Deng Cai

AAAI Conference 2026 Conference Paper

Enhancing Spatial Reasoning Through Visual and Textual Thinking

• Xun Liang
• Xin Guo
• Zhongming Jin
• Weihang Pan
• Penghui Shang
• Deng Cai
• Binbin Lin
• Jieping Ye

The spatial reasoning task aims to reason about the spatial relationships in 2D and 3D space, which is a fundamental capability for Visual Question Answering (VQA) and robotics. Although vision language models (VLMs) have developed rapidly in recent years, they are still struggling with the spatial reasoning task. In this paper, we introduce a method that can enhance Spatial reasoning through Visual and Textual thinking Simultaneously (SpatialVTS). In the spatial visual thinking phase, our model is trained to generate location-related specific tokens of important targets automatically. Not only are the objects mentioned in the problem addressed, but also the potential objects related to the reasoning are considered. During the spatial textual thinking phase, our model conducts long-term thinking based on visual cues and dialogues and gradually inferences the answers to spatial reasoning problems. To effectively support the model's training, we made manual corrections to the existing spatial reasoning dataset, eliminating numerous incorrect labels resulting from automatic annotation, restructuring the data input format to enhance generalization, and developing a reasoning framework for model thinking. Without introducing any additional information (such as masks or depth), our model's overall average level in several spatial understanding tasks has significantly improved compared with other models.

TMLR Journal 2025 Journal Article

A Survey on the Honesty of Large Language Models

• Siheng Li
• Cheng Yang
• Taiqiang Wu
• Chufan Shi
• Yuji Zhang
• Xinyu Zhu
• Zesen Cheng
• Deng Cai

Honesty is a fundamental principle for aligning large language models (LLMs) with human values, requiring these models to recognize what they know and don't know and be able to faithfully express their knowledge. Despite promising, current LLMs still exhibit significant dishonest behaviors, such as confidently presenting wrong answers or failing to express what they know. In addition, research on the honesty of LLMs also faces challenges, including varying definitions of honesty, difficulties in distinguishing between known and unknown knowledge, and a lack of comprehensive understanding of related research. To address these issues, we provide a survey on the honesty of LLMs, covering its clarification, evaluation approaches, and strategies for improvement. Moreover, we offer insights for future research, aiming to inspire further exploration in this important area.

NeurIPS Conference 2025 Conference Paper

Controlling Thinking Speed in Reasoning Models

• Zhengkai Lin
• Zhihang Fu
• Ze Chen
• Chao Chen
• Liang Xie
• Wenxiao Wang
• Deng Cai
• Zheng Wang

Human cognition is theorized to operate in two modes: fast, intuitive System 1 thinking and slow, deliberate System 2 thinking. While current Large Reasoning Models (LRMs) excel at System 2 thinking, their inability to perform fast thinking leads to high computational overhead and latency. In this work, we enable LRMs to approximate human intelligence through dynamic thinking speed adjustment, optimizing accuracy-efficiency trade-offs. Our approach addresses two key questions: (1) how to control thinking speed in LRMs, and (2) when to adjust it for optimal performance. For the first question, we identify the steering vector that governs slow-fast thinking transitions in LRMs' representation space. Using this vector, we achieve the first representation editing-based test-time scaling effect, outperforming existing prompt-based scaling methods. For the second question, we apply real-time difficulty estimation to signal reasoning segments of varying complexity. Combining these techniques, we propose the first reasoning strategy that enables fast processing of easy steps and deeper analysis for complex reasoning. Without any training or additional cost, our plug-and-play method yields an average +1. 3\% accuracy with -8. 6\% token usage across leading LRMs and advanced reasoning benchmarks. All of our algorithms are implemented based on vLLM and are expected to support broader applications and inspire future research.

AAAI Conference 2025 Conference Paper

Empowering Self-Learning of LLMs: Inner Knowledge Explicitation as a Catalyst

• Shijue Huang
• Wanjun Zhong
• Deng Cai
• Fanqi Wan
• Chengyi Wang
• Mingxuan Wang
• Mu Qiao
• Ruifeng Xu

Self-learning of Large Language Models (LLMs) facilitates their advancement towards super-intelligence by training with self-synthesized experiences. However, a critical challenge is the amplification of hallucinations in generated data during iterative self-learning, underscoring the need for reliable data selection. To address this, we investigate the mechanism of Inner Knowledge Explicitation, which involves explicitly extracting the inner knowledge from memory of LLMs, to concurrently improves reasoning, and enables reliable self-learning data selection. This paper introduces a Self Knowledge Explicitation Learning (SKE-Learn) framework, which equips the LLMs with meta-skills to explicitly extract, verify and utilize inner knowledge for reasoning. By leveraging these meta-skills, SKE-Learn establishes a self-learning approach that ensures reliable selection of self-synthetic data. This approach enhances performance through iterative self-learning while mitigating the problem of hallucinations. Empirical results from six benchmarks demonstrate that Inner Knowledge Explicitation improves reasoning by serving as a more effective prompting method. Additionally, SKE-Learn, based on the verifiability of explicit knowledge, shows consistent performance improvements over multiple self-training iterations, with an average performance increase from 52.79% to 56.54% across all benchmarks. Furthermore, Inner Knowledge Explicitation provides explanation and intervention space during LLM's generation process.

NeurIPS Conference 2025 Conference Paper

GeoCAD: Local Geometry-Controllable CAD Generation with Large Language Models

• Zhanwei Zhang
• Kaiyuan Liu
• Junjie Liu
• Wenxiao Wang
• Binbin Lin
• Liang Xie
• Chen Shen
• Deng Cai

Local geometry-controllable computer-aided design (CAD) generation aims to modify local parts of CAD models automatically, enhancing design efficiency. It also ensures that the shapes of newly generated local parts follow user-specific geometric instructions (e. g. , an isosceles right triangle or a rectangle with one corner cut off). However, existing methods encounter challenges in achieving this goal. Specifically, they either lack the ability to follow textual instructions or are unable to focus on the local parts. To address this limitation, we introduce GeoCAD, a user-friendly and local geometry-controllable CAD generation method. Specifically, we first propose a complementary captioning strategy to generate geometric instructions for local parts. This strategy involves vertex-based and VLLM-based captioning for systematically annotating simple and complex parts, respectively. In this way, we caption $\sim$221k different local parts in total. In the training stage, given a CAD model, we randomly mask a local part. Then, using its geometric instruction and the remaining parts as input, we prompt large language models (LLMs) to predict the masked part. During inference, users can specify any local part for modification while adhering to a variety of predefined geometric instructions. Extensive experiments demonstrate the effectiveness of GeoCAD in generation quality, validity and text-to-CAD consistency.

AAAI Conference 2025 Conference Paper

STraj: Self-training for Bridging the Cross-Geography Gap in Trajectory Prediction

• Zhanwei Zhang
• Minghao Chen
• Zhihong Gu
• Xinkui Zhao
• Zheng Yang
• Binbin Lin
• Deng Cai
• Wenxiao Wang

Accurate trajectory prediction has prominent significance in autonomous driving scenarios. Most existing methods predict the trajectory of an agent by learning its interaction with other agents and the map within the scenario. However, the heterogeneous distribution of these elements across different geographical scenarios is always ignored. Thus, trajectory predictors might struggle to generalize well when deployed in different geographical scenarios. To bridge the cross-geography gap, in this paper, we propose a plug-and-play self-training pipeline, termed STraj, for cross-geography trajectory prediction. STraj comprises three progressive steps: pseudo label (i.e., time-series trajectory) generation, update, and utilization. First, to generate pseudo labels that generalize to the cross-geography scenarios, STraj pre-trains the predictor through the complementary agent and map augmentations. Second, to facilitate the stable training of the predictor, we design a specific pseudo label update strategy. This strategy selects high-consistency pseudo trajectories from the current and historical epochs to supervise the target domain samples. Third, with generated pseudo trajectories, we introduce trajectory-induced contrastive learning to mitigate the representation bias of cross-geography agents. Extensive experiment results on various cross-geography trajectory prediction benchmarks demonstrate the effectiveness of STraj.

NeurIPS Conference 2025 Conference Paper

TokenSqueeze: Performance-Preserving Compression for Reasoning LLMs

• Yuxiang Zhang
• Zhengxu Yu
• Weihang Pan
• Zhongming Jin
• Qiang Fu
• Deng Cai
• Binbin Lin
• Jieping Ye

Emerging reasoning LLMs such as OpenAI-o1 and DeepSeek-R1 have achieved strong performance on complex reasoning tasks by generating long chain-of-thought (CoT) traces. However, these long CoTs result in increased token usage, leading to higher inference latency and memory consumption. As a result, balancing accuracy and reasoning efficiency has become essential for deploying reasoning LLMs in practical applications. Existing long-to-short (Long2Short) methods aim to reduce inference length but often sacrifice accuracy, revealing a need for an approach that maintains performance while lowering token costs. To address this efficiency-accuracy tradeoff, we propose TokenSqueeze, a novel Long2Short method that condenses reasoning paths while preserving performance and relying exclusively on self-generated data. First, to prevent performance degradation caused by excessive compression of reasoning depth, we propose to select self-generated samples whose reasoning depth is adaptively matched to the complexity of the problem. To further optimize the linguistic expression without altering the underlying reasoning paths, we introduce a distribution-aligned linguistic refinement method that enhances the clarity and conciseness of the reasoning path while preserving its logical integrity. Comprehensive experimental results demonstrated the effectiveness of TokenSqueeze in reducing token usage while maintaining accuracy. Notably, DeepSeek‑R1‑Distill‑Qwen‑7B fine-tuned by using our proposed method achieved a 50\% average token reduction while preserving accuracy on the MATH500 benchmark. TokenSqueeze exclusively utilizes the model's self-generated data, enabling efficient and high-fidelity reasoning without relying on manually curated short-answer datasets across diverse applications. Our code is available at \url{https: //github. com/zhangyx1122/TokenSqueeze}.

