Hang Li

Zhikai Chen, Haitao Mao, Hang Li et al.

Learning on Graphs has attracted immense attention due to its wide real-world applications. The most popular pipeline for learning on graphs with textual node attributes primarily relies on Graph Neural Networks (GNNs), and utilizes shallow text embedding as initial node representations, which has limitations in general knowledge and profound semantic understanding. In recent years, Large Language Models (LLMs) have been proven to possess extensive common knowledge and powerful semantic comprehension abilities that have revolutionized existing workflows to handle text data. In this paper, we aim to explore the potential of LLMs in graph machine learning, especially the node classification task, and investigate two possible pipelines: LLMs-as-Enhancers and LLMs-as-Predictors. The former leverages LLMs to enhance nodes' text attributes with their massive knowledge and then generate predictions through GNNs. The latter attempts to directly employ LLMs as standalone predictors. We conduct comprehensive and systematical studies on these two pipelines under various settings. From comprehensive empirical results, we make original observations and find new insights that open new possibilities and suggest promising directions to leverage LLMs for learning on graphs. Our codes and datasets are available at https://github.com/CurryTang/Graph-LLM.

Hang Li, Jindong Gu, Rajat Koner et al. · deepmind, oxford

The field of multimodal research focusing on the comprehension and creation of both images and text has witnessed significant strides. This progress is exemplified by the emergence of sophisticated models dedicated to image captioning at scale, such as the notable Flamingo model and text-to-image generative models, with DALL-E serving as a prominent example. An interesting question worth exploring in this domain is whether Flamingo and DALL-E understand each other. To study this question, we propose a reconstruction task where Flamingo generates a description for a given image and DALL-E uses this description as input to synthesize a new image. We argue that these models understand each other if the generated image is similar to the given image. Specifically, we study the relationship between the quality of the image reconstruction and that of the text generation. We find that an optimal description of an image is one that gives rise to a generated image similar to the original one. The finding motivates us to propose a unified framework to finetune the text-to-image and image-to-text models. Concretely, the reconstruction part forms a regularization loss to guide the tuning of the models. Extensive experiments on multiple datasets with different image captioning and image generation models validate our findings and demonstrate the effectiveness of our proposed unified framework. As DALL-E and Flamingo are not publicly available, we use Stable Diffusion and BLIP in the remaining work. Project website: https://dalleflamingo.github.io.

Tong Liu, Yushan Liu, Marcel Hildebrandt et al. · bytedance

Graphs can model real-world, complex systems by representing entities and their interactions in terms of nodes and edges. To better exploit the graph structure, graph neural networks have been developed, which learn entity and edge embeddings for tasks such as node classification and link prediction. These models achieve good performance with respect to accuracy, but the confidence scores associated with the predictions might not be calibrated. That means that the scores might not reflect the ground-truth probabilities of the predicted events, which would be especially important for safety-critical applications. Even though graph neural networks are used for a wide range of tasks, the calibration thereof has not been sufficiently explored yet. We investigate the calibration of graph neural networks for node classification, study the effect of existing post-processing calibration methods, and analyze the influence of model capacity, graph density, and a new loss function on calibration. Further, we propose a topology-aware calibration method that takes the neighboring nodes into account and yields improved calibration compared to baseline methods.

Jiabao Wang, Yang Li, Xiu-Shen Wei et al.

Unsupervised learning technology has caught up with or even surpassed supervised learning technology in general object classification (GOC) and person re-identification (re-ID). However, it is found that the unsupervised learning of fine-grained visual classification (FGVC) is more challenging than GOC and person re-ID. In order to bridge the gap between unsupervised and supervised learning for FGVC, we investigate the essential factors (including feature extraction, clustering, and contrastive learning) for the performance gap between supervised and unsupervised FGVC. Furthermore, we propose a simple, effective, and practical method, termed as UFCL, to alleviate the gap. Three key issues are concerned and improved: First, we introduce a robust and powerful backbone, ResNet50-IBN, which has an ability of domain adaptation when we transfer ImageNet pre-trained models to FGVC tasks. Next, we propose to introduce HDBSCAN instead of DBSCAN to do clustering, which can generate better clusters for adjacent categories with fewer hyper-parameters. Finally, we propose a weighted feature agent and its updating mechanism to do contrastive learning by using the pseudo labels with inevitable noise, which can improve the optimization process of learning the parameters of the network. The effectiveness of our UFCL is verified on CUB-200-2011, Oxford-Flowers, Oxford-Pets, Stanford-Dogs, Stanford-Cars and FGVC-Aircraft datasets. Under the unsupervised FGVC setting, we achieve state-of-the-art results, and analyze the key factors and the important parameters to provide a practical guidance.

