Weiping Wang

Bo Fang, Wenhao Wu, Chang Liu et al. · amazon-science

With the explosive growth of web videos and emerging large-scale vision-language pre-training models, e.g., CLIP, retrieving videos of interest with text instructions has attracted increasing attention. A common practice is to transfer text-video pairs to the same embedding space and craft cross-modal interactions with certain entities in specific granularities for semantic correspondence. Unfortunately, the intrinsic uncertainties of optimal entity combinations in appropriate granularities for cross-modal queries are understudied, which is especially critical for modalities with hierarchical semantics, e.g., video, text, etc. In this paper, we propose an Uncertainty-Adaptive Text-Video Retrieval approach, termed UATVR, which models each look-up as a distribution matching procedure. Concretely, we add additional learnable tokens in the encoders to adaptively aggregate multi-grained semantics for flexible high-level reasoning. In the refined embedding space, we represent text-video pairs as probabilistic distributions where prototypes are sampled for matching evaluation. Comprehensive experiments on four benchmarks justify the superiority of our UATVR, which achieves new state-of-the-art results on MSR-VTT (50.8%), VATEX (64.5%), MSVD (49.7%), and DiDeMo (45.8%). The code is available at https://github.com/bofang98/UATVR.

Yuanxin Liu, Fandong Meng, Zheng Lin et al. · pku, tsinghua

Despite the remarkable success of pre-trained language models (PLMs), they still face two challenges: First, large-scale PLMs are inefficient in terms of memory footprint and computation. Second, on the downstream tasks, PLMs tend to rely on the dataset bias and struggle to generalize to out-of-distribution (OOD) data. In response to the efficiency problem, recent studies show that dense PLMs can be replaced with sparse subnetworks without hurting the performance. Such subnetworks can be found in three scenarios: 1) the fine-tuned PLMs, 2) the raw PLMs and then fine-tuned in isolation, and even inside 3) PLMs without any parameter fine-tuning. However, these results are only obtained in the in-distribution (ID) setting. In this paper, we extend the study on PLMs subnetworks to the OOD setting, investigating whether sparsity and robustness to dataset bias can be achieved simultaneously. To this end, we conduct extensive experiments with the pre-trained BERT model on three natural language understanding (NLU) tasks. Our results demonstrate that \textbf{sparse and robust subnetworks (SRNets) can consistently be found in BERT}, across the aforementioned three scenarios, using different training and compression methods. Furthermore, we explore the upper bound of SRNets using the OOD information and show that \textbf{there exist sparse and almost unbiased BERT subnetworks}. Finally, we present 1) an analytical study that provides insights on how to promote the efficiency of SRNets searching process and 2) a solution to improve subnetworks' performance at high sparsity. The code is available at https://github.com/llyx97/sparse-and-robust-PLM.

Jiangnan Li, Fandong Meng, Zheng Lin et al. · tsinghua

Conversational Causal Emotion Entailment aims to detect causal utterances for a non-neutral targeted utterance from a conversation. In this work, we build conversations as graphs to overcome implicit contextual modelling of the original entailment style. Following the previous work, we further introduce the emotion information into graphs. Emotion information can markedly promote the detection of causal utterances whose emotion is the same as the targeted utterance. However, it is still hard to detect causal utterances with different emotions, especially neutral ones. The reason is that models are limited in reasoning causal clues and passing them between utterances. To alleviate this problem, we introduce social commonsense knowledge (CSK) and propose a Knowledge Enhanced Conversation graph (KEC). KEC propagates the CSK between two utterances. As not all CSK is emotionally suitable for utterances, we therefore propose a sentiment-realized knowledge selecting strategy to filter CSK. To process KEC, we further construct the Knowledge Enhanced Directed Acyclic Graph networks. Experimental results show that our method outperforms baselines and infers more causes with different emotions from the targeted utterance.

Qingyi Si, Yuanxin Liu, Zheng Lin et al. · pku

Despite the excellent performance of vision-language pre-trained models (VLPs) on conventional VQA task, they still suffer from two problems: First, VLPs tend to rely on language biases in datasets and fail to generalize to out-of-distribution (OOD) data. Second, they are inefficient in terms of memory footprint and computation. Although promising progress has been made in both problems, most existing works tackle them independently. To facilitate the application of VLP to VQA tasks, it is imperative to jointly study VLP compression and OOD robustness, which, however, has not yet been explored. This paper investigates whether a VLP can be compressed and debiased simultaneously by searching sparse and robust subnetworks. To this end, we systematically study the design of a training and compression pipeline to search the subnetworks, as well as the assignment of sparsity to different modality-specific modules. Our experiments involve 3 VLPs, 2 compression methods, 4 training methods, 2 datasets and a range of sparsity levels and random seeds. Our results show that there indeed exist sparse and robust subnetworks, which are competitive with the debiased full VLP and clearly outperform the debiasing SoTAs with fewer parameters on OOD datasets VQA-CP v2 and VQA-VS. The codes can be found at https://github.com/PhoebusSi/Compress-Robust-VQA.

Qingyi Si, Yuanxin Liu, Fandong Meng et al. · pku, tsinghua

Models for Visual Question Answering (VQA) often rely on the spurious correlations, i.e., the language priors, that appear in the biased samples of training set, which make them brittle against the out-of-distribution (OOD) test data. Recent methods have achieved promising progress in overcoming this problem by reducing the impact of biased samples on model training. However, these models reveal a trade-off that the improvements on OOD data severely sacrifice the performance on the in-distribution (ID) data (which is dominated by the biased samples). Therefore, we propose a novel contrastive learning approach, MMBS, for building robust VQA models by Making the Most of Biased Samples. Specifically, we construct positive samples for contrastive learning by eliminating the information related to spurious correlation from the original training samples and explore several strategies to use the constructed positive samples for training. Instead of undermining the importance of biased samples in model training, our approach precisely exploits the biased samples for unbiased information that contributes to reasoning. The proposed method is compatible with various VQA backbones. We validate our contributions by achieving competitive performance on the OOD dataset VQA-CP v2 while preserving robust performance on the ID dataset VQA v2.