NeurIPS Conference 2024 Conference Paper

Delving into the Reversal Curse: How Far Can Large Language Models Generalize?

• Zhengkai Lin
• Zhihang Fu
• Kai Liu
• Liang Xie
• Binbin Lin
• Wenxiao Wang
• Deng Cai
• Yue Wu

While large language models (LLMs) showcase unprecedented capabilities, they also exhibit certain inherent limitations when facing seemingly trivial tasks. A prime example is the recently debated "reversal curse", which surfaces when models, having been trained on the fact "A is B", struggle to generalize this knowledge to infer that "B is A". In this paper, we examine the manifestation of the reversal curse across various tasks and delve into both the generalization abilities and the problem-solving mechanisms of LLMs. This investigation leads to a series of significant insights: (1) LLMs are able to generalize to "B is A" when both A and B are presented in the context as in the case of a multiple-choice question. (2) This generalization ability is highly correlated to the structure of the fact "A is B" in the training documents. For example, this generalization only applies to biographies structured in "[Name] is [Description]" but not to "[Description] is [Name]". (3) We propose and verify the hypothesis that LLMs possess an inherent bias in fact recalling during knowledge application, which explains and underscores the importance of the document structure to successful learning. (4) The negative impact of this bias on the downstream performance of LLMs can hardly be mitigated through training alone. Based on these intriguing findings, our work not only presents a novel perspective for interpreting LLMs' generalization abilities from their intrinsic working mechanism but also provides new insights for the development of more effective learning methods for LLMs.

IJCAI Conference 2024 Conference Paper

G2LTraj: A Global-to-Local Generation Approach for Trajectory Prediction

• Zhanwei Zhang
• Zishuo Hua
• Minghao Chen
• Wei Lu
• Binbin Lin
• Deng Cai
• Wenxiao Wang

Predicting future trajectories of traffic agents accurately holds substantial importance in various applications such as autonomous driving. Previous methods commonly infer all future steps of an agent either recursively or simultaneously. However, the recursive strategy suffers from the accumulated error, while the simultaneous strategy overlooks the constraints among future steps, resulting in kinematically infeasible predictions. To address these issues, in this paper, we propose G2LTraj, a plug-and-play global-to-local generation approach for trajectory prediction. Specifically, we generate a series of global key steps that uniformly cover the entire future time range. Subsequently, the local intermediate steps between the adjacent key steps are recursively filled in. In this way, we prevent the accumulated error from propagating beyond the adjacent key steps. Moreover, to boost the kinematical feasibility, we not only introduce the spatial constraints among key steps but also strengthen the temporal constraints among the intermediate steps. Finally, to ensure the optimal granularity of key steps, we design a selectable granularity strategy that caters to each predicted trajectory. Our G2LTraj significantly improves the performance of seven existing trajectory predictors across the ETH, UCY and nuScenes datasets. Experimental results demonstrate its effectiveness. Code will be available at https: //github. com/Zhanwei-Z/G2LTraj.

NeurIPS Conference 2024 Conference Paper

GLBench: A Comprehensive Benchmark for Graph with Large Language Models

• Yuhan Li
• Peisong Wang
• Xiao Zhu
• Aochuan Chen
• Haiyun Jiang
• Deng Cai
• Victor W. Chan
• Jia Li

The emergence of large language models (LLMs) has revolutionized the way we interact with graphs, leading to a new paradigm called GraphLLM. Despite the rapid development of GraphLLM methods in recent years, the progress and understanding of this field remain unclear due to the lack of a benchmark with consistent experimental protocols. To bridge this gap, we introduce GLBench, the first comprehensive benchmark for evaluating GraphLLM methods in both supervised and zero-shot scenarios. GLBench provides a fair and thorough evaluation of different categories of GraphLLM methods, along with traditional baselines such as graph neural networks. Through extensive experiments on a collection of real-world datasets with consistent data processing and splitting strategies, we have uncovered several key findings. Firstly, GraphLLM methods outperform traditional baselines in supervised settings, with LLM-as-enhancers showing the most robust performance. However, using LLMs as predictors is less effective and often leads to uncontrollable output issues. We also notice that no clear scaling laws exist for current GraphLLM methods. In addition, both structures and semantics are crucial for effective zero-shot transfer, and our proposed simple baseline can even outperform several models tailored for zero-shot scenarios. The data and code of the benchmark can be found at https: //github. com/NineAbyss/GLBench.

NeurIPS Conference 2024 Conference Paper

On the Worst Prompt Performance of Large Language Models

• Bowen Cao
• Deng Cai
• Zhisong Zhang
• Yuexian Zou
• Wai Lam

The performance of large language models (LLMs) is acutely sensitive to the phrasing of prompts, which raises significant concerns about their reliability in real-world scenarios. Existing studies often divide prompts into task-level instructions and case-level inputs and primarily focus on evaluating and improving robustness against variations in tasks-level instructions. However, this setup fails to fully address the diversity of real-world user queries and assumes the existence of task-specific datasets. To address these limitations, we introduce RobustAlpacaEval, a new benchmark that consists of semantically equivalent case-level queries and emphasizes the importance of using the worst prompt performance to gauge the lower bound of model performance. Extensive experiments on RobustAlpacaEval with ChatGPT and six open-source LLMs from the Llama, Mistral, and Gemma families uncover substantial variability in model performance; for instance, a difference of 45. 48% between the worst and best performance for the Llama-2-70B-chat model, with its worst performance dipping as low as 9. 38%. We further illustrate the difficulty in identifying the worst prompt from both model-agnostic and model-dependent perspectives, emphasizing the absence of a shortcut to characterize the worst prompt. We also attempt to enhance the worst prompt performance using existing prompt engineering and prompt consistency methods, but find that their impact is limited. These findings underscore the need to create more resilient LLMs that can maintain high performance across diverse prompts.

AAAI Conference 2024 Conference Paper

Regulating Intermediate 3D Features for Vision-Centric Autonomous Driving

• Junkai Xu
• Liang Peng
• Haoran Cheng
• Linxuan Xia
• Qi Zhou
• Dan Deng
• Wei Qian
• Wenxiao Wang

Multi-camera perception tasks have gained significant attention in the field of autonomous driving. However, existing frameworks based on Lift-Splat-Shoot (LSS) in the multi-camera setting cannot produce suitable dense 3D features due to the projection nature and uncontrollable densification process. To resolve this problem, we propose to regulate intermediate dense 3D features with the help of volume rendering. Specifically, we employ volume rendering to process the dense 3D features to obtain corresponding 2D features (e.g., depth maps, semantic maps), which are supervised by associated labels in the training. This manner regulates the generation of dense 3D features on the feature level, providing appropriate dense and unified features for multiple perception tasks. Therefore, our approach is termed Vampire, stands for ``Volume rendering As Multi-camera Perception Intermediate feature REgulator''. Experimental results on the Occ3D and nuScenes datasets demonstrate that Vampire facilitates fine-grained and appropriate extraction of dense 3D features, and is competitive with existing SOTA methods across diverse downstream perception tasks like 3D occupancy prediction, LiDAR segmentation and 3D objection detection, while utilizing moderate GPU resources. We provide a video demonstration in the supplementary materials and Codes are available at github.com/cskkxjk/Vampire.

AAAI Conference 2024 Conference Paper

Semi-supervised 3D Object Detection with PatchTeacher and PillarMix

• Xiaopei Wu
• Liang Peng
• Liang Xie
• Yuenan Hou
• Binbin Lin
• Xiaoshui Huang
• Haifeng Liu
• Deng Cai

Semi-supervised learning aims to leverage numerous unlabeled data to improve the model performance. Current semi-supervised 3D object detection methods typically use a teacher to generate pseudo labels for a student, and the quality of the pseudo labels is essential for the final performance. In this paper, we propose PatchTeacher, which focuses on partial scene 3D object detection to provide high-quality pseudo labels for the student. Specifically, we divide a complete scene into a series of patches and feed them to our PatchTeacher sequentially. PatchTeacher leverages the low memory consumption advantage of partial scene detection to process point clouds with a high-resolution voxelization, which can minimize the information loss of quantization and extract more fine-grained features. However, it is non-trivial to train a detector on fractions of the scene. Therefore, we introduce three key techniques, i.e., Patch Normalizer, Quadrant Align, and Fovea Selection, to improve the performance of PatchTeacher. Moreover, we devise PillarMix, a strong data augmentation strategy that mixes truncated pillars from different LiDAR scans to generate diverse training samples and thus help the model learn more general representation. Extensive experiments conducted on Waymo and ONCE datasets verify the effectiveness and superiority of our method and we achieve new state-of-the-art results, surpassing existing methods by a large margin. Codes are available at https://github.com/LittlePey/PTPM.