Ugur Sahin, Hang Li, Qadeer Khan et al.

Contemporary large-scale visual language models (VLMs) exhibit strong representation capacities, making them ubiquitous for enhancing image and text understanding tasks. They are often trained in a contrastive manner on a large and diverse corpus of images and corresponding text captions scraped from the internet. Despite this, VLMs often struggle with compositional reasoning tasks which require a fine-grained understanding of the complex interactions of objects and their attributes. This failure can be attributed to two main factors: 1) Contrastive approaches have traditionally focused on mining negative examples from existing datasets. However, the mined negative examples might not be difficult for the model to discriminate from the positive. An alternative to mining would be negative sample generation 2) But existing generative approaches primarily focus on generating hard negative texts associated with a given image. Mining in the other direction, i.e., generating negative image samples associated with a given text has been ignored. To overcome both these limitations, we propose a framework that not only mines in both directions but also generates challenging negative samples in both modalities, i.e., images and texts. Leveraging these generative hard negative samples, we significantly enhance VLMs' performance in tasks involving multimodal compositional reasoning. Our code and dataset are released at https://ugorsahin.github.io/enhancing-multimodal-compositional-reasoning-of-vlm.html.

Yuanshun Yao, Chong Wang, Hang Li

Modern recommender systems face an increasing need to explain their recommendations. Despite considerable progress in this area, evaluating the quality of explanations remains a significant challenge for researchers and practitioners. Prior work mainly conducts human study to evaluate explanation quality, which is usually expensive, time-consuming, and prone to human bias. In this paper, we propose an offline evaluation method that can be computed without human involvement. To evaluate an explanation, our method quantifies its counterfactual impact on the recommendation. To validate the effectiveness of our method, we carry out an online user study. We show that, compared to conventional methods, our method can produce evaluation scores more correlated with the real human judgments, and therefore can serve as a better proxy for human evaluation. In addition, we show that explanations with high evaluation scores are considered better by humans. Our findings highlight the promising direction of using the counterfactual approach as one possible way to evaluate recommendation explanations.

Yuanshun Yao, Chong Wang, Hang Li

Recommender system practitioners are facing increasing pressure to explain recommendations. We explore how to explain recommendations using counterfactual logic, i.e. "Had you not interacted with the following items, we would not recommend it." Compared to the traditional explanation logic, counterfactual explanations are easier to understand, more technically verifiable, and more informative in terms of giving users control over recommendations. The major challenge of generating such explanations is the computational cost because it requires repeatedly retraining the models to obtain the effect on a recommendation caused by the absence of user history. We propose a learning-based framework to generate counterfactual explanations. The key idea is to train a surrogate model to learn the effect of removing a subset of user history on the recommendation. To this end, we first artificially simulate the counterfactual outcomes on the recommendation after deleting subsets of history. Then we train a surrogate model to learn the mapping between a history deletion and the corresponding change of the recommendation caused by the deletion. Finally, to generate an explanation, we find the history subset predicted by the surrogate model that is most likely to remove the recommendation. Through offline experiments and online user studies, we show our method, compared to baselines, can generate explanations that are more counterfactually valid and more satisfactory considered by users.

Hang Li, Tianlong Xu, Ethan Chang et al.