Xugong Qin, Pengyuan Lyu, Chengquan Zhang et al.

Due to the flexible representation of arbitrary-shaped scene text and simple pipeline, bottom-up segmentation-based methods begin to be mainstream in real-time scene text detection. Despite great progress, these methods show deficiencies in robustness and still suffer from false positives and instance adhesion. Different from existing methods which integrate multiple-granularity features or multiple outputs, we resort to the perspective of representation learning in which auxiliary tasks are utilized to enable the encoder to jointly learn robust features with the main task of per-pixel classification during optimization. For semantic representation learning, we propose global-dense semantic contrast (GDSC), in which a vector is extracted for global semantic representation, then used to perform element-wise contrast with the dense grid features. To learn instance-aware representation, we propose to combine top-down modeling (TDM) with the bottom-up framework to provide implicit instance-level clues for the encoder. With the proposed GDSC and TDM, the encoder network learns stronger representation without introducing any parameters and computations during inference. Equipped with a very light decoder, the detector can achieve more robust real-time scene text detection. Experimental results on four public datasets show that the proposed method can outperform or be comparable to the state-of-the-art on both accuracy and speed. Specifically, the proposed method achieves 87.2% F-measure with 48.2 FPS on Total-Text and 89.6% F-measure with 36.9 FPS on MSRA-TD500 on a single GeForce RTX 2080 Ti GPU.

Chenxu Yang, Chuanyu Qin, Qingyi Si et al.

On-policy distillation (OPD) has become a popular training paradigm in the LLM community. This paradigm selects a larger model as the teacher to provide dense, fine-grained signals for each sampled trajectory, in contrast to reinforcement learning with verifiable rewards (RLVR), which only obtains sparse signals from verifiable outcomes in the environment. Recently, the community has explored on-policy self-distillation (OPSD), where the same model serves as both teacher and student, with the teacher receiving additional privileged information such as reference answers to enable self-evolution. This paper demonstrates that learning signals solely derived from the privileged teacher result in severe information leakage and unstable long-term training. Accordingly, we identify the optimal niche for self-distillation and propose \textbf{RLSD} (\textbf{RL}VR with \textbf{S}elf-\textbf{D}istillation). Specifically, we leverage self-distillation to obtain token-level policy differences for determining fine-grained update magnitudes, while continuing to use RLVR to derive reliable update directions from environmental feedback (e.g., response correctness). This enables RLSD to simultaneously harness the strengths of both RLVR and OPSD, achieving a higher convergence ceiling and superior training stability.

Yifei Zhang, Chang Liu, Yu Zhou et al.

Contrastive self-supervised learning (CSL) based on instance discrimination typically attracts positive samples while repelling negatives to learn representations with pre-defined binary self-supervision. However, vanilla CSL is inadequate in modeling sophisticated instance relations, limiting the learned model to retain fine semantic structure. On the one hand, samples with the same semantic category are inevitably pushed away as negatives. On the other hand, differences among samples cannot be captured. In this paper, we present relation-aware contrastive self-supervised learning (ReCo) to integrate instance relations, i.e., global distribution relation and local interpolation relation, into the CSL framework in a plug-and-play fashion. Specifically, we align similarity distributions calculated between the positive anchor views and the negatives at the global level to exploit diverse similarity relations among instances. Local-level interpolation consistency between the pixel space and the feature space is applied to quantitatively model the feature differences of samples with distinct apparent similarities. Through explicitly instance relation modeling, our ReCo avoids irrationally pushing away semantically identical samples and carves a well-structured feature space. Extensive experiments conducted on commonly used benchmarks justify that our ReCo consistently gains remarkable performance improvements.

Youhui Guo, Yu Zhou, Xugong Qin et al.

Recent scene text detection methods are almost based on deep learning and data-driven. Synthetic data is commonly adopted for pre-training due to expensive annotation cost. However, there are obvious domain discrepancies between synthetic data and real-world data. It may lead to sub-optimal performance to directly adopt the model initialized by synthetic data in the fine-tuning stage. In this paper, we propose a new training paradigm for scene text detection, which introduces an \textbf{UN}supervised \textbf{I}ntermediate \textbf{T}raining \textbf{S}tage (UNITS) that builds a buffer path to real-world data and can alleviate the gap between the pre-training stage and fine-tuning stage. Three training strategies are further explored to perceive information from real-world data in an unsupervised way. With UNITS, scene text detectors are improved without introducing any parameters and computations during inference. Extensive experimental results show consistent performance improvements on three public datasets.

Chengyang Fang, Gangyan Zeng, Yu Zhou et al.

Texts in scene images convey critical information for scene understanding and reasoning. The abilities of reading and reasoning matter for the model in the text-based visual question answering (TextVQA) process. However, current TextVQA models do not center on the text and suffer from several limitations. The model is easily dominated by language biases and optical character recognition (OCR) errors due to the absence of semantic guidance in the answer prediction process. In this paper, we propose a novel Semantics-Centered Network (SC-Net) that consists of an instance-level contrastive semantic prediction module (ICSP) and a semantics-centered transformer module (SCT). Equipped with the two modules, the semantics-centered model can resist the language biases and the accumulated errors from OCR. Extensive experiments on TextVQA and ST-VQA datasets show the effectiveness of our model. SC-Net surpasses previous works with a noticeable margin and is more reasonable for the TextVQA task.

Yufan Liu, Wanqian Zhang, Dayan Wu et al.