NeurIPS Conference 2024 Conference Paper

StrategyLLM: Large Language Models as Strategy Generators, Executors, Optimizers, and Evaluators for Problem Solving

• Chang Gao
• Haiyun Jiang
• Deng Cai
• Shuming Shi
• Wai Lam

Most existing prompting methods suffer from the issues of generalizability and consistency, as they often rely on instance-specific solutions that may not be applicable to other instances and lack task-level consistency across the selected few-shot examples. To address these limitations, we propose a comprehensive framework, StrategyLLM, allowing LLMs to perform inductive reasoning, deriving general strategies from specific task instances, and deductive reasoning, applying these general strategies to particular task examples, for constructing generalizable and consistent few-shot prompts. It employs four LLM-based agents: strategy generator, executor, optimizer, and evaluator, working together to generate, evaluate, and select promising strategies for a given task. Experimental results demonstrate that StrategyLLM outperforms the competitive baseline CoT-SC that requires human-annotated solutions on 13 datasets across 4 challenging tasks without human involvement, including math reasoning (34. 2\% $\rightarrow$ 38. 8\%), commonsense reasoning (70. 3\% $\rightarrow$ 72. 5\%), algorithmic reasoning (73. 7\% $\rightarrow$ 85. 0\%), and symbolic reasoning (30. 0\% $\rightarrow$ 79. 2\%). Further analysis reveals that StrategyLLM is applicable to various LLMs and demonstrates advantages across numerous scenarios.

AAAI Conference 2024 Conference Paper

TagCLIP: A Local-to-Global Framework to Enhance Open-Vocabulary Multi-Label Classification of CLIP without Training

• Yuqi Lin
• Minghao Chen
• Kaipeng Zhang
• Hengjia Li
• Mingming Li
• Zheng Yang
• Dongqin Lv
• Binbin Lin

Contrastive Language-Image Pre-training (CLIP) has demonstrated impressive capabilities in open-vocabulary classification. The class token in the image encoder is trained to capture the global features to distinguish different text descriptions supervised by contrastive loss, making it highly effective for single-label classification. However, it shows poor performance on multi-label datasets because the global feature tends to be dominated by the most prominent class and the contrastive nature of softmax operation aggravates it. In this study, we observe that the multi-label classification results heavily rely on discriminative local features but are overlooked by CLIP. As a result, we dissect the preservation of patch-wise spatial information in CLIP and proposed a local-to-global framework to obtain image tags. It comprises three steps: (1) patch-level classification to obtain coarse scores; (2) dual-masking attention refinement (DMAR) module to refine the coarse scores; (3) class-wise reidentification (CWR) module to remedy predictions from a global perspective. This framework is solely based on frozen CLIP and significantly enhances its multi-label classification performance on various benchmarks without dataset-specific training. Besides, to comprehensively assess the quality and practicality of generated tags, we extend their application to the downstream task, i.e., weakly supervised semantic segmentation (WSSS) with generated tags as image-level pseudo labels. Experiments demonstrate that this classify-then-segment paradigm dramatically outperforms other annotation-free segmentation methods and validates the effectiveness of generated tags. Our code is available at https://github.com/linyq2117/TagCLIP.

AAAI Conference 2024 Conference Paper

Towards Fine-Grained HBOE with Rendered Orientation Set and Laplace Smoothing

• Ruisi Zhao
• Mingming Li
• Zheng Yang
• Binbin Lin
• Xiaohui Zhong
• Xiaobo Ren
• Deng Cai
• Boxi Wu

Human body orientation estimation (HBOE) aims to estimate the orientation of a human body relative to the camera’s frontal view. Despite recent advancements in this field, there still exist limitations in achieving fine-grained results. We identify certain defects and propose corresponding approaches as follows: 1). Existing datasets suffer from non-uniform angle distributions, resulting in sparse image data for certain angles. To provide comprehensive and high-quality data, we introduce RMOS (Rendered Model Orientation Set), a rendered dataset comprising 150K accurately labeled human instances with a wide range of orientations. 2). Directly using one-hot vector as labels may overlook the similarity between angle labels, leading to poor supervision. And converting the predictions from radians to degrees enlarges the regression error. To enhance supervision, we employ Laplace smoothing to vectorize the label, which contains more information. For fine-grained predictions, we adopt weighted Smooth-L1-loss to align predictions with the smoothed-label, thus providing robust supervision. 3). Previous works ignore body-part-specific information, resulting in coarse predictions. By employing local-window self-attention, our model could utilize different body part information for more precise orientation estimations. We validate the effectiveness of our method in the benchmarks with extensive experiments and show that our method outperforms state-of-the-art. Project is available at: https://github.com/Whalesong-zrs/Towards-Fine-grained-HBOE.

NeurIPS Conference 2024 Conference Paper

Unchosen Experts Can Contribute Too: Unleashing MoE Models’ Power by Self-Contrast

• Chufan Shi
• Cheng Yang
• Xinyu Zhu
• Jiahao Wang
• Taiqiang Wu
• Siheng Li
• Deng Cai
• Yujiu Yang

Mixture-of-Experts (MoE) has emerged as a prominent architecture for scaling model size while maintaining computational efficiency. In MoE, each token in the input sequence activates a different subset of experts determined by a routing mechanism. However, the unchosen experts in MoE models do not contribute to the output, potentially leading to underutilization of the model's capacity. In this work, we first conduct exploratory studies to demonstrate that increasing the number of activated experts does not necessarily improve and can even degrade the output quality. Then, we show that output distributions from an MoE model using different routing strategies substantially differ, indicating that different experts do not always act synergistically. Motivated by these findings, we propose S elf- C ontrast M ixture- o f- E xperts (SCMoE), a training-free strategy that utilizes unchosen experts in a self-contrast manner during inference. In SCMoE, the next-token probabilities are determined by contrasting the outputs from strong and weak activation using the same MoE model. Our method is conceptually simple and computationally lightweight, as it incurs minimal latency compared to greedy decoding. Experiments on several benchmarks (GSM8K, StrategyQA, MBPP and HumanEval) demonstrate that SCMoE can consistently enhance Mixtral 8x7B’s reasoning capability across various domains. For example, it improves the accuracy on GSM8K from 61. 79 to 66. 94. Moreover, combining SCMoE with self-consistency yields additional gains, increasing major@20 accuracy from 75. 59 to 78. 31.

NeurIPS Conference 2023 Conference Paper

Repetition In Repetition Out: Towards Understanding Neural Text Degeneration from the Data Perspective

• Huayang Li
• Tian Lan
• Zihao Fu
• Deng Cai
• Lemao Liu
• Nigel Collier
• Taro Watanabe
• Yixuan Su

There are a number of diverging hypotheses about the neural text degeneration problem, i. e. , generating repetitive and dull loops, which makes this problem both interesting and confusing. In this work, we aim to advance our understanding by presenting a straightforward and fundamental explanation from the data perspective. Our preliminary investigation reveals a strong correlation between the degeneration issue and the presence of repetitions in training data. Subsequent experiments also demonstrate that by selectively dropping out the attention to repetitive words in training data, degeneration can be significantly minimized. Furthermore, our empirical analysis illustrates that prior works addressing the degeneration issue from various standpoints, such as the high-inflow words, the likelihood objective, and the self-reinforcement phenomenon, can be interpreted by one simple explanation. That is, penalizing the repetitions in training data is a common and fundamental factor for their effectiveness. Moreover, our experiments reveal that penalizing the repetitions in training data remains critical even when considering larger model sizes and instruction tuning.

AAAI Conference 2023 Conference Paper

Towards In-Distribution Compatible Out-of-Distribution Detection

• Boxi Wu
• Jie Jiang
• Haidong Ren
• Zifan Du
• Wenxiao Wang
• Zhifeng Li
• Deng Cai
• Xiaofei He

Deep neural network, despite its remarkable capability of discriminating targeted in-distribution samples, shows poor performance on detecting anomalous out-of-distribution data. To address this defect, state-of-the-art solutions choose to train deep networks on an auxiliary dataset of outliers. Various training criteria for these auxiliary outliers are proposed based on heuristic intuitions. However, we find that these intuitively designed outlier training criteria can hurt in-distribution learning and eventually lead to inferior performance. To this end, we identify three causes of the in-distribution incompatibility: contradictory gradient, false likelihood, and distribution shift. Based on our new understandings, we propose a new out-of-distribution detection method by adapting both the top-design of deep models and the loss function. Our method achieves in-distribution compatibility by pursuing less interference with the probabilistic characteristic of in-distribution features. On several benchmarks, our method not only achieves the state-of-the-art out-of-distribution detection performance but also improves the in-distribution accuracy.

AAAI Conference 2022 Conference Paper

DMN4: Few-Shot Learning via Discriminative Mutual Nearest Neighbor Neural Network

• Yang Liu
• Tu Zheng
• Jie Song
• Deng Cai
• Xiaofei He

Few-shot learning (FSL) aims to classify images under lowdata regimes, where the conventional pooled global feature is likely to lose useful local characteristics. Recent work has achieved promising performances by using deep descriptors. They generally take all deep descriptors from neural networks into consideration while ignoring that some of them are useless in classification due to their limited receptive field, e. g. , task-irrelevant descriptors could be misleading and multiple aggregative descriptors from background clutter could even overwhelm the object’s presence. In this paper, we argue that a Mutual Nearest Neighbor (MNN) relation should be established to explicitly select the query descriptors that are most relevant to each task and discard less relevant ones from aggregative clutters in FSL. Specifically, we propose Discriminative Mutual Nearest Neighbor Neural Network (DMN4) for FSL. Extensive experiments demonstrate that our method outperforms the existing state-of-the-arts on both fine-grained and generalized datasets.