Knowledge tagging for questions is vital in modern intelligent educational applications, including learning progress diagnosis, practice question recommendations, and course content organization. Traditionally, these annotations have been performed by pedagogical experts, as the task demands not only a deep semantic understanding of question stems and knowledge definitions but also a strong ability to link problem-solving logic with relevant knowledge concepts. With the advent of advanced natural language processing (NLP) algorithms, such as pre-trained language models and large language models (LLMs), pioneering studies have explored automating the knowledge tagging process using various machine learning models. In this paper, we investigate the use of a multi-agent system to address the limitations of previous algorithms, particularly in handling complex cases involving intricate knowledge definitions and strict numerical constraints. By demonstrating its superior performance on the publicly available math question knowledge tagging dataset, MathKnowCT, we highlight the significant potential of an LLM-based multi-agent system in overcoming the challenges that previous methods have encountered. Finally, through an in-depth discussion of the implications of automating knowledge tagging, we underscore the promising results of deploying LLM-based algorithms in educational contexts.

Boren Li, Hang Li, Hangxin Liu

Animatronic robots hold the promise of enabling natural human-robot interaction through lifelike facial expressions. However, generating realistic, speech-synchronized robot expressions poses significant challenges due to the complexities of facial biomechanics and the need for responsive motion synthesis. This paper introduces a novel, skinning-centric approach to drive animatronic robot facial expressions from speech input. At its core, the proposed approach employs linear blend skinning (LBS) as a unifying representation, guiding innovations in both embodiment design and motion synthesis. LBS informs the actuation topology, facilitates human expression retargeting, and enables efficient speech-driven facial motion generation. This approach demonstrates the capability to produce highly realistic facial expressions on an animatronic face in real-time at over 4000 fps on a single Nvidia RTX 4090, significantly advancing robots' ability to replicate nuanced human expressions for natural interaction. To foster further research and development in this field, the code has been made publicly available at: \url{https://github.com/library87/OpenRoboExp}.

Feng Wang, Tao Kong, Rufeng Zhang et al.

We present TWIST, a simple and theoretically explainable self-supervised representation learning method by classifying large-scale unlabeled datasets in an end-to-end way. We employ a siamese network terminated by a softmax operation to produce twin class distributions of two augmented images. Without supervision, we enforce the class distributions of different augmentations to be consistent. However, simply minimizing the divergence between augmentations will cause collapsed solutions, i.e., outputting the same class probability distribution for all images. In this case, no information about the input image is left. To solve this problem, we propose to maximize the mutual information between the input and the class predictions. Specifically, we minimize the entropy of the distribution for each sample to make the class prediction for each sample assertive and maximize the entropy of the mean distribution to make the predictions of different samples diverse. In this way, TWIST can naturally avoid the collapsed solutions without specific designs such as asymmetric network, stop-gradient operation, or momentum encoder. As a result, TWIST outperforms state-of-the-art methods on a wide range of tasks. Especially, TWIST performs surprisingly well on semi-supervised learning, achieving 61.2% top-1 accuracy with 1% ImageNet labels using a ResNet-50 as backbone, surpassing previous best results by an absolute improvement of 6.2%. Codes and pre-trained models are given on: https://github.com/bytedance/TWIST

Hang Li, Tianlong Xu, Kaiqi Yang et al.

The rise of large language models (LLMs) offers new opportunities for automatic error detection in education, particularly for math word problems (MWPs). While prior studies demonstrate the promise of LLMs as error detectors, they overlook the presence of multiple valid solutions for a single MWP. Our preliminary analysis reveals a significant performance gap between conventional and alternative solutions in MWPs, a phenomenon we term conformity bias in this work. To mitigate this bias, we introduce the Ask-Before-Detect (AskBD) framework, which generates adaptive reference solutions using LLMs to enhance error detection. Experiments on 200 examples of GSM8K show that AskBD effectively mitigates bias and improves performance, especially when combined with reasoning-enhancing techniques like chain-of-thought prompting.

Yi-Fan Zhang, Hang Li, Dingjie Song et al.