Model inversion (MI) attack reconstructs the private training data of a target model given its output, posing a significant threat to deep learning models and data privacy. On one hand, most of existing MI methods focus on searching for latent codes to represent the target identity, yet this iterative optimization-based scheme consumes a huge number of queries to the target model, making it unrealistic especially in black-box scenario. On the other hand, some training-based methods launch an attack through a single forward inference, whereas failing to directly learn high-level mappings from prediction vectors to images. Addressing these limitations, we propose a novel Prediction-to-Image (P2I) method for black-box MI attack. Specifically, we introduce the Prediction Alignment Encoder to map the target model's output prediction into the latent code of StyleGAN. In this way, prediction vector space can be well aligned with the more disentangled latent space, thus establishing a connection between prediction vectors and the semantic facial features. During the attack phase, we further design the Aligned Ensemble Attack scheme to integrate complementary facial attributes of target identity for better reconstruction. Experimental results show that our method outperforms other SOTAs, e.g.,compared with RLB-MI, our method improves attack accuracy by 8.5% and reduces query numbers by 99% on dataset CelebA.

Mingyu Zheng, Xinwei Feng, Qingyi Si et al.

Although great progress has been made by previous table understanding methods including recent approaches based on large language models (LLMs), they rely heavily on the premise that given tables must be converted into a certain text sequence (such as Markdown or HTML) to serve as model input. However, it is difficult to access such high-quality textual table representations in some real-world scenarios, and table images are much more accessible. Therefore, how to directly understand tables using intuitive visual information is a crucial and urgent challenge for developing more practical applications. In this paper, we propose a new problem, multimodal table understanding, where the model needs to generate correct responses to various table-related requests based on the given table image. To facilitate both the model training and evaluation, we construct a large-scale dataset named MMTab, which covers a wide spectrum of table images, instructions and tasks. On this basis, we develop Table-LLaVA, a generalist tabular multimodal large language model (MLLM), which significantly outperforms recent open-source MLLM baselines on 23 benchmarks under held-in and held-out settings. The code and data is available at this https://github.com/SpursGoZmy/Table-LLaVA

Qingyi Si, Yuchen Mo, Zheng Lin et al.

Outside-knowledge visual question answering is a challenging task that requires both the acquisition and the use of open-ended real-world knowledge. Some existing solutions draw external knowledge into the cross-modality space which overlooks the much vaster textual knowledge in natural-language space, while others transform the image into a text that further fuses with the textual knowledge into the natural-language space and completely abandons the use of visual features. In this paper, we are inspired to constrain the cross-modality space into the same space of natural-language space which makes the visual features preserved directly, and the model still benefits from the vast knowledge in natural-language space. To this end, we propose a novel framework consisting of a multimodal encoder, a textual encoder and an answer decoder. Such structure allows us to introduce more types of knowledge including explicit and implicit multimodal and textual knowledge. Extensive experiments validate the superiority of the proposed method which outperforms the state-of-the-art by 6.17% accuracy. We also conduct comprehensive ablations of each component, and systematically study the roles of varying types of knowledge. Codes and knowledge data can be found at https://github.com/PhoebusSi/Thinking-while-Observing.

Zhi Qiao, Yu Zhou, Jin Wei et al.

Nowadays, scene text recognition has attracted more and more attention due to its various applications. Most state-of-the-art methods adopt an encoder-decoder framework with attention mechanism, which generates text autoregressively from left to right. Despite the convincing performance, the speed is limited because of the one-by-one decoding strategy. As opposed to autoregressive models, non-autoregressive models predict the results in parallel with a much shorter inference time, but the accuracy falls behind the autoregressive counterpart considerably. In this paper, we propose a Parallel, Iterative and Mimicking Network (PIMNet) to balance accuracy and efficiency. Specifically, PIMNet adopts a parallel attention mechanism to predict the text faster and an iterative generation mechanism to make the predictions more accurate. In each iteration, the context information is fully explored. To improve learning of the hidden layer, we exploit the mimicking learning in the training phase, where an additional autoregressive decoder is adopted and the parallel decoder mimics the autoregressive decoder with fitting outputs of the hidden layer. With the shared backbone between the two decoders, the proposed PIMNet can be trained end-to-end without pre-training. During inference, the branch of the autoregressive decoder is removed for a faster speed. Extensive experiments on public benchmarks demonstrate the effectiveness and efficiency of PIMNet. Our code will be available at https://github.com/Pay20Y/PIMNet.

Yifei Zhang, Chang Liu, Yu Zhou et al.

In unsupervised feature learning, sample specificity based methods ignore the inter-class information, which deteriorates the discriminative capability of representation models. Clustering based methods are error-prone to explore the complete class boundary information due to the inevitable class inconsistent samples in each cluster. In this work, we propose a novel clustering based method, which, by iteratively excluding class inconsistent samples during progressive cluster formation, alleviates the impact of noise samples in a simple-yet-effective manner. Our approach, referred to as Progressive Cluster Purification (PCP), implements progressive clustering by gradually reducing the number of clusters during training, while the sizes of clusters continuously expand consistently with the growth of model representation capability. With a well-designed cluster purification mechanism, it further purifies clusters by filtering noise samples which facilitate the subsequent feature learning by utilizing the refined clusters as pseudo-labels. Experiments on commonly used benchmarks demonstrate that the proposed PCP improves baseline method with significant margins. Our code will be available at https://github.com/zhangyifei0115/PCP.

Chenxu Yang, Qingyi Si, Yongjie Duan et al.

Recent advances in large reasoning language models (LRLMs) rely on test-time scaling, which extends long chain-of-thought (CoT) generation to solve complex tasks. However, overthinking in long CoT not only slows down the efficiency of problem solving, but also risks accuracy loss due to the extremely detailed or redundant reasoning steps. We propose a simple yet effective method that allows LLMs to self-truncate CoT sequences by early exit during generation. Instead of relying on fixed heuristics, the proposed method monitors model behavior at potential reasoning transition points and dynamically terminates the next reasoning chain's generation when the model exhibits high confidence in a trial answer. Our method requires no additional training and can be seamlessly integrated into existing o1-like reasoning LLMs. Experiments on 10 reasoning benchmarks (e.g., GSM8K, MATH-500, AMC, GPQA, AIME and LiveCodeBench) show that the proposed method is consistently effective on 11 cutting-edge reasoning LLMs of varying series and sizes, reducing the length of CoT sequences by an average of 19.1% to 80.1% while improving accuracy by 0.3% to 5.0%.

