The Visual Question Answering (VQA) task utilizes both visual image and language analysis to answer a textual question with respect to an image. It has been a popular research topic with an increasing number of real-world applications in the last decade. This paper introduces a novel hierarchical integration of vision and language AliceMind-MMU (ALIbaba's Collection of Encoder-decoders from Machine IntelligeNce lab of Damo academy - MultiMedia Understanding), which leads to similar or even slightly better results than a human being does on VQA. A hierarchical framework is designed to tackle the practical problems of VQA in a cascade manner including: (1) diverse visual semantics learning for comprehensive image content understanding; (2) enhanced multi-modal pre-training with modality adaptive attention; and (3) a knowledge-guided model integration with three specialized expert modules for the complex VQA task. Treating different types of visual questions with corresponding expertise needed plays an important role in boosting the performance of our VQA architecture up to the human level. An extensive set of experiments and analysis are conducted to demonstrate the effectiveness of the new research work.

Vision-Language Pre-training (VLP) methods based on object detection enjoy the rich knowledge of fine-grained object-text alignment but at the cost of computationally expensive inference. Recent Visual-Transformer (ViT)-based approaches circumvent this issue while struggling with long visual sequences without detailed cross-modal alignment information. This paper introduces a ViT-based VLP technique that efficiently incorporates object information through a novel patch-text alignment mechanism. Specifically, we convert object-level signals into patch-level ones and devise a Patch-Text Alignment pre-training task (PTA) to learn a text-aware patch detector. By using off-the-shelf delicate object annotations in 5% training images, we jointly train PTA with other conventional VLP objectives in an end-to-end manner, bypassing the high computational cost of object detection and yielding an effective patch detector that accurately detects text-relevant patches, thus considerably reducing patch sequences and accelerating computation within the ViT backbone. Our experiments on a variety of widely-used benchmarks reveal that our method achieves a speedup of nearly 88% compared to prior VLP models while maintaining competitive or superior performance on downstream tasks with similar model size and data scale.

Vision Transformer (ViT) based Vision-Language Pre-training (VLP) models have demonstrated impressive performance in various tasks. However, the lengthy visual token sequences fed into ViT can lead to training inefficiency and ineffectiveness. Existing efforts address the challenge by either bottom-level patch extraction in the ViT backbone or top-level patch abstraction outside, not balancing training efficiency and effectiveness well. Inspired by text summarization in natural language processing, we propose a Bottom-Up Patch Summarization approach named BUS, coordinating bottom-level extraction and top-level abstraction to learn a concise summary of lengthy visual token sequences efficiently. Specifically, We incorporate a Text-Semantics-Aware Patch Selector (TSPS) into the ViT backbone to perform a coarse-grained visual token extraction and then attach a flexible Transformer-based Patch Abstraction Decoder (PAD) upon the backbone for top-level visual abstraction. This bottom-up collaboration enables our BUS to yield high training efficiency while maintaining or even improving effectiveness. We evaluate our approach on various visual-language understanding and generation tasks and show competitive downstream task performance while boosting the training efficiency by 50%. Additionally, our model achieves state-of-the-art performance on many downstream tasks by increasing input image resolution without increasing computational costs over baselines.

Recent years have witnessed a big convergence of language, vision, and multi-modal pretraining. In this work, we present mPLUG-2, a new unified paradigm with modularized design for multi-modal pretraining, which can benefit from modality collaboration while addressing the problem of modality entanglement. In contrast to predominant paradigms of solely relying on sequence-to-sequence generation or encoder-based instance discrimination, mPLUG-2 introduces a multi-module composition network by sharing common universal modules for modality collaboration and disentangling different modality modules to deal with modality entanglement. It is flexible to select different modules for different understanding and generation tasks across all modalities including text, image, and video. Empirical study shows that mPLUG-2 achieves state-of-the-art or competitive results on a broad range of over 30 downstream tasks, spanning multi-modal tasks of image-text and video-text understanding and generation, and uni-modal tasks of text-only, image-only, and video-only understanding. Notably, mPLUG-2 shows new state-of-the-art results of 48.0 top-1 accuracy and 80.3 CIDEr on the challenging MSRVTT video QA and video caption tasks with a far smaller model size and data scale. It also demonstrates strong zero-shot transferability on vision-language and video-language tasks. Code and models will be released in https://github.com/alibaba/AliceMind.