Large Language Models (LLMs), such as GPT-4, have demonstrated impressive mathematical reasoning capabilities, achieving near-perfect performance on benchmarks like GSM8K. However, their application in personalized education remains limited due to an overemphasis on correctness over error diagnosis and feedback generation. Current models fail to provide meaningful insights into the causes of student mistakes, limiting their utility in educational contexts. To address these challenges, we present three key contributions. First, we introduce \textbf{MathCCS} (Mathematical Classification and Constructive Suggestions), a multi-modal benchmark designed for systematic error analysis and tailored feedback. MathCCS includes real-world problems, expert-annotated error categories, and longitudinal student data. Evaluations of state-of-the-art models, including \textit{Qwen2-VL}, \textit{LLaVA-OV}, \textit{Claude-3.5-Sonnet} and \textit{GPT-4o}, reveal that none achieved classification accuracy above 30\% or generated high-quality suggestions (average scores below 4/10), highlighting a significant gap from human-level performance. Second, we develop a sequential error analysis framework that leverages historical data to track trends and improve diagnostic precision. Finally, we propose a multi-agent collaborative framework that combines a Time Series Agent for historical analysis and an MLLM Agent for real-time refinement, enhancing error classification and feedback generation. Together, these contributions provide a robust platform for advancing personalized education, bridging the gap between current AI capabilities and the demands of real-world teaching.

Shaohua Zhang, Yuan Lin, Hang Li

The remarkable success of large language models (LLMs) stems from their ability to consolidate vast amounts of knowledge into the memory during pre-training and to retrieve it from the memory during inference, enabling advanced capabilities such as knowledge memorization, instruction-following and reasoning. However, the mechanisms of memory retrieval and consolidation in LLMs remain poorly understood. In this paper, we propose the function token hypothesis to explain the workings of LLMs: During inference, function tokens activate the most predictive features from context and govern next token prediction (memory retrieval). During pre-training, predicting the next tokens (usually content tokens) that follow function tokens increases the number of learned features of LLMs and updates the model parameters (memory consolidation). Function tokens here roughly correspond to function words in linguistics, including punctuation marks, articles, prepositions, and conjunctions, in contrast to content tokens. We provide extensive experimental evidence supporting this hypothesis. Using bipartite graph analysis, we show that a small number of function tokens activate the majority of features. Case studies further reveal how function tokens activate the most predictive features from context to direct next token prediction. We also find that during pre-training, the training loss is dominated by predicting the next content tokens following function tokens, which forces the function tokens to select the most predictive features from context.

Bo Ma, Hang Li, ZeHua Hu et al.

Multi-agent systems powered by large language models have demonstrated remarkable capabilities across diverse domains, yet existing automated design approaches seek monolithic solutions that fail to adapt resource allocation based on query complexity and domain requirements. This paper introduces AutoMaAS, a self-evolving multi-agent architecture search framework that leverages neural architecture search principles to automatically discover optimal agent configurations through dynamic operator lifecycle management and automated machine learning techniques. Our approach incorporates four key innovations: (1) automatic operator generation, fusion, and elimination based on performance-cost analysis, (2) dynamic cost-aware optimization with real-time parameter adjustment, (3) online feedback integration for continuous architecture refinement, and (4) enhanced interpretability through decision tracing mechanisms. Extensive experiments across six benchmarks demonstrate that AutoMaAS achieves 1.0-7.1\% performance improvement while reducing inference costs by 3-5\% compared to state-of-the-art methods. The framework shows superior transferability across datasets and LLM backbones, establishing a new paradigm for automated multi-agent system design in the era of large language models.

Bo Ma, Hang Li, ZeHua Hu et al.

Contemporary generative recommendation systems face significant challenges in handling multimodal data, eliminating algorithmic biases, and providing transparent decision-making processes. This paper introduces an enhanced generative recommendation framework that addresses these limitations through five key innovations: multimodal fusion architecture, retrieval-augmented generation mechanisms, causal inference-based debiasing, explainable recommendation generation, and real-time adaptive learning capabilities. Our framework leverages advanced large language models as the backbone while incorporating specialized modules for cross-modal understanding, contextual knowledge integration, bias mitigation, explanation synthesis, and continuous model adaptation. Extensive experiments on three benchmark datasets (MovieLens-25M, Amazon-Electronics, Yelp-2023) demonstrate consistent improvements in recommendation accuracy, fairness, and diversity compared to existing approaches. The proposed framework achieves up to 2.3% improvement in NDCG@10 and 1.4% enhancement in diversity metrics while maintaining computational efficiency through optimized inference strategies.