NeurIPS Conference 2022 Conference Paper

Learning to Break the Loop: Analyzing and Mitigating Repetitions for Neural Text Generation

• Jin Xu
• Xiaojiang Liu
• Jianhao Yan
• Deng Cai
• Huayang Li
• Jian Li

While large-scale neural language models, such as GPT2 and BART, have achieved impressive results on various text generation tasks, they tend to get stuck in undesirable sentence-level loops with maximization-based decoding algorithms (\textit{e. g. }, greedy search). This phenomenon is counter-intuitive since there are few consecutive sentence-level repetitions in the human corpus (e. g. , 0. 02\% in Wikitext-103). To investigate the underlying reasons for generating consecutive sentence-level repetitions, we study the relationship between the probability of repetitive tokens and their previous repetitions in context. Through our quantitative experiments, we find that 1) Models have a preference to repeat the previous sentence; 2) The sentence-level repetitions have a \textit{self-reinforcement effect}: the more times a sentence is repeated in the context, the higher the probability of continuing to generate that sentence; 3) The sentences with higher initial probabilities usually have a stronger self-reinforcement effect. Motivated by our findings, we propose a simple and effective training method \textbf{DITTO} (Pseu\underline{D}o-Repet\underline{IT}ion Penaliza\underline{T}i\underline{O}n), where the model learns to penalize probabilities of sentence-level repetitions from synthetic repetitive data. Although our method is motivated by mitigating repetitions, our experiments show that DITTO not only mitigates the repetition issue without sacrificing perplexity, but also achieves better generation quality. Extensive experiments on open-ended text generation (Wikitext-103) and text summarization (CNN/DailyMail) demonstrate the generality and effectiveness of our method.

NeurIPS Conference 2022 Conference Paper

Measuring and Reducing Model Update Regression in Structured Prediction for NLP

• Deng Cai
• Elman Mansimov
• Yi-An Lai
• Yixuan Su
• Lei Shu
• Yi Zhang

Recent advance in deep learning has led to rapid adoption of machine learning based NLP models in a wide range of applications. Despite the continuous gain in accuracy, backward compatibility is also an important aspect for industrial applications, yet it received little research attention. Backward compatibility requires that the new model does not regress on cases that were correctly handled by its predecessor. This work studies model update regression in structured prediction tasks. We choose syntactic dependency parsing and conversational semantic parsing as representative examples of structured prediction tasks in NLP. First, we measure and analyze model update regression in different model update settings. Next, we explore and benchmark existing techniques for reducing model update regression including model ensemble and knowledge distillation. We further propose a simple and effective method, Backward-Congruent Re-ranking (BCR), by taking into account the characteristics of structured output. Experiments show that BCR can better mitigate model update regression than model ensemble and knowledge distillation approaches.

AAAI Conference 2022 Conference Paper

SCALoss: Side and Corner Aligned Loss for Bounding Box Regression

• Tu Zheng
• Shuai Zhao
• Yang Liu
• Zili Liu
• Deng Cai

Bounding box regression is an important component in object detection. Recent work achieves promising performance by optimizing the Intersection over Union (IoU). However, IoU-based loss has the gradient vanish problem in the case of low overlapping bounding boxes, and the model could easily ignore these simple cases. In this paper, we propose Side Overlap (SO) loss by maximizing the side overlap of two bounding boxes, which puts more penalty for low overlapping bounding box cases. Besides, to speed up the convergence, the Corner Distance (CD) is added into the objective function. Combining the Side Overlap and Corner Distance, we get a new regression objective function, Side and Corner Align Loss (SCALoss). The SCALoss is well-correlated with IoU loss, which also benefits the evaluation metric but produces more penalty for low-overlapping cases. It can serve as a comprehensive similarity measure, leading to better localization performance and faster convergence speed. Experiments on COCO, PASCAL VOC, and LVIS benchmarks show that SCALoss can bring consistent improvement and outperform `n loss and IoU based loss with popular object detectors such as YOLOV3, SSD, Faster-RCNN. Code is available at: https: //github. com/Turoad/SCALoss.

NeurIPS Conference 2021 Conference Paper

Do Wider Neural Networks Really Help Adversarial Robustness?

• Boxi Wu
• Jinghui Chen
• Deng Cai
• Xiaofei He
• Quanquan Gu

Adversarial training is a powerful type of defense against adversarial examples. Previous empirical results suggest that adversarial training requires wider networks for better performances. However, it remains elusive how does neural network width affect model robustness. In this paper, we carefully examine the relationship between network width and model robustness. Specifically, we show that the model robustness is closely related to the tradeoff between natural accuracy and perturbation stability, which is controlled by the robust regularization parameter λ. With the same λ, wider networks can achieve better natural accuracy but worse perturbation stability, leading to a potentially worse overall model robustness. To understand the origin of this phenomenon, we further relate the perturbation stability with the network's local Lipschitzness. By leveraging recent results on neural tangent kernels, we theoretically show that wider networks tend to have worse perturbation stability. Our analyses suggest that: 1) the common strategy of first fine-tuning λ on small networks and then directly use it for wide model training could lead to deteriorated model robustness; 2) one needs to properly enlarge λ to unleash the robustness potential of wider models fully. Finally, we propose a new Width Adjusted Regularization (WAR) method that adaptively enlarges λ on wide models and significantly saves the tuning time.

AAAI Conference 2021 Conference Paper

RESA: Recurrent Feature-Shift Aggregator for Lane Detection

• Tu Zheng
• Hao Fang
• Yi Zhang
• Wenjian Tang
• Zheng Yang
• Haifeng Liu
• Deng Cai

Lane detection is one of the most important tasks in selfdriving. Due to various complex scenarios (e. g. , severe occlusion, ambiguous lanes, etc.) and the sparse supervisory signals inherent in lane annotations, lane detection task is still challenging. Thus, it is difficult for the ordinary convolutional neural network (CNN) to train in general scenes to catch subtle lane feature from the raw image. In this paper, we present a novel module named REcurrent Feature-Shift Aggregator (RESA) to enrich lane feature after preliminary feature extraction with an ordinary CNN. RESA takes advantage of strong shape priors of lanes and captures spatial relationships of pixels across rows and columns. It shifts sliced feature map recurrently in vertical and horizontal directions and enables each pixel to gather global information. RESA can conjecture lanes accurately in challenging scenarios with weak appearance clues by aggregating sliced feature map. Moreover, we propose a Bilateral Up-Sampling Decoder that combines coarse-grained and fine-detailed features in the upsampling stage. It can recover the low-resolution feature map into pixel-wise prediction meticulously. Our method achieves state-of-the-art results on two popular lane detection benchmarks (CULane and Tusimple). Code has been made available at: https: //github. com/ZJULearning/resa.

AAAI Conference 2020 Conference Paper

Adversarial-Learned Loss for Domain Adaptation

• Minghao Chen
• Shuai Zhao
• Haifeng Liu
• Deng Cai

Recently, remarkable progress has been made in learning transferable representation across domains. Previous works in domain adaptation are majorly based on two techniques: domain-adversarial learning and self-training. However, domain-adversarial learning only aligns feature distributions between domains but does not consider whether the target features are discriminative. On the other hand, selftraining utilizes the model predictions to enhance the discrimination of target features, but it is unable to explicitly align domain distributions. In order to combine the strengths of these two methods, we propose a novel method called Adversarial-Learned Loss for Domain Adaptation (ALDA). We first analyze the pseudo-label method, a typical selftraining method. Nevertheless, there is a gap between pseudolabels and the ground truth, which can cause incorrect training. Thus we introduce the confusion matrix, which is learned through an adversarial manner in ALDA, to reduce the gap and align the feature distributions. Finally, a new loss function is auto-constructed from the learned confusion matrix, which serves as the loss for unlabeled target samples. Our ALDA outperforms state-of-the-art approaches in four standard domain adaptation datasets. Our code is available at https: //github. com/ZJULearning/ALDA.

AAAI Conference 2020 Conference Paper

Graph Transformer for Graph-to-Sequence Learning

• Deng Cai
• Wai Lam

The dominant graph-to-sequence transduction models employ graph neural networks for graph representation learning, where the structural information is reflected by the receptive field of neurons. Unlike graph neural networks that restrict the information exchange between immediate neighborhood, we propose a new model, known as Graph Transformer, that uses explicit relation encoding and allows direct communication between two distant nodes. It provides a more efficient way for global graph structure modeling. Experiments on the applications of text generation from Abstract Meaning Representation (AMR) and syntax-based neural machine translation show the superiority of our proposed model. Specifically, our model achieves 27. 4 BLEU on LDC2015E86 and 29. 7 BLEU on LDC2017T10 for AMR-to-text generation, outperforming the state-of-the-art results by up to 2. 2 points. On the syntax-based translation tasks, our model establishes new single-model state-of-the-art BLEU scores, 21. 3 for Englishto-German and 14. 1 for English-to-Czech, improving over the existing best results, including ensembles, by over 1 BLEU.

IJCAI Conference 2020 Conference Paper

MaCAR: Urban Traffic Light Control via Active Multi-agent Communication and Action Rectification

• Zhengxu Yu
• Shuxian Liang
• Long Wei
• Zhongming Jin
• Jianqiang Huang
• Deng Cai
• Xiaofei He
• Xian-Sheng Hua

Urban traffic light control is an important and challenging real-world problem. By regarding intersections as agents, most of the Reinforcement Learning (RL) based methods generate actions of agents independently. They can cause action conflict and result in overflow or road resource waste in adjacent intersections. Recently, some collaborative methods have alleviated the above problems by extending the observable surroundings of agents, which can be considered as inactive cross-agent communication methods. However, when agents act synchronously in these works, the perceived action value is biased and the information exchanged is insufficient. In this work, we propose a novel Multi-agent Communication and Action Rectification (MaCAR) framework. It enables active communication between agents by considering the impact of synchronous actions of agents. MaCAR consists of two parts: (1) an active Communication Agent Network (CAN) involving a Message Propagation Graph Neural Network (MPGNN); (2) a Traffic Forecasting Network (TFN) which learns to predict the traffic after agents' synchronous actions and the corresponding action values. By using predicted information, we mitigate the action value bias during training to help rectify agents' future actions. In experiments, we show that our proposal can outperforms state-of-the-art methods on both synthetic and real-world datasets.

