generate the target. We conduct extensive experiments spanning language understanding and generation tasks upon the state-of-the-art seq2seq pretrained language model BART. We show that E2S2 can consistently boost the performance, including 1.0% averaged gain on GLUE benchmark and 1.75% F_0.5 score improvement on CoNLL2014 dataset, validating the effectiveness and robustness of our E2S2.

A Contrastive Cross-Channel Data Augmentation Framework for Aspect-based Sentiment Analysis

Aspect-Based Sentiment Analysis is a fine-grained sentiment analysis task, which focuses on detecting the sentiment polarity towards the aspect in a sentence. However, it is always sensitive to the multi-aspect challenge, where features of multiple aspects in a sentence will affect each other. To mitigate this issue, we design a novel training framework, called Contrastive Cross-Channel Data Augmentation (C3DA). A source sentence will be fed a domain-specific generator to obtain some synthetic sentences and is concatenated with these generated sentences to conduct supervised training and proposed contrastive training. To be specific, considering the limited ABSA labeled data, we also introduce some parameter-efficient approaches to complete sentences generation. This novel generation method consists of an Aspect Augmentation Channel (AAC) to generate aspect-specific sentences and a Polarity Augmentation (PAC) to generate polarity-inverted sentences. According to our extensive experiments, our C3DA framework can outperform those baselines without any augmentations by about 1\% on accuracy and Macro-F1.

Knowledge Graph Augmented Network Towards Multiview Representation Learning for Aspect-based Sentiment Analysis

Aspect-based sentiment analysis (ABSA) is a fine-grained task of sentiment analysis. To better comprehend long complicated sentences and obtain accurate aspect-specific information, linguistic and commonsense knowledge are generally required in this task. However, most methods employ complicated and inefficient approaches to incorporate external knowledge, e.g., directly searching the graph nodes. Additionally, the complementarity between external knowledge and linguistic information has not been thoroughly studied. To this end, we propose a knowledge graph augmented network (KGAN), which aims to effectively incorporate external knowledge with explicitly syntactic and contextual information. In particular, KGAN captures the sentiment feature representations from multiple different perspectives, i.e., context-, syntax- and knowledge-based. First, KGAN learns the contextual and syntactic representations in parallel to fully extract the semantic features. Then, KGAN integrates the knowledge graphs into the embedding space, based on which the aspect-specific knowledge representations are further obtained via an attention mechanism. Last, we propose a hierarchical fusion module to complement these multiview representations in a local-to-global manner. Extensive experiments on three popular ABSA benchmarks demonstrate the effectiveness and robustness of our KGAN. Notably, with the help of the pretrained model of RoBERTa, KGAN achieves a new record of state-of-the-art performance.

Unified Instance and Knowledge Alignment Pretraining for Aspect-based Sentiment Analysis

Aspect-based Sentiment Analysis (ABSA) aims to determine the sentiment polarity towards an aspect. Because of the expensive and limited labelled data, the pretraining strategy has become the de-facto standard for ABSA. However, there always exists severe domain shift between the pretraining and downstream ABSA datasets, hindering the effective knowledge transfer when directly finetuning and making the downstream task performs sub-optimal. To mitigate such domain shift, we introduce a unified alignment pretraining framework into the vanilla pretrain-finetune pipeline with both instance- and knowledge-level alignments. Specifically, we first devise a novel coarse-to-fine retrieval sampling approach to select target domain-related instances from the large-scale pretraining dataset, thus aligning the instances between pretraining and target domains (\textit{First Stage}). Then, we introduce a knowledge guidance-based strategy to further bridge the domain gap at the knowledge level. In practice, we formulate the model pretrained on the sampled instances into a knowledge guidance model and a learner model, respectively. On the target dataset, we design an on-the-fly teacher-student joint fine-tuning approach to progressively transfer the knowledge from the knowledge guidance model to the learner model (\textit{Second Stage}). Thereby, the learner model can maintain more domain-invariant knowledge when learning new knowledge from the target dataset. In the \textit{Third Stage,} the learner model is finetuned to better adapt its learned knowledge to the target dataset. Extensive experiments and analyses on several ABSA benchmarks demonstrate the effectiveness and universality of our proposed pretraining framework. Notably, our pretraining framework pushes several strong baseline models up to the new state-of-the-art records. We release our code and models.