Hang Li, Shengyao Zhuang, Ahmed Mourad et al.

Pseudo-Relevance Feedback (PRF) utilises the relevance signals from the top-k passages from the first round of retrieval to perform a second round of retrieval aiming to improve search effectiveness. A recent research direction has been the study and development of PRF methods for deep language models based rankers, and in particular in the context of dense retrievers. Dense retrievers, compared to more complex neural rankers, provide a trade-off between effectiveness, which is often reduced compared to more complex neural rankers, and query latency, which also is reduced making the retrieval pipeline more efficient. The introduction of PRF methods for dense retrievers has been motivated as an attempt to further improve their effectiveness. In this paper, we reproduce and study a recent method for PRF with dense retrievers, called ANCE-PRF. This method concatenates the query text and that of the top-k feedback passages to form a new query input, which is then encoded into a dense representation using a newly trained query encoder based on the original dense retriever used for the first round of retrieval. While the method can potentially be applied to any of the existing dense retrievers, prior work has studied it only in the context of the ANCE dense retriever. We study the reproducibility of ANCE-PRF in terms of both its training (encoding of the PRF signal) and inference (ranking) steps. We further extend the empirical analysis provided in the original work to investigate the effect of the hyper-parameters that govern the training process and the robustness of the method across these different settings. Finally, we contribute a study of the generalisability of the ANCE-PRF method when dense retrievers other than ANCE are used for the first round of retrieval and for encoding the PRF signal.

Yan Zeng, Xinsong Zhang, Hang Li

Most existing methods in vision language pre-training rely on object-centric features extracted through object detection and make fine-grained alignments between the extracted features and texts. It is challenging for these methods to learn relations among multiple objects. To this end, we propose a new method called X-VLM to perform `multi-grained vision language pre-training.' The key to learning multi-grained alignments is to locate visual concepts in the image given the associated texts, and in the meantime align the texts with the visual concepts, where the alignments are in multi-granularity. Experimental results show that X-VLM effectively leverages the learned multi-grained alignments to many downstream vision language tasks and consistently outperforms state-of-the-art methods.

Rajat Koner, Hang Li, Marcel Hildebrandt et al.

Visual Question Answering (VQA) is concerned with answering free-form questions about an image. Since it requires a deep semantic and linguistic understanding of the question and the ability to associate it with various objects that are present in the image, it is an ambitious task and requires multi-modal reasoning from both computer vision and natural language processing. We propose Graphhopper, a novel method that approaches the task by integrating knowledge graph reasoning, computer vision, and natural language processing techniques. Concretely, our method is based on performing context-driven, sequential reasoning based on the scene entities and their semantic and spatial relationships. As a first step, we derive a scene graph that describes the objects in the image, as well as their attributes and their mutual relationships. Subsequently, a reinforcement learning agent is trained to autonomously navigate in a multi-hop manner over the extracted scene graph to generate reasoning paths, which are the basis for deriving answers. We conduct an experimental study on the challenging dataset GQA, based on both manually curated and automatically generated scene graphs. Our results show that we keep up with a human performance on manually curated scene graphs. Moreover, we find that Graphhopper outperforms another state-of-the-art scene graph reasoning model on both manually curated and automatically generated scene graphs by a significant margin.

Hang Li, Dong Wei, Shilei Cao et al.