AAAI Conference 2020 Conference Paper

PI-RCNN: An Efficient Multi-Sensor 3D Object Detector with Point-Based Attentive Cont-Conv Fusion Module

• Liang Xie
• Chao Xiang
• Zhengxu Yu
• Guodong Xu
• Zheng Yang
• Deng Cai
• Xiaofei He

LIDAR point clouds and RGB-images are both extremely essential for 3D object detection. So many state-of-the-art 3D detection algorithms dedicate in fusing these two types of data effectively. However, their fusion methods based on Bird’s Eye View (BEV) or voxel format are not accurate. In this paper, we propose a novel fusion approach named Pointbased Attentive Cont-conv Fusion(PACF) module, which fuses multi-sensor features directly on 3D points. Except for continuous convolution, we additionally add a Point-Pooling and an Attentive Aggregation to make the fused features more expressive. Moreover, based on the PACF module, we propose a 3D multi-sensor multi-task network called Pointcloud- Image RCNN(PI-RCNN as brief), which handles the image segmentation and 3D object detection tasks. PI-RCNN employs a segmentation sub-network to extract full-resolution semantic feature maps from images and then fuses the multisensor features via powerful PACF module. Beneficial from the effectiveness of the PACF module and the expressive semantic features from the segmentation module, PI-RCNN can improve much in 3D object detection. We demonstrate the effectiveness of the PACF module and PI-RCNN on the KITTI 3D Detection benchmark, and our method can achieve stateof-the-art on the metric of 3D AP.

AAAI Conference 2020 Conference Paper

Question-Driven Purchasing Propensity Analysis for Recommendation

• Long Chen
• Ziyu Guan
• Qibin Xu
• Qiong Zhang
• Huan Sun
• Guangyue Lu
• Deng Cai

Merchants of e-commerce Websites expect recommender systems to entice more consumption which is highly correlated with the customers’ purchasing propensity. However, most existing recommender systems focus on customers’ general preference rather than purchasing propensity often governed by instant demands which we deem to be well conveyed by the questions asked by customers. A typical recommendation scenario is: Bob wants to buy a cell phone which can play the game PUBG. He is interested in HUAWEI P20 and asks “can PUBG run smoothly on this phone? ” under it. Then our system will be triggered to recommend the most eligible cell phones to him. Intuitively, diverse user questions could probably be addressed in reviews written by other users who have similar concerns. To address this recommendation problem, we propose a novel Question-Driven Attentive Neural Network (QDANN) to assess the instant demands of questioners and the eligibility of products based on user generated reviews, and do recommendation accordingly. Without supervision, QDANN can well exploit reviews to achieve this goal. The attention mechanisms can be used to provide explanations for recommendations. We evaluate QDANN in three domains of Taobao. The results show the efficacy of our method and its superiority over baseline methods.

AAAI Conference 2020 Conference Paper

Training-Time-Friendly Network for Real-Time Object Detection

• Zili Liu
• Tu Zheng
• Guodong Xu
• Zheng Yang
• Haifeng Liu
• Deng Cai

Modern object detectors can rarely achieve short training time, fast inference speed, and high accuracy at the same time. To strike a balance among them, we propose the Training-Time-Friendly Network (TTFNet). In this work, we start with light-head, single-stage, and anchor-free designs, which enable fast inference speed. Then, we focus on shortening training time. We notice that encoding more training samples from annotated boxes plays a similar role as increasing batch size, which helps enlarge the learning rate and accelerate the training process. To this end, we introduce a novel approach using Gaussian kernels to encode training samples. Besides, we design the initiative sample weights for better information utilization. Experiments on MS COCO show that our TTFNet has great advantages in balancing training time, inference speed, and accuracy. It has reduced training time by more than seven times compared to previous real-time detectors while maintaining state-of-the-art performances. In addition, our super-fast version of TTFNet-18 and TTFNet-53 can outperform SSD300 and YOLOv3 by less than one-tenth of their training time, respectively. The code has been made available at https: //github. com/ZJULearning/ttfnet.

IJCAI Conference 2019 Conference Paper

COP: Customized Deep Model Compression via Regularized Correlation-Based Filter-Level Pruning

• Wenxiao Wang
• Cong Fu
• Jishun Guo
• Deng Cai
• Xiaofei He

Neural network compression empowers the effective yet unwieldy deep convolutional neural networks (CNN) to be deployed in resource-constrained scenarios. Most state-of-the-art approaches prune the model in filter-level according to the "importance" of filters. Despite their success, we notice they suffer from at least two of the following problems: 1) The redundancy among filters is not considered because the importance is evaluated independently. 2) Cross-layer filter comparison is unachievable since the importance is defined locally within each layer. Consequently, we must manually specify layer-wise pruning ratios. 3) They are prone to generate sub-optimal solutions because they neglect the inequality between reducing parameters and reducing computational cost. Reducing the same number of parameters in different positions in the network may reduce different computational cost. To address the above problems, we develop a novel algorithm named as COP (correlation-based pruning), which can detect the redundant filters efficiently. We enable the cross-layer filter comparison through global normalization. We add parameter-quantity and computational-cost regularization terms to the importance, which enables the users to customize the compression according to their preference (smaller or faster). Extensive experiments have shown COP outperforms the others significantly. The code is released at https: //github. com/ZJULearning/COP.

IJCAI Conference 2019 Conference Paper

Localizing Unseen Activities in Video via Image Query

• Zhu Zhang
• Zhou Zhao
• Zhijie Lin
• Jingkuan Song
• Deng Cai

Action localization in untrimmed videos is an important topic in the field of video understanding. However, existing action localization methods are restricted to a pre-defined set of actions and cannot localize unseen activities. Thus, we consider a new task to localize unseen activities in videos via image queries, named Image-Based Activity Localization. This task faces three inherent challenges: (1) how to eliminate the influence of semantically inessential contents in image queries; (2) how to deal with the fuzzy localization of inaccurate image queries; (3) how to determine the precise boundaries of target segments. We then propose a novel self-attention interaction localizer to retrieve unseen activities in an end-to-end fashion. Specifically, we first devise a region self-attention method with relative position encoding to learn fine-grained image region representations. Then, we employ a local transformer encoder to build multi-step fusion and reasoning of image and video contents. We next adopt an order-sensitive localizer to directly retrieve the target segment. Furthermore, we construct a new dataset ActivityIBAL by reorganizing the ActivityNet dataset. The extensive experiments show the effectiveness of our method.

IJCAI Conference 2019 Conference Paper

Progressive Transfer Learning for Person Re-identification

• Zhengxu Yu
• Zhongming Jin
• Long Wei
• Jishun Guo
• Jianqiang Huang
• Deng Cai
• Xiaofei He
• Xian-Sheng Hua

Model fine-tuning is a widely used transfer learning approach in person Re-identification (ReID) applications, which fine-tuning a pre-trained feature extraction model into the target scenario instead of training a model from scratch. It is challenging due to the significant variations inside the target scenario, e. g. , different camera viewpoint, illumination changes, and occlusion. These variations result in a gap between the distribution of each mini-batch and the distribution of the whole dataset when using mini-batch training. In this paper, we study model fine-tuning from the perspective of the aggregation and utilization of the global information of the dataset when using mini-batch training. Specifically, we introduce a novel network structure called Batch-related Convolutional Cell (BConv-Cell), which progressively collects the global information of the dataset into a latent state and uses this latent state to rectify the extracted feature. Based on BConv-Cells, we further proposed the Progressive Transfer Learning (PTL) method to facilitate the model fine-tuning process by joint training the BConv-Cells and the pre-trained ReID model. Empirical experiments show that our proposal can improve the performance of the ReID model greatly on MSMT17, Market-1501, CUHK03 and DukeMTMC-reID datasets. The code will be released later on at \url{https: //github. com/ZJULearning/PTL}

NeurIPS Conference 2019 Conference Paper

Region Mutual Information Loss for Semantic Segmentation

• Shuai Zhao
• Yang Wang
• Zheng Yang
• Deng Cai

Semantic segmentation is a fundamental problem in computer vision. It is considered as a pixel-wise classification problem in practice, and most segmentation models use a pixel-wise loss as their optimization criterion. However, the pixel-wise loss ignores the dependencies between pixels in an image. Several ways to exploit the relationship between pixels have been investigated, \eg, conditional random fields (CRF) and pixel affinity based methods. Nevertheless, these methods usually require additional model branches, large extra memories, or more inference time. In this paper, we develop a region mutual information (RMI) loss to model the dependencies among pixels more simply and efficiently. In contrast to the pixel-wise loss which treats the pixels as independent samples, RMI uses one pixel and its neighbour pixels to represent this pixel. Then for each pixel in an image, we get a multi-dimensional point that encodes the relationship between pixels, and the image is cast into a multi-dimensional distribution of these high-dimensional points. The prediction and ground truth thus can achieve high order consistency through maximizing the mutual information (MI) between their multi-dimensional distributions. Moreover, as the actual value of the MI is hard to calculate, we derive a lower bound of the MI and maximize the lower bound to maximize the real value of the MI. RMI only requires a few extra computational resources in the training stage, and there is no overhead during testing. Experimental results demonstrate that RMI can achieve substantial and consistent improvements in performance on PASCAL VOC 2012 and CamVid datasets. The code is available at \url{https: //github. com/ZJULearning/RMI}.