Pre-trained language models (PLMs) such as BERT, RoBERTa, and DeBERTa have achieved state-of-the-art performance on various downstream tasks. The enormous sizes of PLMs hinder their deployment in resource-constrained scenarios, e.g., on edge and mobile devices. To address this issue, many model compression approaches have been proposed to reduce the number of model parameters. This paper focuses on compressing the token embedding matrices of PLMs, which typically make up a large proportion (around 20-30%) of the entire model parameters. Existing efforts to compress token embedding usually require the introduction of customized compression architectures or the optimization of model compression processes for individual downstream tasks, limiting their applicability in both model and task dimensions. To overcome these limitations and adhere to the principle of "one-for-all", we propose a lightweight token embedding framework named LightToken, which is able to produce compressed token embedding in a task and model-agnostic fashion. LightToken is generally compatible with different architectures and applicable to any downstream task. Specifically, through an integration of low-rank approximation, novel residual binary autoencoder, and a new compression loss function, LightToken can significantly improve the model compression ratio. To demonstrate the effectiveness of LightToken, we conduct comprehensive experiments on natural language understanding and question answering tasks. In particular, LightToken improves the state-of-the-art token embedding compression ratio from 5 to 25 and outperforms the existing token embedding compression approaches by 11% and 5% on GLUE and SQuAD v1.1 benchmarks, respectively.

Large-scale pretrained foundation models have been an emerging paradigm for building artificial intelligence (AI) systems, which can be quickly adapted to a wide range of downstream tasks. This paper presents mPLUG, a new vision-language foundation model for both cross-modal understanding and generation. Most existing pre-trained models suffer from the problems of low computational efficiency and information asymmetry brought by the long visual sequence in cross-modal alignment. To address these problems, mPLUG introduces an effective and efficient vision-language architecture with novel cross-modal skip-connections, which creates inter-layer shortcuts that skip a certain number of layers for time-consuming full self-attention on the vision side. mPLUG is pre-trained end-to-end on large-scale image-text pairs with both discriminative and generative objectives. It achieves state-of-the-art results on a wide range of vision-language downstream tasks, such as image captioning, image-text retrieval, visual grounding and visual question answering. mPLUG also demonstrates strong zero-shot transferability when directly transferred to multiple video-language tasks.

Existing work in multilingual pretraining has demonstrated the potential of cross-lingual transferability by training a unified Transformer encoder for multiple languages. However, much of this work only relies on the shared vocabulary and bilingual contexts to encourage the correlation across languages, which is loose and implicit for aligning the contextual representations between languages. In this paper, we plug a cross-attention module into the Transformer encoder to explicitly build the interdependence between languages. It can effectively avoid the degeneration of predicting masked words only conditioned on the context in its own language. More importantly, when fine-tuning on downstream tasks, the cross-attention module can be plugged in or out on-demand, thus naturally benefiting a wider range of cross-lingual tasks, from language understanding to generation. As a result, the proposed cross-lingual model delivers new state-of-the-art results on various cross-lingual understanding tasks of the XTREME benchmark, covering text classification, sequence labeling, question answering, and sentence retrieval. For cross-lingual generation tasks, it also outperforms all existing cross-lingual models and state-of-the-art Transformer variants on WMT14 English-to-German and English-to-French translation datasets, with gains of up to 1~2 BLEU.

Pretrained language models have recently advanced a wide range of natural language processing tasks. Nowadays, the application of pretrained language models to IR tasks has also achieved impressive results. Typical methods either directly apply a pretrained model to improve the re-ranking stage, or use it to conduct passage expansion and term weighting for first-stage retrieval. We observe that the passage ranking and passage expansion tasks share certain inherent relations, and can benefit from each other. Therefore, in this paper, we propose a general pretraining framework to enhance both tasks with Unified Encoder-Decoder networks (UED). The overall ranking framework consists of two parts in a cascade manner: (1) passage expansion with a pretraining-based query generation method; (2) re-ranking of passage candidates from a traditional retrieval method with a pretrained transformer encoder. Both the two parts are based on the same pretrained UED model, where we jointly train the passage ranking and query generation tasks for further improving the full ranking pipeline. An extensive set of experiments have been conducted on two large-scale passage retrieval datasets to demonstrate the state-of-the-art results of the proposed framework in both the first-stage retrieval and the final re-ranking. In addition, we successfully deploy the framework to our online production system, which can stably serve industrial applications with a request volume of up to 100 QPS in less than 300ms.

Cold-start is the most difficult and time-consuming phase when building a question answering based chatbot for a new business scenario because of the collection of sufficient training data. In this paper, we propose AliMe DA, a practical data augmentation (DA) framework that consists of data production, denoising and consumption, to alleviate this problem. We show how our DA approach can be used to substantially enhance annotation productivity and also improve downstream model performance. More importantly, we provide best practices for data augmentation, including how to choose and employ appropriate methods at each stage of our framework, and share our observation on the applicable scene of data augmentation in the era of pre-trained language models.