Segmentation of objects of interest is one of the central tasks in medical image analysis, which is indispensable for quantitative analysis. When developing machine-learning based methods for automated segmentation, manual annotations are usually used as the ground truth toward which the models learn to mimic. While the bulky parts of the segmentation targets are relatively easy to label, the peripheral areas are often difficult to handle due to ambiguous boundaries and the partial volume effect, etc., and are likely to be labeled with uncertainty. This uncertainty in labeling may, in turn, result in unsatisfactory performance of the trained models. In this paper, we propose superpixel-based label softening to tackle the above issue. Generated by unsupervised over-segmentation, each superpixel is expected to represent a locally homogeneous area. If a superpixel intersects with the annotation boundary, we consider a high probability of uncertain labeling within this area. Driven by this intuition, we soften labels in this area based on signed distances to the annotation boundary and assign probability values within [0, 1] to them, in comparison with the original "hard", binary labels of either 0 or 1. The softened labels are then used to train the segmentation models together with the hard labels. Experimental results on a brain MRI dataset and an optical coherence tomography dataset demonstrate that this conceptually simple and implementation-wise easy method achieves overall superior segmentation performances to baseline and comparison methods for both 3D and 2D medical images.

Marcel Hildebrandt, Hang Li, Rajat Koner et al.

Visual question answering is concerned with answering free-form questions about an image. Since it requires a deep linguistic understanding of the question and the ability to associate it with various objects that are present in the image, it is an ambitious task and requires techniques from both computer vision and natural language processing. We propose a novel method that approaches the task by performing context-driven, sequential reasoning based on the objects and their semantic and spatial relationships present in the scene. As a first step, we derive a scene graph which describes the objects in the image, as well as their attributes and their mutual relationships. A reinforcement agent then learns to autonomously navigate over the extracted scene graph to generate paths, which are then the basis for deriving answers. We conduct a first experimental study on the challenging GQA dataset with manually curated scene graphs, where our method almost reaches the level of human performance.

Hang Li, Chen Ma, Wei Xu et al.

Building compact convolutional neural networks (CNNs) with reliable performance is a critical but challenging task, especially when deploying them in real-world applications. As a common approach to reduce the size of CNNs, pruning methods delete part of the CNN filters according to some metrics such as $l1$-norm. However, previous methods hardly leverage the information variance in a single feature map and the similarity characteristics among feature maps. In this paper, we propose a novel filter pruning method, which incorporates two kinds of feature map selections: diversity-aware selection (DFS) and similarity-aware selection (SFS). DFS aims to discover features with low information diversity while SFS removes features that have high similarities with others. We conduct extensive empirical experiments with various CNN architectures on publicly available datasets. The experimental results demonstrate that our model obtains up to 91.6% parameter decrease and 83.7% FLOPs reduction with almost no accuracy loss.

Ziniu Hu, Yang Wang, Qu Peng et al.

Although click data is widely used in search systems in practice, so far the inherent bias, most notably position bias, has prevented it from being used in training of a ranker for search, i.e., learning-to-rank. Recently, a number of authors have proposed new techniques referred to as 'unbiased learning-to-rank', which can reduce position bias and train a relatively high-performance ranker using click data. Most of the algorithms, based on the inverse propensity weighting (IPW) principle, first estimate the click bias at each position, and then train an unbiased ranker with the estimated biases using a learning-to-rank algorithm. However, there has not been a method for pairwise learning-to-rank that can jointly conduct debiasing of click data and training of a ranker using a pairwise loss function. In this paper, we propose a novel algorithm, which can jointly estimate the biases at click positions and the biases at unclick positions, and learn an unbiased ranker. Experiments on benchmark data show that our algorithm can significantly outperform existing algorithms. In addition, an online A/B Testing at a commercial search engine shows that our algorithm can effectively conduct debiasing of click data and enhance relevance ranking.

Qian Wang, Hang Li, Zhi Chen et al.

From the viewpoint of physical-layer authentication, spoofing attacks can be foiled by checking channel state information (CSI). Existing CSI-based authentication algorithms mostly require a deep knowledge of the channel to deliver decent performance. In this paper, we investigate CSI-based authenticators that can spare the effort to predetermine channel properties by utilizing deep neural networks (DNNs). We first propose a convolutional neural network (CNN)-enabled authenticator that is able to extract the local features in CSI. Next, we employ the recurrent neural network (RNN) to capture the dependencies between different frequencies in CSI. In addition, we propose to use the convolutional recurrent neural network (CRNN)---a combination of the CNN and the RNN---to learn local and contextual information in CSI for user authentication. To effectively train these DNNs, one needs a large amount of labeled channel records. However, it is often expensive to label large channel observations in the presence of a spoofer. In view of this, we further study a case in which only a small part of the the channel observations are labeled. To handle it, we extend these DNNs-enabled approaches into semi-supervised ones. This extension is based on a semi-supervised learning technique that employs both the labeled and unlabeled data to train a DNN. To be specific, our semi-supervised method begins by generating pseudo labels for the unlabeled channel samples through implementing the K-means algorithm in a semi-supervised manner. Subsequently, both the labeled and pseudo labeled data are exploited to pre-train a DNN, which is then fine-tuned based on the labeled channel records.