JMLR Journal 2019 Journal Article

Scaling Up Sparse Support Vector Machines by Simultaneous Feature and Sample Reduction

• Bin Hong
• Weizhong Zhang
• Wei Liu
• Jieping Ye
• Deng Cai
• Xiaofei He
• Jie Wang

Sparse support vector machine (SVM) is a popular classification technique that can simultaneously learn a small set of the most interpretable features and identify the support vectors. It has achieved great successes in many real-world applications. However, for large-scale problems involving a huge number of samples and ultra-high dimensional features, solving sparse SVMs remains challenging. By noting that sparse SVMs induce sparsities in both feature and sample spaces, we propose a novel approach, which is based on accurate estimations of the primal and dual optima of sparse SVMs, to simultaneously identify the inactive features and samples that are guaranteed to be irrelevant to the outputs. Thus, we can remove the identified inactive samples and features from the training phase, leading to substantial savings in the computational cost without sacrificing the accuracy. Moreover, we show that our method can be extended to multi-class sparse support vector machines. To the best of our knowledge, the proposed method is the first static feature and sample reduction method for sparse SVMs and multi-class sparse SVMs. Experiments on both synthetic and real data sets demonstrate that our approach significantly outperforms state-of-the-art methods and the speedup gained by our approach can be orders of magnitude. [abs] [ pdf ][ bib ] &copy JMLR 2019. ( edit, beta )

AAAI Conference 2019 Conference Paper

Unsupervised Learning Helps Supervised Neural Word Segmentation

• Xiaobin Wang
• Deng Cai
• Linlin Li
• Guangwei Xu
• Hai Zhao
• Luo Si

By exploiting unlabeled data for further performance improvement for Chinese word segmentation, this work makes the first attempt at exploring adding unsupervised segmentation information into neural supervised segmenter. We survey various effective strategies, including extending the character embedding, augmenting the word score and applying multi-task learning, for leveraging unsupervised information derived from abundant unlabeled data. Experiments on standard data sets show that the explored strategies indeed improve the recall rate of out-of-vocabulary words and thus boost the segmentation accuracy. Moreover, the model enhanced by the proposed methods outperforms state-of-theart models in closed test and shows promising improvement trend when adopting three different strategies with the help of a large unlabeled data set. Our thorough empirical study eventually verifies the proposed approach outperforms the widelyused pre-training approach in terms of effectively making use of freely abundant unlabeled data.

IJCAI Conference 2018 Conference Paper

A Brand-level Ranking System with the Customized Attention-GRU Model

• Yu Zhu
• Junxiong Zhu
• Jie Hou
• Yongliang Li
• Beidou Wang
• Ziyu Guan
• Deng Cai

In e-commerce websites like Taobao, brand is playing a more important role in influencing users' decision of click/purchase, partly because users are now attaching more importance to the quality of products and brand is an indicator of quality. However, existing ranking systems are not specifically designed to satisfy this kind of demand. Some design tricks may partially alleviate this problem, but still cannot provide satisfactory results or may create additional interaction cost. In this paper, we design the first brand-level ranking system to address this problem. The key challenge of this system is how to sufficiently exploit users' rich behavior in e-commerce websites to rank the brands. In our solution, we firstly conduct the feature engineering specifically tailored for the personalized brand ranking problem and then rank the brands by an adapted Attention-GRU model containing three important modifications. Note that our proposed modifications can also apply to many other machine learning models on various tasks. We conduct a series of experiments to evaluate the effectiveness of our proposed ranking model and test the response to the brand-level ranking system from real users on a large-scale e-commerce platform, i. e. Taobao.

IJCAI Conference 2018 Conference Paper

Attentional Image Retweet Modeling via Multi-Faceted Ranking Network Learning

• Zhou Zhao
• Lingtao Meng
• Jun Xiao
• Min Yang
• Fei Wu
• Deng Cai
• Xiaofei He
• Yueting Zhuang

Retweet prediction is a challenging problem in social media sites (SMS). In this paper, we study the problem of image retweet prediction in social media, which predicts the image sharing behavior that the user reposts the image tweets from their followees. Unlike previous studies, we learn user preference ranking model from their past retweeted image tweets in SMS. We first propose heterogeneous image retweet modeling network (IRM) that exploits users' past retweeted image tweets with associated contexts, their following relations in SMS and preference of their followees. We then develop a novel attentional multi-faceted ranking network learning framework with multi-modal neural networks for the proposed heterogenous IRM network to learn the joint image tweet representations and user preference representations for prediction task. The extensive experiments on a large-scale dataset from Twitter site shows that our method achieves better performance than other state-of-the-art solutions to the problem.

NeurIPS Conference 2018 Conference Paper

MacNet: Transferring Knowledge from Machine Comprehension to Sequence-to-Sequence Models

• Boyuan Pan
• Yazheng Yang
• Hao Li
• Zhou Zhao
• Yueting Zhuang
• Deng Cai
• Xiaofei He

Machine Comprehension (MC) is one of the core problems in natural language processing, requiring both understanding of the natural language and knowledge about the world. Rapid progress has been made since the release of several benchmark datasets, and recently the state-of-the-art models even surpass human performance on the well-known SQuAD evaluation. In this paper, we transfer knowledge learned from machine comprehension to the sequence-to-sequence tasks to deepen the understanding of the text. We propose MacNet: a novel encoder-decoder supplementary architecture to the widely used attention-based sequence-to-sequence models. Experiments on neural machine translation (NMT) and abstractive text summarization show that our proposed framework can significantly improve the performance of the baseline models, and our method for the abstractive text summarization achieves the state-of-the-art results on the Gigaword dataset.

IJCAI Conference 2018 Conference Paper

Multi-Turn Video Question Answering via Multi-Stream Hierarchical Attention Context Network

• Zhou Zhao
• Xinghua Jiang
• Deng Cai
• Jun Xiao
• Xiaofei He
• Shiliang Pu

Conversational video question answering is a challenging task in visual information retrieval, which generates the accurate answer from the referenced video contents according to the visual conversation context and given question. However, the existing visual question answering methods mainly tackle the problem of single-turn video question answering, which may be ineffectively applied for multi-turn video question answering directly, due to the insufficiency of modeling the sequential conversation context. In this paper, we study the problem of multi-turn video question answering from the viewpoint of multi-step hierarchical attention context network learning. We first propose the hierarchical attention context network for context-aware question understanding by modeling the hierarchically sequential conversation context structure. We then develop the multi-stream spatio-temporal attention network for learning the joint representation of the dynamic video contents and context-aware question embedding. We next devise the hierarchical attention context network learning method with multi-step reasoning process for multi-turn video question answering. We construct two large-scale multi-turn video question answering datasets. The extensive experiments show the effectiveness of our method.

IJCAI Conference 2018 Conference Paper

Open-Ended Long-form Video Question Answering via Adaptive Hierarchical Reinforced Networks

• Zhou Zhao
• Zhu Zhang
• Shuwen Xiao
• Zhou Yu
• Jun Yu
• Deng Cai
• Fei Wu
• Yueting Zhuang

Open-ended long-form video question answering is challenging problem in visual information retrieval, which automatically generates the natural language answer from the referenced long-form video content according to the question. However, the existing video question answering works mainly focus on the short-form video question answering, due to the lack of modeling the semantic representation of long-form video contents. In this paper, we consider the problem of long-form video question answering from the viewpoint of adaptive hierarchical reinforced encoder-decoder network learning. We propose the adaptive hierarchical encoder network to learn the joint representation of the long-form video contents according to the question with adaptive video segmentation. we then develop the reinforced decoder network to generate the natural language answer for open-ended video question answering. We construct a large-scale long-form video question answering dataset. The extensive experiments show the effectiveness of our method.

AAAI Conference 2018 Conference Paper

PixelLink: Detecting Scene Text via Instance Segmentation

• Dan Deng
• Haifeng Liu
• Xuelong Li
• Deng Cai

Most state-of-the-art scene text detection algorithms are deep learning based methods that depend on bounding box regression and perform at least two kinds of predictions: text/nontext classification and location regression. Regression plays a key role in the acquisition of bounding boxes in these methods, but it is not indispensable because text/non-text prediction can also be considered as a kind of semantic segmentation that contains full location information in itself. However, text instances in scene images often lie very close to each other, making them very difficult to separate via semantic segmentation. Therefore, instance segmentation is needed to address this problem. In this paper, PixelLink, a novel scene text detection algorithm based on instance segmentation, is proposed. Text instances are first segmented out by linking pixels within the same instance together. Text bounding boxes are then extracted directly from the segmentation result without location regression. Experiments show that, compared with regression-based methods, PixelLink can achieve better or comparable performance on several benchmarks, while requiring many fewer training iterations and less training data.

IJCAI Conference 2018 Conference Paper

Translating Embeddings for Knowledge Graph Completion with Relation Attention Mechanism

• Wei Qian
• Cong Fu
• Yu Zhu
• Deng Cai
• Xiaofei He

Knowledge graph embedding is an essential problem in knowledge extraction. Recently, translation based embedding models (e. g. , TransE) have received increasingly attentions. These methods try to interpret the relations among entities as translations from head entity to tail entity and achieve promising performance on knowledge graph completion. Previous researchers attempt to transform the entity embedding concerning the given relation for distinguishability. Also, they naturally think the relation-related transforming should reflect attention mechanism, which means it should focus on only a part of the attributes. However, we found previous methods are failed with creating attention mechanism, and the reason is that they ignore the hierarchical routine of human cognition. When predicting whether a relation holds between two entities, people first check the category of entities, then they focus on fined-grained relation-related attributes to make the decision. In other words, the attention should take effect on entities filtered by the right category. In this paper, we propose a novel knowledge graph embedding method named TransAt to learn the translation based embedding, relation-related categories of entities and relation-related attention simultaneously. Extensive experiments show that our approach outperforms state-of-the-art methods significantly on public datasets, and our method can learn the true attention varying among relations.