Yufan Liu, Jinyang An, Wanqian Zhang et al.

The remarkable development of text-to-image generation models has raised notable security concerns, such as the infringement of portrait rights and the generation of inappropriate content. Concept erasure has been proposed to remove the model's knowledge about protected and inappropriate concepts. Although many methods have tried to balance the efficacy (erasing target concepts) and specificity (retaining irrelevant concepts), they can still generate abundant erasure concepts under the steering of semantically related inputs. In this work, we propose RealEra to address this "concept residue" issue. Specifically, we first introduce the mechanism of neighbor-concept mining, digging out the associated concepts by adding random perturbation into the embedding of erasure concept, thus expanding the erasing range and eliminating the generations even through associated concept inputs. Furthermore, to mitigate the negative impact on the generation of irrelevant concepts caused by the expansion of erasure scope, RealEra preserves the specificity through the beyond-concept regularization. This makes irrelevant concepts maintain their corresponding spatial position, thereby preserving their normal generation performance. We also employ the closed-form solution to optimize weights of U-Net for the cross-attention alignment, as well as the prediction noise alignment with the LoRA module. Extensive experiments on multiple benchmarks demonstrate that RealEra outperforms previous concept erasing methods in terms of superior erasing efficacy, specificity, and generality. More details are available on our project page https://realerasing.github.io/RealEra/ .

Lanrui Wang, Mingyu Zheng, Hongyin Tang et al.

Processing structured tabular data, particularly large and lengthy tables, constitutes a fundamental yet challenging task for large language models (LLMs). However, existing long-context benchmarks like Needle-in-a-Haystack primarily focus on unstructured text, neglecting the challenge of diverse structured tables. Meanwhile, previous tabular benchmarks mainly consider downstream tasks that require high-level reasoning abilities, and overlook models' underlying fine-grained perception of individual table cells, which is crucial for practical and robust LLM-based table applications. To address this gap, we introduce \textsc{NeedleInATable} (NIAT), a new long-context tabular benchmark that treats each table cell as a ``needle'' and requires models to extract the target cell based on cell locations or lookup questions. Our comprehensive evaluation of various LLMs and multimodal LLMs reveals a substantial performance gap between popular downstream tabular tasks and the simpler NIAT task, suggesting that they may rely on dataset-specific correlations or shortcuts to obtain better benchmark results but lack truly robust long-context understanding towards structured tables. Furthermore, we demonstrate that using synthesized NIAT training data can effectively improve performance on both NIAT task and downstream tabular tasks, which validates the importance of NIAT capability for LLMs' genuine table understanding ability.

Jiaxuan Zhao, Naibin Gu, Yuchen Feng et al.

The high costs of customizing large language models (LLMs) fundamentally limit their adaptability to user-specific needs. Consequently, LLMs are increasingly offered as cloud-based services, a paradigm that introduces critical limitations: providers struggle to support personalized customization at scale, while users face privacy risks when exposing sensitive data. To address this dual challenge, we propose Customized Black-box Prompt Tuning (CBP-Tuning), a novel framework that facilitates efficient local customization while preserving bidirectional privacy. Specifically, we design a two-stage framework: (1) a prompt generator trained on the server-side to capture domain-specific and task-agnostic capabilities, and (2) user-side gradient-free optimization that tailors soft prompts for individual tasks. This approach eliminates the need for users to access model weights or upload private data, requiring only a single customized vector per task while achieving effective adaptation. Furthermore, the evaluation of CBP-Tuning in the commonsense reasoning, medical and financial domain settings demonstrates superior performance compared to baselines, showcasing its advantages in task-agnostic processing and privacy preservation.

Yisu Liu, Chenxing Li, Wanqian Zhang et al.

Controllable text-to-audio generation aims to synthesize audio from textual descriptions while satisfying user-specified constraints, including event types, temporal sequences, and onset and offset timestamps. This enables precise control over both the content and temporal structure of the generated audio. Despite recent progress, existing methods still face inherent trade-offs among accurate temporal localization, open-vocabulary scalability, and practical efficiency. To address these challenges, we propose DegDiT, a novel dynamic event graph-guided diffusion transformer framework for open-vocabulary controllable audio generation. DegDiT encodes the events in the description as structured dynamic graphs. The nodes in each graph are designed to represent three aspects: semantic features, temporal attributes, and inter-event connections. A graph transformer is employed to integrate these nodes and produce contextualized event embeddings that serve as guidance for the diffusion model. To ensure high-quality and diverse training data, we introduce a quality-balanced data selection pipeline that combines hierarchical event annotation with multi-criteria quality scoring, resulting in a curated dataset with semantic diversity. Furthermore, we present consensus preference optimization, facilitating audio generation through consensus among multiple reward signals. Extensive experiments on AudioCondition, DESED, and AudioTime datasets demonstrate that DegDiT achieves state-of-the-art performances across a variety of objective and subjective evaluation metrics.

Rong Yin, Ruyue Liu, Xiaoshuai Hao et al.