Zhenguo Li, Fengwei Zhou, Fei Chen et al.

Few-shot learning is challenging for learning algorithms that learn each task in isolation and from scratch. In contrast, meta-learning learns from many related tasks a meta-learner that can learn a new task more accurately and faster with fewer examples, where the choice of meta-learners is crucial. In this paper, we develop Meta-SGD, an SGD-like, easily trainable meta-learner that can initialize and adapt any differentiable learner in just one step, on both supervised learning and reinforcement learning. Compared to the popular meta-learner LSTM, Meta-SGD is conceptually simpler, easier to implement, and can be learned more efficiently. Compared to the latest meta-learner MAML, Meta-SGD has a much higher capacity by learning to learn not just the learner initialization, but also the learner update direction and learning rate, all in a single meta-learning process. Meta-SGD shows highly competitive performance for few-shot learning on regression, classification, and reinforcement learning.

Hao Zhou, Zhaopeng Tu, Shujian Huang et al.

In typical neural machine translation~(NMT), the decoder generates a sentence word by word, packing all linguistic granularities in the same time-scale of RNN. In this paper, we propose a new type of decoder for NMT, which splits the decode state into two parts and updates them in two different time-scales. Specifically, we first predict a chunk time-scale state for phrasal modeling, on top of which multiple word time-scale states are generated. In this way, the target sentence is translated hierarchically from chunks to words, with information in different granularities being leveraged. Experiments show that our proposed model significantly improves the translation performance over the state-of-the-art NMT model.

Zhaopeng Tu, Yang Liu, Lifeng Shang et al.

Although end-to-end Neural Machine Translation (NMT) has achieved remarkable progress in the past two years, it suffers from a major drawback: translations generated by NMT systems often lack of adequacy. It has been widely observed that NMT tends to repeatedly translate some source words while mistakenly ignoring other words. To alleviate this problem, we propose a novel encoder-decoder-reconstructor framework for NMT. The reconstructor, incorporated into the NMT model, manages to reconstruct the input source sentence from the hidden layer of the output target sentence, to ensure that the information in the source side is transformed to the target side as much as possible. Experiments show that the proposed framework significantly improves the adequacy of NMT output and achieves superior translation result over state-of-the-art NMT and statistical MT systems.

Xing Wang, Zhengdong Lu, Zhaopeng Tu et al.

Neural Machine Translation (NMT) is a new approach to machine translation that has made great progress in recent years. However, recent studies show that NMT generally produces fluent but inadequate translations (Tu et al. 2016b; Tu et al. 2016a; He et al. 2016; Tu et al. 2017). This is in contrast to conventional Statistical Machine Translation (SMT), which usually yields adequate but non-fluent translations. It is natural, therefore, to leverage the advantages of both models for better translations, and in this work we propose to incorporate SMT model into NMT framework. More specifically, at each decoding step, SMT offers additional recommendations of generated words based on the decoding information from NMT (e.g., the generated partial translation and attention history). Then we employ an auxiliary classifier to score the SMT recommendations and a gating function to combine the SMT recommendations with NMT generations, both of which are jointly trained within the NMT architecture in an end-to-end manner. Experimental results on Chinese-English translation show that the proposed approach achieves significant and consistent improvements over state-of-the-art NMT and SMT systems on multiple NIST test sets.

Zhaopeng Tu, Yang Liu, Zhengdong Lu et al.