Recent studies about learning multilingual representations have achieved significant performance gains across a wide range of downstream cross-lingual tasks. They train either an encoder-only Transformer mainly for understanding tasks, or an encoder-decoder Transformer specifically for generation tasks, ignoring the correlation between the two tasks and frameworks. In contrast, this paper presents a variable encoder-decoder (VECO) pre-training approach to unify the two mainstreams in both model architectures and pre-training tasks. VECO splits the standard Transformer block into several sub-modules trained with both inner-sequence and cross-sequence masked language modeling, and correspondingly reorganizes certain sub-modules for understanding and generation tasks during inference. Such a workflow not only ensures to train the most streamlined parameters necessary for two kinds of tasks, but also enables them to boost each other via sharing common sub-modules. As a result, VECO delivers new state-of-the-art results on various cross-lingual understanding tasks of the XTREME benchmark covering text classification, sequence labeling, question answering, and sentence retrieval. For generation tasks, VECO also outperforms all existing cross-lingual models and state-of-the-art Transformer variants on WMT14 English-to-German and English-to-French translation datasets, with gains of up to 1~2 BLEU.

Large pre-trained language models achieve state-of-the-art results when fine-tuned on downstream NLP tasks. However, they almost exclusively focus on text-only representation, while neglecting cell-level layout information that is important for form image understanding. In this paper, we propose a new pre-training approach, StructuralLM, to jointly leverage cell and layout information from scanned documents. Specifically, we pre-train StructuralLM with two new designs to make the most of the interactions of cell and layout information: 1) each cell as a semantic unit; 2) classification of cell positions. The pre-trained StructuralLM achieves new state-of-the-art results in different types of downstream tasks, including form understanding (from 78.95 to 85.14), document visual question answering (from 72.59 to 83.94) and document image classification (from 94.43 to 96.08).

Vision-language pre-training (VLP) on largescale image-text pairs has achieved huge success for the cross-modal downstream tasks. The most existing pre-training methods mainly adopt a two-step training procedure, which firstly employs a pre-trained object detector to extract region-based visual features, then concatenates the image representation and text embedding as the input of Transformer to train. However, these methods face problems of using task-specific visual representation of the specific object detector for generic crossmodal understanding, and the computation inefficiency of two-stage pipeline. In this paper, we propose the first end-to-end vision-language pre-trained model for both V+L understanding and generation, namely E2E-VLP, where we build a unified Transformer framework to jointly learn visual representation, and semantic alignments between image and text. We incorporate the tasks of object detection and image captioning into pretraining with a unified Transformer encoderdecoder architecture for enhancing visual learning. An extensive set of experiments have been conducted on well-established visionlanguage downstream tasks to demonstrate the effectiveness of this novel VLP paradigm.

Vision-language pre-training (VLP) on largescale image-text pairs has recently witnessed rapid progress for learning cross-modal representations. Existing pre-training methods either directly concatenate image representation and text representation at a feature level as input to a single-stream Transformer, or use a two-stream cross-modal Transformer to align the image-text representation at a highlevel semantic space. In real-world imagetext data, we observe that it is easy for some of the image-text pairs to align simple semantics on both modalities, while others may be related after higher-level abstraction. Therefore, in this paper, we propose a new pre-training method SemVLP, which jointly aligns both the low-level and highlevel semantics between image and text representations. The model is pre-trained iteratively with two prevalent fashions: singlestream pre-training to align at a fine-grained feature level and two-stream pre-training to align high-level semantics, by employing a shared Transformer network with a pluggable cross-modal attention module. An extensive set of experiments have been conducted on four well-established vision-language understanding tasks to demonstrate the effectiveness of the proposed SemVLP in aligning cross-modal representations towards different semantic granularities.

Recently, question answering (QA) based on machine reading comprehension has become popular. This work focuses on generative QA which aims to generate an abstractive answer to a given question instead of extracting an answer span from a provided passage. Generative QA often suffers from two critical problems: (1) summarizing content irrelevant to a given question, (2) drifting away from a correct answer during generation. In this paper, we address these problems by a novel Rationale-Enriched Answer Generator (REAG), which incorporates an extractive mechanism into a generative model. Specifically, we add an extraction task on the encoder to obtain the rationale for an answer, which is the most relevant piece of text in an input document to a given question. Based on the extracted rationale and original input, the decoder is expected to generate an answer with high confidence. We jointly train REAG on the MS MARCO QA+NLG task and the experimental results show that REAG improves the quality and semantic accuracy of answers over baseline models.