Accurate extraction of molecular representations is a critical step in the drug discovery process. In recent years, significant progress has been made in molecular representation learning methods, among which multi-modal molecular representation methods based on images, and 2D/3D topologies have become increasingly mainstream. However, existing these multi-modal approaches often directly fuse information from different modalities, overlooking the potential of intermodal interactions and failing to adequately capture the complex higher-order relationships and invariant features between molecules. To overcome these challenges, we propose a structure-awareness-based multi-modal self-supervised molecular representation pre-training framework (MMSA) designed to enhance molecular graph representations by leveraging invariant knowledge between molecules. The framework consists of two main modules: the multi-modal molecular representation learning module and the structure-awareness module. The multi-modal molecular representation learning module collaboratively processes information from different modalities of the same molecule to overcome intermodal differences and generate a unified molecular embedding. Subsequently, the structure-awareness module enhances the molecular representation by constructing a hypergraph structure to model higher-order correlations between molecules. This module also introduces a memory mechanism for storing typical molecular representations, aligning them with memory anchors in the memory bank to integrate invariant knowledge, thereby improving the model generalization ability. Extensive experiments have demonstrated the effectiveness of MMSA, which achieves state-of-the-art performance on the MoleculeNet benchmark, with average ROC-AUC improvements ranging from 1.8% to 9.6% over baseline methods.

Xunyu Zhu, Jian Li, Can Ma et al.

Large Language Models (LLMs) demonstrate exceptional reasoning capabilities, often achieving state-of-the-art performance in various tasks. However, their substantial computational and memory demands, due to billions of parameters, hinder deployment in resource-constrained environments. A promising solution is knowledge distillation, where LLMs transfer reasoning capabilities to Small Language Models (SLMs, $\le$ 1B parameters), enabling wider deployment on low-resource devices. Existing methods primarily focus on generating high-quality reasoning rationales for distillation datasets but often neglect the critical role of data quantity and quality. To address these challenges, we propose a Feedback-Driven Distillation (FDD) framework to enhance SLMs' mathematical reasoning capabilities. In the initialization stage, a distillation dataset is constructed by prompting LLMs to pair mathematical problems with corresponding reasoning rationales. We classify problems into easy and hard categories based on SLM performance. For easy problems, LLMs generate more complex variations, while for hard problems, new questions of similar complexity are synthesized. In addition, we propose a multi-round distillation paradigm to iteratively enrich the distillation datasets, thereby progressively improving the mathematical reasoning abilities of SLMs. Experimental results demonstrate that our method can make SLMs achieve SOTA mathematical reasoning performance.

Xiaohua Chen, Yucan Zhou, Dayan Wu et al.

Real-world data often follows a long-tailed distribution, which makes the performance of existing classification algorithms degrade heavily. A key issue is that samples in tail categories fail to depict their intra-class diversity. Humans can imagine a sample in new poses, scenes, and view angles with their prior knowledge even if it is the first time to see this category. Inspired by this, we propose a novel reasoning-based implicit semantic data augmentation method to borrow transformation directions from other classes. Since the covariance matrix of each category represents the feature transformation directions, we can sample new directions from similar categories to generate definitely different instances. Specifically, the long-tailed distributed data is first adopted to train a backbone and a classifier. Then, a covariance matrix for each category is estimated, and a knowledge graph is constructed to store the relations of any two categories. Finally, tail samples are adaptively enhanced via propagating information from all the similar categories in the knowledge graph. Experimental results on CIFAR-100-LT, ImageNet-LT, and iNaturalist 2018 have demonstrated the effectiveness of our proposed method compared with the state-of-the-art methods.

Xiaoni Li, Yu Zhou, Yifei Zhang et al.

Self-supervised representation learning for visual pre-training has achieved remarkable success with sample (instance or pixel) discrimination and semantics discovery of instance, whereas there still exists a non-negligible gap between pre-trained model and downstream dense prediction tasks. Concretely, these downstream tasks require more accurate representation, in other words, the pixels from the same object must belong to a shared semantic category, which is lacking in the previous methods. In this work, we present Dense Semantic Contrast (DSC) for modeling semantic category decision boundaries at a dense level to meet the requirement of these tasks. Furthermore, we propose a dense cross-image semantic contrastive learning framework for multi-granularity representation learning. Specially, we explicitly explore the semantic structure of the dataset by mining relations among pixels from different perspectives. For intra-image relation modeling, we discover pixel neighbors from multiple views. And for inter-image relations, we enforce pixel representation from the same semantic class to be more similar than the representation from different classes in one mini-batch. Experimental results show that our DSC model outperforms state-of-the-art methods when transferring to downstream dense prediction tasks, including object detection, semantic segmentation, and instance segmentation. Code will be made available.

Youhui Guo, Yu Zhou, Xugong Qin et al.

Scene text detection has drawn the close attention of researchers. Though many methods have been proposed for horizontal and oriented texts, previous methods may not perform well when dealing with arbitrary-shaped texts such as curved texts. In particular, confusion problem arises in the case of nearby text instances. In this paper, we propose a simple yet effective method for accurate arbitrary-shaped nearby scene text detection. Firstly, a One-to-Many Training Scheme (OMTS) is designed to eliminate confusion and enable the proposals to learn more appropriate groundtruths in the case of nearby text instances. Secondly, we propose a Proposal Feature Attention Module (PFAM) to exploit more effective features for each proposal, which can better adapt to arbitrary-shaped text instances. Finally, we propose a baseline that is based on Faster R-CNN and outputs the curve representation directly. Equipped with PFAM and OMTS, the detector can achieve state-of-the-art or competitive performance on several challenging benchmarks.

Xugong Qin, Yu Zhou, Youhui Guo et al.

Due to the large success in object detection and instance segmentation, Mask R-CNN attracts great attention and is widely adopted as a strong baseline for arbitrary-shaped scene text detection and spotting. However, two issues remain to be settled. The first is dense text case, which is easy to be neglected but quite practical. There may exist multiple instances in one proposal, which makes it difficult for the mask head to distinguish different instances and degrades the performance. In this work, we argue that the performance degradation results from the learning confusion issue in the mask head. We propose to use an MLP decoder instead of the "deconv-conv" decoder in the mask head, which alleviates the issue and promotes robustness significantly. And we propose instance-aware mask learning in which the mask head learns to predict the shape of the whole instance rather than classify each pixel to text or non-text. With instance-aware mask learning, the mask branch can learn separated and compact masks. The second is that due to large variations in scale and aspect ratio, RPN needs complicated anchor settings, making it hard to maintain and transfer across different datasets. To settle this issue, we propose an adaptive label assignment in which all instances especially those with extreme aspect ratios are guaranteed to be associated with enough anchors. Equipped with these components, the proposed method named MAYOR achieves state-of-the-art performance on five benchmarks including DAST1500, MSRA-TD500, ICDAR2015, CTW1500, and Total-Text.