AAAI Conference 2017 Conference Paper

Community-Based Question Answering via Asymmetric Multi-Faceted Ranking Network Learning

• Zhou Zhao
• Hanqing Lu
• Vincent Zheng
• Deng Cai
• Xiaofei He
• Yueting Zhuang

Nowadays the community-based question answering (CQA) sites become the popular Internet-based web service, which have accumulated millions of questions and their posted answers over time. Thus, question answering becomes an essential problem in CQA sites, which ranks the high-quality answers to the given question. Currently, most of the existing works study the problem of question answering based on the deep semantic matching model to rank the answers based on their semantic relevance, while ignoring the authority of answerers to the given question. In this paper, we consider the problem of community-based question answering from the viewpoint of asymmetric multi-faceted ranking network embedding. We propose a novel asymmetric multi-faceted ranking network learning framework for community-based question answering by jointly exploiting the deep semantic relevance between question-answer pairs and the answerers’ authority to the given question. We then develop an asymmetric ranking network learning method with deep recurrent neural networks by integrating both answers’ relative quality rank to the given question and the answerers’ following relations in CQA sites. The extensive experiments on a large-scale dataset from a real world CQA site show that our method achieves better performance than other state-of-the-art solutions to the problem.

IJCAI Conference 2017 Conference Paper

Link Prediction via Ranking Metric Dual-Level Attention Network Learning

• Zhou Zhao
• Ben Gao
• Vincent W. Zheng
• Deng Cai
• Xiaofei He
• Yueting Zhuang

Link prediction is a challenging problem for complex network analysis, arising in many disciplines such as social networks and telecommunication networks. Currently, many existing approaches estimate the proximity of the link endpoints for link prediction from their feature or the local neighborhood around them, which suffer from the localized view of network connections and insufficiency of discriminative feature representation. In this paper, we consider the problem of link prediction from the viewpoint of learning discriminative path-based proximity ranking metric embedding. We propose a novel ranking metric network learning framework by jointly exploiting both node-level and path-level attentional proximity of the endpoints for link prediction. We then develop the path-based dual-level reasoning attentional learning method with recurrent neural network for proximity ranking metric embedding. The extensive experiments on two large-scale datasets show that our method achieves better performance than other state-of-the-art solutions to the problem.

IJCAI Conference 2017 Conference Paper

Microblog Sentiment Classification via Recurrent Random Walk Network Learning

• Zhou Zhao
• Hanqing Lu
• Deng Cai
• Xiaofei He
• Yueting Zhuang

Microblog Sentiment Classification (MSC) is a challenging task in microblog mining, arising in many applications such as stock price prediction and crisis management. Currently, most of the existing approaches learn the user sentiment model from their posted tweets in microblogs, which suffer from the insufficiency of discriminative tweet representation. In this paper, we consider the problem of microblog sentiment classification from the viewpoint of heterogeneous MSC network embedding. We propose a novel recurrent random walk network learning framework for the problem by exploiting both users’ posted tweets and their social relations in microblogs. We then introduce the deep recurrent neural networks with random-walk layer for heterogeneous MSC network embedding, which can be trained end-to-end from the scratch. Weemploytheback-propagationmethodfortraining the proposed recurrent random walk network model. The extensive experiments on the large-scale public datasets from Twitter show that our method achieves better performance than other state-of-the-art solutions to the problem.

IJCAI Conference 2017 Conference Paper

Robust Asymmetric Bayesian Adaptive Matrix Factorization

• Xin Guo
• Boyuan Pan
• Deng Cai
• Xiaofei He

Low rank matrix factorizations(LRMF) have attracted much attention due to its wide range of applications in computer vision, such as image impainting and video denoising. Most of the existing methods assume that the loss between an observed measurement matrix and its bilinear factorization follows symmetric distribution, like gaussian or gamma families. However, in real-world situations, this assumption is often found too idealized, because pictures under various illumination and angles may suffer from multi-peaks, asymmetric and irregular noises. To address these problems, this paper assumes that the loss follows a mixture of Asymmetric Laplace distributions and proposes robust Asymmetric Laplace Adaptive Matrix Factorization model(ALAMF) under bayesian matrix factorization framework. The assumption of Laplace distribution makes our model more robust and the asymmetric attribute makes our model more flexible and adaptable to real-world noise. A variational method is then devised for model inference. We compare ALAMF with other state-of-the-art matrix factorization methods both on data sets ranging from synthetic and real-world application. The experimental results demonstrate the effectiveness of our proposed approach.

IJCAI Conference 2017 Conference Paper

Stacked Similarity-Aware Autoencoders

• Wenqing Chu
• Deng Cai

As one of the most popular unsupervised learning approaches, the autoencoder aims at transforming the inputs to the outputs with the least discrepancy. The conventional autoencoder and most of its variants only consider the one-to-one reconstruction, which ignores the intrinsic structure of the data and may lead to overfitting. In order to preserve the latent geometric information in the data, we propose the stacked similarity-aware autoencoders. To train each single autoencoder, we first obtain the pseudo class label of each sample by clustering the input features. Then the hidden codes of those samples sharing the same category label will be required to satisfy an additional similarity constraint. Specifically, the similarity constraint is implemented based on an extension of the recently proposed center loss. With this joint supervision of the autoencoder reconstruction error and the center loss, the learned feature representations not only can reconstruct the original data, but also preserve the geometric structure of the data. Furthermore, a stacked framework is introduced to boost the representation capacity. The experimental results on several benchmark datasets show the remarkable performance improvement of the proposed algorithm compared with other autoencoder based approaches.

IJCAI Conference 2017 Conference Paper

Video Question Answering via Hierarchical Spatio-Temporal Attention Networks

• Zhou Zhao
• Qifan Yang
• Deng Cai
• Xiaofei He
• Yueting Zhuang

Open-ended video question answering is a challenging problem in visual information retrieval, which automatically generates the natural language answer from the referenced video content according to the question. However, the existing visual question answering works only focus on the static image, which may be ineffectively applied to video question answering due to the temporal dynamics of video contents. In this paper, we consider the problem of open-ended video question answering from the viewpoint of spatio-temporal attentional encoder-decoder learning framework. We propose the hierarchical spatio-temporal attention network for learning the joint representation of the dynamic video contents according to the given question. We then develop the encoder-decoder learning method with reasoning recurrent neural networks for open-ended video question answering. We construct a large-scale video question answering dataset. The extensive experiments show the effectiveness of our method.

IJCAI Conference 2017 Conference Paper

What to Do Next: Modeling User Behaviors by Time-LSTM

• Yu Zhu
• Hao Li
• Yikang Liao
• Beidou Wang
• Ziyu Guan
• Haifeng Liu
• Deng Cai

Recently, Recurrent Neural Network (RNN) solutions for recommender systems (RS) are becoming increasingly popular. The insight is that, there exist some intrinsic patterns in the sequence of users' actions, and RNN has been proved to perform excellently when modeling sequential data. In traditional tasks such as language modeling, RNN solutions usually only consider the sequential order of objects without the notion of interval. However, in RS, time intervals between users' actions are of significant importance in capturing the relations of users' actions and the traditional RNN architectures are not good at modeling them. In this paper, we propose a new LSTM variant, i. e. Time-LSTM, to model users' sequential actions. Time-LSTM equips LSTM with time gates to model time intervals. These time gates are specifically designed, so that compared to the traditional RNN solutions, Time-LSTM better captures both of users' short-term and long-term interests, so as to improve the recommendation performance. Experimental results on two real-world datasets show the superiority of the recommendation method using Time-LSTM over the traditional methods.

AAAI Conference 2016 Conference Paper

Accelerated Sparse Linear Regression via Random Projection

• Weizhong Zhang
• Lijun Zhang
• Rong Jin
• Deng Cai
• Xiaofei He

In this paper, we present an accelerated numerical method based on random projection for sparse linear regression. Previous studies have shown that under appropriate conditions, gradient-based methods enjoy a geometric convergence rate when applied to this problem. However, the time complexity of evaluating the gradient is as large as O(nd), where n is the number of data points and d is the dimensionality, making those methods inefficient for large-scale and highdimensional dataset. To address this limitation, we first utilize random projection to find a rank-k approximator for the data matrix, and reduce the cost of gradient evaluation to O(nk + dk), a significant improvement when k is much smaller than d and n. Then, we solve the sparse linear regression problem via a proximal gradient method with a homotopy strategy to generate sparse intermediate solutions. Theoretical analysis shows that our method also achieves a global geometric convergence rate, and moreover the sparsity of all the intermediate solutions are well-bounded over the iterations. Finally, we conduct experiments to demonstrate the ef- ficiency of the proposed method.

IJCAI Conference 2016 Conference Paper

Expert Finding for Community-Based Question Answering via Ranking Metric Network Learning

• Zhou Zhao
• Qifan Yang
• Deng Cai
• Xiaofei He
• Yueting Zhuang

Expert finding for question answering is a challenging problem in Community-based Question Answering(CQA) site, arising in many applications such as question routing and the identification of best answers. In order to provide high-quality experts, many existing approaches learn the user model mainly from their past question-answering activities in CQA sites, which suffer from the sparsity problem of CQA data. In this paper, we consider the problem of expert finding from the viewpoint of learning ranking metric embedding. We propose a novel ranking metric network learning framework for expert finding by exploiting both users' relative quality rank to given questions and their social relations. We then develop a random-walk based learning method with recurrent neural networks for ranking metric network embedding. The extensive experiments on a large-scale dataset from a real world CQA site show that our method achieves better performance than other state-of-the-art solutions to the problem.