Existing approaches to vision-language pre-training (VLP) heavily rely on an object detector based on bounding boxes (regions), where salient objects are first detected from images and then a Transformer-based model is used for cross-modal fusion. Despite their superior performance, these approaches are bounded by the capability of the object detector in terms of both effectiveness and efficiency. Besides, the presence of object detection imposes unnecessary constraints on model designs and makes it difficult to support end-to-end training. In this paper, we revisit grid-based convolutional features for vision-language pre-training, skipping the expensive region-related steps. We propose a simple yet effective grid-based VLP method that works surprisingly well with the grid features. By pre-training only with in-domain datasets, the proposed Grid-VLP method can outperform most competitive region-based VLP methods on three examined vision-language understanding tasks. We hope that our findings help to further advance the state of the art of vision-language pre-training, and provide a new direction towards effective and efficient VLP.

Self-supervised pre-training, such as BERT, MASS and BART, has emerged as a powerful technique for natural language understanding and generation. Existing pre-training techniques employ autoencoding and/or autoregressive objectives to train Transformer-based models by recovering original word tokens from corrupted text with some masked tokens. The training goals of existing techniques are often inconsistent with the goals of many language generation tasks, such as generative question answering and conversational response generation, for producing new text given context. This work presents PALM with a novel scheme that jointly pre-trains an autoencoding and autoregressive language model on a large unlabeled corpus, specifically designed for generating new text conditioned on context. The new scheme alleviates the mismatch introduced by the existing denoising scheme between pre-training and fine-tuning where generation is more than reconstructing original text. An extensive set of experiments show that PALM achieves new state-of-the-art results on a variety of language generation benchmarks covering generative question answering (Rank 1 on the official MARCO leaderboard), abstractive summarization on CNN/DailyMail as well as Gigaword, question generation on SQuAD, and conversational response generation on Cornell Movie Dialogues.

Question answering (QA) based on machine reading comprehension has been a recent surge in popularity, yet most work has focused on extractive methods. We instead address a more challenging QA problem of generating a well-formed answer by reading and summarizing the paragraph for a given question. For the generative QA task, we introduce a new neural architecture, LatentQA. in which a novel stochastic selector network composes a well-formed answer with words selected from the question. the paragraph and the global vocabulary. based on a sequence of discrete latent variables. Bayesian inference for the latent variables is performed to train the LatentQA model. The experiments on public datasets of natural answer generation confirm the effectiveness of LatentQA in generating high-quality well-formed answers.

Recently, the pre-trained language model, BERT (and its robustly optimized version RoBERTa), has attracted a lot of attention in natural language understanding (NLU), and achieved state-of-the-art accuracy in various NLU tasks, such as sentiment classification, natural language inference, semantic textual similarity and question answering. Inspired by the linearization exploration work of Elman [8], we extend BERT to a new model, StructBERT, by incorporating language structures into pre-training. Specifically, we pre-train StructBERT with two auxiliary tasks to make the most of the sequential order of words and sentences, which leverage language structures at the word and sentence levels, respectively. As a result, the new model is adapted to different levels of language understanding required by downstream tasks. The StructBERT with structural pre-training gives surprisingly good empirical results on a variety of downstream tasks, including pushing the state-of-the-art on the GLUE benchmark to 89.0 (outperforming all published models), the F1 score on SQuAD v1.1 question answering to 93.0, the accuracy on SNLI to 91.7.

Learning to control the structure of sentences is a challenging problem in text generation. Existing work either relies on simple deterministic approaches or RL-based hard structures. We explore the use of structured variational autoencoders to infer latent templates for sentence generation using a soft, continuous relaxation in order to utilize reparameterization for training. Specifically, we propose a Gumbel-CRF, a continuous relaxation of the CRF sampling algorithm using a relaxed Forward-Filtering Backward-Sampling (FFBS) approach. As a reparameterized gradient estimator, the Gumbel-CRF gives more stable gradients than score-function based estimators. As a structured inference network, we show that it learns interpretable templates during training, which allows us to control the decoder during testing. We demonstrate the effectiveness of our methods with experiments on data-to-text generation and unsupervised paraphrase generation.

A fundamental trade-off between effectiveness and efficiency needs to be balanced when designing an online question answering system. Effectiveness comes from sophisticated functions such as extractive machine reading comprehension (MRC), while efficiency is obtained from improvements in preliminary retrieval components such as candidate document selection and paragraph ranking. Given the complexity of the real-world multi-document MRC scenario, it is difficult to jointly optimize both in an end-to-end system. To address this problem, we develop a novel deep cascade learning model, which progressively evolves from the document-level and paragraph-level ranking of candidate texts to more precise answer extraction with machine reading comprehension. Specifically, irrelevant documents and paragraphs are first filtered out with simple functions for efficiency consideration. Then we jointly train three modules on the remaining texts for better tracking the answer: the document extraction, the paragraph extraction and the answer extraction. Experiment results show that the proposed method outperforms the previous state-of-the-art methods on two large-scale multi-document benchmark datasets, i.e., TriviaQA and DuReader. In addition, our online system can stably serve typical scenarios with millions of daily requests in less than 50ms.