Wei Li, Dezhao Luo, Bo Fang et al.

Most of the existing video self-supervised methods mainly leverage temporal signals of videos, ignoring that the semantics of moving objects and environmental information are all critical for video-related tasks. In this paper, we propose a novel self-supervised method for video representation learning, referred to as Video 3D Sampling (V3S). In order to sufficiently utilize the information (spatial and temporal) provided in videos, we pre-process a video from three dimensions (width, height, time). As a result, we can leverage the spatial information (the size of objects), temporal information (the direction and magnitude of motions) as our learning target. In our implementation, we combine the sampling of the three dimensions and propose the scale and projection transformations in space and time respectively. The experimental results show that, when applied to action recognition, video retrieval and action similarity labeling, our approach improves the state-of-the-arts with significant margins.

Xiaoni Li, Yucan Zhou, Yu Zhou et al.

Hierarchical classification is significant for complex tasks by providing multi-granular predictions and encouraging better mistakes. As the label structure decides its performance, many existing approaches attempt to construct an excellent label structure for promoting the classification results. In this paper, we consider that different label structures provide a variety of prior knowledge for category recognition, thus fusing them is helpful to achieve better hierarchical classification results. Furthermore, we propose a multi-task multi-structure fusion model to integrate different label structures. It contains two kinds of branches: one is the traditional classification branch to classify the common subclasses, the other is responsible for identifying the heterogeneous superclasses defined by different label structures. Besides the effect of multiple label structures, we also explore the architecture of the deep model for better hierachical classification and adjust the hierarchical evaluation metrics for multiple label structures. Experimental results on CIFAR100 and Car196 show that our method obtains significantly better results than using a flat classifier or a hierarchical classifier with any single label structure.

Qingyi Si, Zheng Lin, Mingyu Zheng et al.

While sophisticated Visual Question Answering models have achieved remarkable success, they tend to answer questions only according to superficial correlations between question and answer. Several recent approaches have been developed to address this language priors problem. However, most of them predict the correct answer according to one best output without checking the authenticity of answers. Besides, they only explore the interaction between image and question, ignoring the semantics of candidate answers. In this paper, we propose a select-and-rerank (SAR) progressive framework based on Visual Entailment. Specifically, we first select the candidate answers relevant to the question or the image, then we rerank the candidate answers by a visual entailment task, which verifies whether the image semantically entails the synthetic statement of the question and each candidate answer. Experimental results show the effectiveness of our proposed framework, which establishes a new state-of-the-art accuracy on VQA-CP v2 with a 7.55% improvement.

Yifei Zhang, Yu Zhou, Weiping Wang

Despite the great progress achieved in unsupervised feature embedding, existing contrastive learning methods typically pursue view-invariant representations through attracting positive sample pairs and repelling negative sample pairs in the embedding space, while neglecting to systematically explore instance relations. In this paper, we explore instance relations including intra-instance multi-view relation and inter-instance interpolation relation for unsupervised feature embedding. Specifically, we embed intra-instance multi-view relation by aligning the distribution of the distance between an instance's different augmented samples and negative samples. We explore inter-instance interpolation relation by transferring the ratio of information for image sample interpolation from pixel space to feature embedding space. The proposed approach, referred to as EIR, is simple-yet-effective and can be easily inserted into existing view-invariant contrastive learning based methods. Experiments conducted on public benchmarks for image classification and retrieval report state-of-the-art or comparable performance.

Yilin Kang, Yong Liu, Jian Li et al.

Pairwise learning focuses on learning tasks with pairwise loss functions, depends on pairs of training instances, and naturally fits for modeling relationships between pairs of samples. In this paper, we focus on the privacy of pairwise learning and propose a new differential privacy paradigm for pairwise learning, based on gradient perturbation. Except for the privacy guarantees, we also analyze the excess population risk and give corresponding bounds under both expectation and high probability conditions. We use the \textit{on-average stability} and the \textit{pairwise locally elastic stability} theories to analyze the expectation bound and the high probability bound, respectively. Moreover, our analyzed utility bounds do not require convex pairwise loss functions, which means that our method is general to both convex and non-convex conditions. Under these circumstances, the utility bounds are similar to (or better than) previous bounds under convexity or strongly convexity assumption, which are attractive results.

Shu Zhao, Dayan Wu, Yucan Zhou et al.

Deep hashing methods have shown great retrieval accuracy and efficiency in large-scale image retrieval. How to optimize discrete hash bits is always the focus in deep hashing methods. A common strategy in these methods is to adopt an activation function, e.g. $\operatorname{sigmoid}(\cdot)$ or $\operatorname{tanh}(\cdot)$, and minimize a quantization loss to approximate discrete values. However, this paradigm may make more and more hash bits stuck into the wrong saturated area of the activation functions and never escaped. We call this problem "Dead Bits Problem~(DBP)". Besides, the existing quantization loss will aggravate DBP as well. In this paper, we propose a simple but effective gradient amplifier which acts before activation functions to alleviate DBP. Moreover, we devise an error-aware quantization loss to further alleviate DBP. It avoids the negative effect of quantization loss based on the similarity between two images. The proposed gradient amplifier and error-aware quantization loss are compatible with a variety of deep hashing methods. Experimental results on three datasets demonstrate the efficiency of the proposed gradient amplifier and the error-aware quantization loss.

Wei Chen, Yu Liu, Weiping Wang et al.