In neural machine translation (NMT), generation of a target word depends on both source and target contexts. We find that source contexts have a direct impact on the adequacy of a translation while target contexts affect the fluency. Intuitively, generation of a content word should rely more on the source context and generation of a functional word should rely more on the target context. Due to the lack of effective control over the influence from source and target contexts, conventional NMT tends to yield fluent but inadequate translations. To address this problem, we propose context gates which dynamically control the ratios at which source and target contexts contribute to the generation of target words. In this way, we can enhance both the adequacy and fluency of NMT with more careful control of the information flow from contexts. Experiments show that our approach significantly improves upon a standard attention-based NMT system by +2.3 BLEU points.

Mingxuan Wang, Zhengdong Lu, Hang Li et al.

We propose to enhance the RNN decoder in a neural machine translator (NMT) with external memory, as a natural but powerful extension to the state in the decoding RNN. This memory-enhanced RNN decoder is called \textsc{MemDec}. At each time during decoding, \textsc{MemDec} will read from this memory and write to this memory once, both with content-based addressing. Unlike the unbounded memory in previous work\cite{RNNsearch} to store the representation of source sentence, the memory in \textsc{MemDec} is a matrix with pre-determined size designed to better capture the information important for the decoding process at each time step. Our empirical study on Chinese-English translation shows that it can improve by $4.8$ BLEU upon Groundhog and $5.3$ BLEU upon on Moses, yielding the best performance achieved with the same training set.

Fandong Meng, Zhengdong Lu, Zhaopeng Tu et al.

We propose DEEPMEMORY, a novel deep architecture for sequence-to-sequence learning, which performs the task through a series of nonlinear transformations from the representation of the input sequence (e.g., a Chinese sentence) to the final output sequence (e.g., translation to English). Inspired by the recently proposed Neural Turing Machine (Graves et al., 2014), we store the intermediate representations in stacked layers of memories, and use read-write operations on the memories to realize the nonlinear transformations between the representations. The types of transformations are designed in advance but the parameters are learned from data. Through layer-by-layer transformations, DEEPMEMORY can model complicated relations between sequences necessary for applications such as machine translation between distant languages. The architecture can be trained with normal back-propagation on sequenceto-sequence data, and the learning can be easily scaled up to a large corpus. DEEPMEMORY is broad enough to subsume the state-of-the-art neural translation model in (Bahdanau et al., 2015) as its special case, while significantly improving upon the model with its deeper architecture. Remarkably, DEEPMEMORY, being purely neural network-based, can achieve performance comparable to the traditional phrase-based machine translation system Moses with a small vocabulary and a modest parameter size.

Lin Ma, Zhengdong Lu, Hang Li

In this paper, we propose to employ the convolutional neural network (CNN) for the image question answering (QA). Our proposed CNN provides an end-to-end framework with convolutional architectures for learning not only the image and question representations, but also their inter-modal interactions to produce the answer. More specifically, our model consists of three CNNs: one image CNN to encode the image content, one sentence CNN to compose the words of the question, and one multimodal convolution layer to learn their joint representation for the classification in the space of candidate answer words. We demonstrate the efficacy of our proposed model on the DAQUAR and COCO-QA datasets, which are two benchmark datasets for the image QA, with the performances significantly outperforming the state-of-the-art.

Lin Ma, Zhengdong Lu, Lifeng Shang et al.

In this paper, we propose multimodal convolutional neural networks (m-CNNs) for matching image and sentence. Our m-CNN provides an end-to-end framework with convolutional architectures to exploit image representation, word composition, and the matching relations between the two modalities. More specifically, it consists of one image CNN encoding the image content, and one matching CNN learning the joint representation of image and sentence. The matching CNN composes words to different semantic fragments and learns the inter-modal relations between image and the composed fragments at different levels, thus fully exploit the matching relations between image and sentence. Experimental results on benchmark databases of bidirectional image and sentence retrieval demonstrate that the proposed m-CNNs can effectively capture the information necessary for image and sentence matching. Specifically, our proposed m-CNNs for bidirectional image and sentence retrieval on Flickr30K and Microsoft COCO databases achieve the state-of-the-art performances.