In recent years a vast amount of visual content has been generated and shared from many fields, such as social media platforms, medical imaging, and robotics. This abundance of content creation and sharing has introduced new challenges, particularly that of searching databases for similar content-Content Based Image Retrieval (CBIR)-a long-established research area in which improved efficiency and accuracy are needed for real-time retrieval. Artificial intelligence has made progress in CBIR and has significantly facilitated the process of instance search. In this survey we review recent instance retrieval works that are developed based on deep learning algorithms and techniques, with the survey organized by deep network architecture types, deep features, feature embedding and aggregation methods, and network fine-tuning strategies. Our survey considers a wide variety of recent methods, whereby we identify milestone work, reveal connections among various methods and present the commonly used benchmarks, evaluation results, common challenges, and propose promising future directions.

Qingyi Si, Yuanxin Liu, Peng Fu et al.

Zero-shot intent detection (ZSID) aims to deal with the continuously emerging intents without annotated training data. However, existing ZSID systems suffer from two limitations: 1) They are not good at modeling the relationship between seen and unseen intents. 2) They cannot effectively recognize unseen intents under the generalized intent detection (GZSID) setting. A critical problem behind these limitations is that the representations of unseen intents cannot be learned in the training stage. To address this problem, we propose a novel framework that utilizes unseen class labels to learn Class-Transductive Intent Representations (CTIR). Specifically, we allow the model to predict unseen intents during training, with the corresponding label names serving as input utterances. On this basis, we introduce a multi-task learning objective, which encourages the model to learn the distinctions among intents, and a similarity scorer, which estimates the connections among intents more accurately. CTIR is easy to implement and can be integrated with existing methods. Experiments on two real-world datasets show that CTIR brings considerable improvement to the baseline systems.

Wei Chen, Weiping Wang, Li Liu et al.

The focus of this survey is on the analysis of two modalities of multimodal deep learning: image and text. Unlike classic reviews of deep learning where monomodal image classifiers such as VGG, ResNet and Inception module are central topics, this paper will examine recent multimodal deep models and structures, including auto-encoders, generative adversarial nets and their variants. These models go beyond the simple image classifiers in which they can do uni-directional (e.g. image captioning, image generation) and bi-directional (e.g. cross-modal retrieval, visual question answering) multimodal tasks. Besides, we analyze two aspects of the challenge in terms of better content understanding in deep multimodal applications. We then introduce current ideas and trends in deep multimodal feature learning, such as feature embedding approaches and objective function design, which are crucial in overcoming the aforementioned challenges. Finally, we include several promising directions for future research.

Dezhao Luo, Bo Fang, Yu Zhou et al.

Existing video self-supervised learning methods mainly rely on trimmed videos for model training. However, trimmed datasets are manually annotated from untrimmed videos. In this sense, these methods are not really self-supervised. In this paper, we propose a novel self-supervised method, referred to as Exploring Relations in Untrimmed Videos (ERUV), which can be straightforwardly applied to untrimmed videos (real unlabeled) to learn spatio-temporal features. ERUV first generates single-shot videos by shot change detection. Then a designed sampling strategy is used to model relations for video clips. The strategy is saved as our self-supervision signals. Finally, the network learns representations by predicting the category of relations between the video clips. ERUV is able to compare the differences and similarities of videos, which is also an essential procedure for action and video related tasks. We validate our learned models with action recognition and video retrieval tasks with three kinds of 3D CNNs. Experimental results show that ERUV is able to learn richer representations and it outperforms state-of-the-art self-supervised methods with significant margins.

Yucan Zhou, Yu Wang, Jianfei Cai et al.

Deep neural networks are highly effective when a large number of labeled samples are available but fail with few-shot classification tasks. Recently, meta-learning methods have received much attention, which train a meta-learner on massive additional tasks to gain the knowledge to instruct the few-shot classification. Usually, the training tasks are randomly sampled and performed indiscriminately, often making the meta-learner stuck into a bad local optimum. Some works in the optimization of deep neural networks have shown that a better arrangement of training data can make the classifier converge faster and perform better. Inspired by this idea, we propose an easy-to-hard expert meta-training strategy to arrange the training tasks properly, where easy tasks are preferred in the first phase, then, hard tasks are emphasized in the second phase. A task hardness aware module is designed and integrated into the training procedure to estimate the hardness of a task based on the distinguishability of its categories. In addition, we explore multiple hardness measurements including the semantic relation, the pairwise Euclidean distance, the Hausdorff distance, and the Hilbert-Schmidt independence criterion. Experimental results on the miniImageNet and tieredImageNetSketch datasets show that the meta-learners can obtain better results with our expert training strategy.

Xugong Qin, Yu Zhou, Dayan Wu et al.

Recent scene text detection works mainly focus on curve text detection. However, in real applications, the curve texts are more scarce than the multi-oriented ones. Accurate detection of multi-oriented text with large variations of scales, orientations, and aspect ratios is of great significance. Among the multi-oriented detection methods, direct regression for the geometry of scene text shares a simple yet powerful pipeline and gets popular in academic and industrial communities, but it may produce imperfect detections, especially for long texts due to the limitation of the receptive field. In this work, we aim to improve this while keeping the pipeline simple. A fully convolutional corner refinement network (FC2RN) is proposed for accurate multi-oriented text detection, in which an initial corner prediction and a refined corner prediction are obtained at one pass. With a novel quadrilateral RoI convolution operation tailed for multi-oriented scene text, the initial quadrilateral prediction is encoded into the feature maps which can be further used to predict offset between the initial prediction and the ground-truth as well as output a refined confidence score. Experimental results on four public datasets including MSRA-TD500, ICDAR2017-RCTW, ICDAR2015, and COCO-Text demonstrate that FC2RN can outperform the state-of-the-art methods. The ablation study shows the effectiveness of corner refinement and scoring for accurate text localization.

Yudi Chen, Wei Wang, Yu Zhou et al.

Though deep learning based scene text detection has achieved great progress, well-trained detectors suffer from severe performance degradation for different domains. In general, a tremendous amount of data is indispensable to train the detector in the target domain. However, data collection and annotation are expensive and time-consuming. To address this problem, we propose a self-training framework to automatically mine hard examples with pseudo-labels from unannotated videos or images. To reduce the noise of hard examples, a novel text mining module is implemented based on the fusion of detection and tracking results. Then, an image-to-video generation method is designed for the tasks that videos are unavailable and only images can be used. Experimental results on standard benchmarks, including ICDAR2015, MSRA-TD500, ICDAR2017 MLT, demonstrate the effectiveness of our self-training method. The simple Mask R-CNN adapted with self-training and fine-tuned on real data can achieve comparable or even superior results with the state-of-the-art methods.

Zhi Qiao, Yu Zhou, Dongbao Yang et al.

Scene text recognition is a hot research topic in computer vision. Recently, many recognition methods based on the encoder-decoder framework have been proposed, and they can handle scene texts of perspective distortion and curve shape. Nevertheless, they still face lots of challenges like image blur, uneven illumination, and incomplete characters. We argue that most encoder-decoder methods are based on local visual features without explicit global semantic information. In this work, we propose a semantics enhanced encoder-decoder framework to robustly recognize low-quality scene texts. The semantic information is used both in the encoder module for supervision and in the decoder module for initializing. In particular, the state-of-the art ASTER method is integrated into the proposed framework as an exemplar. Extensive experiments demonstrate that the proposed framework is more robust for low-quality text images, and achieves state-of-the-art results on several benchmark datasets.

Rui Liu, Zheng Lin, Weiping Wang

Recently, generative methods have been widely used in keyphrase prediction, thanks to their capability to produce both present keyphrases that appear in the source text and absent keyphrases that do not match any source text. However, the absent keyphrases are generated at the cost of the performance on present keyphrase prediction, since previous works mainly use generative models that rely on the copying mechanism and select words step by step. Besides, the extractive model that directly extracts a text span is more suitable for predicting the present keyphrase. Considering the different characteristics of extractive and generative methods, we propose to divide the keyphrase prediction into two subtasks, i.e., present keyphrase extraction (PKE) and absent keyphrase generation (AKG), to fully exploit their respective advantages. On this basis, a joint inference framework is proposed to make the most of BERT in two subtasks. For PKE, we tackle this task as a sequence labeling problem with the pre-trained language model BERT. For AKG, we introduce a Transformer-based architecture, which fully integrates the present keyphrase knowledge learned from PKE by the fine-tuned BERT. The experimental results show that our approach can achieve state-of-the-art results on both tasks on benchmark datasets.

Yilin Kang, Yong Liu, Ben Niu et al.

Traditionally, there are two models on differential privacy: the central model and the local model. The central model focuses on the machine learning model and the local model focuses on the training data. In this paper, we study the \textit{input perturbation} method in differentially private empirical risk minimization (DP-ERM), preserving privacy of the central model. By adding noise to the original training data and training with the `perturbed data', we achieve ($ε$,$δ$)-differential privacy on the final model, along with some kind of privacy on the original data. We observe that there is an interesting connection between the local model and the central model: the perturbation on the original data causes the perturbation on the gradient, and finally the model parameters. This observation means that our method builds a bridge between local and central model, protecting the data, the gradient and the model simultaneously, which is more superior than previous central methods. Detailed theoretical analysis and experiments show that our method achieves almost the same (or even better) performance as some of the best previous central methods with more protections on privacy, which is an attractive result. Moreover, we extend our method to a more general case: the loss function satisfies the Polyak-Lojasiewicz condition, which is more general than strong convexity, the constraint on the loss function in most previous work.

Dezhao Luo, Chang Liu, Yu Zhou et al.

We propose a novel self-supervised method, referred to as Video Cloze Procedure (VCP), to learn rich spatial-temporal representations. VCP first generates "blanks" by withholding video clips and then creates "options" by applying spatio-temporal operations on the withheld clips. Finally, it fills the blanks with "options" and learns representations by predicting the categories of operations applied on the clips. VCP can act as either a proxy task or a target task in self-supervised learning. As a proxy task, it converts rich self-supervised representations into video clip operations (options), which enhances the flexibility and reduces the complexity of representation learning. As a target task, it can assess learned representation models in a uniform and interpretable manner. With VCP, we train spatial-temporal representation models (3D-CNNs) and apply such models on action recognition and video retrieval tasks. Experiments on commonly used benchmarks show that the trained models outperform the state-of-the-art self-supervised models with significant margins.

Yilin Kang, Yong Liu, Weiping Wang

Distributed machine learning is an approach allowing different parties to learn a model over all data sets without disclosing their own data. In this paper, we propose a weighted distributed differential privacy (WD-DP) empirical risk minimization (ERM) method to train a model in distributed setting, considering different weights of different clients. We guarantee differential privacy by gradient perturbation, adding Gaussian noise, and advance the state-of-the-art on gradient perturbation method in distributed setting. By detailed theoretical analysis, we show that in distributed setting, the noise bound and the excess empirical risk bound can be improved by considering different weights held by multiple parties. Moreover, considering that the constraint of convex loss function in ERM is not easy to achieve in some situations, we generalize our method to non-convex loss functions which satisfy Polyak-Lojasiewicz condition. Experiments on real data sets show that our method is more reliable and we improve the performance of distributed differential privacy ERM, especially in the case that data scale on different clients is uneven.

Yong Liu, Jian Li, Weiping Wang

We study the risk performance of distributed learning for the regularization empirical risk minimization with fast convergence rate, substantially improving the error analysis of the existing divide-and-conquer based distributed learning. An interesting theoretical finding is that the larger the diversity of each local estimate is, the tighter the risk bound is. This theoretical analysis motivates us to devise an effective maxdiversity distributed learning algorithm (MDD). Experimental results show that MDD can outperform the existing divide-andconquer methods but with a bit more time. Theoretical analysis and empirical results demonstrate that our proposed MDD is sound and effective.