IJCAI Conference 2016 Conference Paper

Non-Negative Matrix Factorization with Sinkhorn Distance

• Wei Qian
• Bin Hong
• Deng Cai
• Xiaofei He
• Xuelong Li

Non-negative Matrix Factorization (NMF) has received considerable attentions in various areas for its psychological and physiological interpretation of naturally occurring data whose representation may be parts-based in the human brain. Despite its good practical performance, one shortcoming of original NMF is that it ignores intrinsic structure of data set. On one hand, samples might be on a manifold and thus one may hope that geometric information can be exploited to improve NMF's performance. On the other hand, features might correlate with each other, thus conventional L 2 distance can not well measure the distance between samples. Although some works have been proposed to solve these problems, rare connects them together. In this paper, we propose a novel method that exploits knowledge in both data manifold and features correlation. We adopt an approximation of Earth Mover's Distance (EMD) as metric and add a graph regularized term based on EMD to NMF. Furthermore, we propose an efficient multiplicative iteration algorithm to solve it. Our empirical study shows the encouraging results of the proposed algorithm comparing with other NMF methods.

IJCAI Conference 2016 Conference Paper

Weakly-Supervised Deep Learning for Customer Review Sentiment Classification

• Ziyu Guan
• Long Chen
• Wei Zhao
• Yi Zheng
• Shulong Tan
• Deng Cai

Sentiment analysis is one of the key challenges for mining online user generated content. In this work, we focus on customer reviews which are an important form of opinionated content. The goal is to identify each sentence's semantic orientation (e. g. positive or negative) of a review. Traditional sentiment classification methods often involve substantial human efforts, e. g. lexicon construction, feature engineering. In recent years, deep learning has emerged as an effective means for solving sentiment classification problems. A neural network intrinsically learns a useful representation automatically without human efforts. However, the success of deep learning highly relies on the availability of large-scale training data. In this paper, we propose a novel deep learning framework for review sentiment classification which employs prevalently available ratings as weak supervision signals. The framework consists of two steps: (1) learn a high level representation (embedding space) which captures the general sentiment distribution of sentences through rating information; (2) add a classification layer on top of the embedding layer and use labeled sentences for supervised fine-tuning. Experiments on review data obtained from Amazon show the efficacy of our method and its superiority over baseline methods.

IJCAI Conference 2015 Conference Paper

Compressed Spectral Regression for Efficient Nonlinear Dimensionality Reduction

• Deng Cai

Spectral dimensionality reduction methods have recently emerged as powerful tools for various applications in pattern recognition, data mining and computer vision. These methods use information contained in the eigenvectors of a data affinity (i. e. , item-item similarity) matrix to reveal the low dimensional structure of the high dimensional data. One of the limitations of various spectral dimensionality reduction methods is their high computational complexity. They all need to construct a data affinity matrix and compute the top eigenvectors. This leads to O(n2 ) computational complexity, where n is the number of samples. Moreover, when the data are highly non-linear distributed, some linear methods have to be performed in a reproducing kernel Hilbert space (leads to the corresponding kernel methods) to learn an effective non-linear mapping. The computational complexity of these kernel methods is O(n3 ). In this paper, we propose a novel nonlinear dimensionality reduction algorithm, called Compressed Spectral Regression, with O(n) computational complexity. Extensive experiments on data clustering demonstrate the effectiveness and efficiency of the proposed approach.

AAAI Conference 2014 Conference Paper

Mapping Users across Networks by Manifold Alignment on Hypergraph

• Shulong Tan
• Ziyu Guan
• Deng Cai
• Xuzhen Qin
• Jiajun Bu
• Chun Chen

Nowadays many people are members of multiple online social networks simultaneously, such as Facebook, Twitter and some other instant messaging circles. But these networks are usually isolated from each other. Mapping common users across these social networks will benefit many applications. Methods based on username comparison perform well on parts of users, however they can not work in the following situations: (a) users choose different usernames in different networks; (b) a unique username corresponds to different individuals. In this paper, we propose to utilize social structures to improve the mapping performance. Specifically, a novel subspace learning algorithm, Manifold Alignment on Hypergraph (MAH), is proposed. Different from traditional semi-supervised manifold alignment methods, we use hypergraph to model high-order relations here. For a target user in one network, the proposed algorithm ranks all users in the other network by their possibilities of being the corresponding user. Moreover, methods based on username comparison can be incorporated into our algorithm easily to further boost the mapping accuracy. Experimental results have demonstrated the effectiveness of our proposed algorithm in mapping users across networks.

AAAI Conference 2014 Conference Paper

Sparse Learning for Stochastic Composite Optimization

• Weizhong Zhang
• Lijun Zhang
• Yao Hu
• Rong Jin
• Deng Cai
• Xiaofei He

In this paper, we focus on Stochastic Composite Optimization (SCO) for sparse learning that aims to learn a sparse solution. Although many SCO algorithms have been developed for sparse learning with an optimal convergence rate O(1/T), they often fail to deliver sparse solutions at the end either because of the limited sparsity regularization during stochastic optimization or due to the limitation in online-to-batch conversion. To improve the sparsity of solutions obtained by SCO, we propose a simple but effective stochastic optimization scheme that adds a novel sparse online-to-batch conversion to the traditional SCO algorithms. The theoretical analysis shows that our scheme can find a solution with better sparse patterns without affecting the convergence rate. Experimental results on both synthetic and real-world data sets show that the proposed methods are more effective in recovering the sparse solution and have comparable convergence rate as the state-of-the-art SCO algorithms for sparse learning.

AAAI Conference 2014 Conference Paper

Who Also Likes It? Generating the Most Persuasive Social Explanations in Recommender Systems

• Beidou Wang
• Martin Ester
• Jiajun Bu
• Deng Cai

IJCAI Conference 2013 Conference Paper

A Unified Approximate Nearest Neighbor Search Scheme by Combining Data Structure and Hashing

• Debing Zhang
• Genmao Yang
• Yao Hu
• Zhongming Jin
• Deng Cai
• Xiaofei He

Nowadays, Nearest Neighbor Search becomes more and more important when facing the challenge of big data. Traditionally, to solve this problem, researchers mainly focus on building effective data structures such as hierarchical k-means tree or using hashing methods to accelerate the query process. In this paper, we propose a novel unified approximate nearest neighbor search scheme to combine the advantages of both the effective data structure and the fast Hamming distance computation in hashing methods. In this way, the searching procedure can be further accelerated. Computational complexity analysis and extensive experiments have demonstrated the effectiveness of our proposed scheme.

IJCAI Conference 2013 Conference Paper

Active Learning via Neighborhood Reconstruction

• Yao Hu
• Debing Zhang
• Zhongming Jin
• Deng Cai
• Xiaofei He

In many real world scenarios, active learning methods are used to select the most informative points for labeling to reduce the expensive human action. One direction for active learning is selecting the most representative points, ie. , selecting the points that other points can be approximated by linear combination of the selected points. However, these methods fails to consider the local geometrical information of the data space. In this paper, we propose a novel framework named Active Learning via Neighborhood Reconstruction (ALNR) by taking into account the locality information directly during the selection. Specifically, for the linear reconstruction of target point, the nearer neighbors should have a greater effect and the selected points distant from the target point should be penalized severely. We further develop an efficient two-stage iterative procedure to solve the final optimization problem. Our empirical study shows encouraging results of the proposed algorithms in comparison to other state-of-the-art active learning algorithms on both synthetic and real visual data sets.

IJCAI Conference 2013 Conference Paper

Bilevel Visual Words Coding for Image Classification

• Jiemi Zhang
• Chenxia Wu
• Deng Cai
• Jianke Zhu

Bag-of-Words approach has played an important role in recent works for image classification. In consideration of efficiency, most methods use kmeans clustering to generate the codebook. The obtained codebooks often lose the cluster size and shape information with distortion errors and low discriminative power. Though some efforts have been made to optimize codebook in sparse coding, they usually incur higher computational cost. Moreover, they ignore the correlations between codes in the following coding stage, that leads to low discriminative power of the final representation. In this paper, we propose a bilevel visual words coding approach in consideration of representation ability, discriminative power and efficiency. In the bilevel codebook generation stage, k-means and an efficient spectral clustering are respectively run in each level by taking both class information and the shapes of each visual word cluster into account. To obtain discriminative representation in the coding stage, we design a certain localized coding rule with bilevel codebook to select local bases. To further achieve an efficient coding referring to this rule, an online method is proposed to efficiently learn a projection of local descriptor to the visual words in the codebook. After projection, coding can be efficiently completed by a low dimensional localized soft-assignment. Experimental results show that our proposed bilevel visual words coding approach outperforms the stateof-the-art approaches for image classification.

IJCAI Conference 2013 Conference Paper

Harmonious Hashing

• Bin Xu
• Jiajun Bu
• Yue Lin
• Chun Chen
• Xiaofei He
• Deng Cai

Hashing-based fast nearest neighbor search technique has attracted great attention in both research and industry areas recently. Many existing hashing approaches encode data with projection-based hash functions and represent each projected dimension by 1-bit. However, the dimensions with high variance hold large energy or information of data but treated equivalently as dimensions with low variance, which leads to a serious information loss. In this paper, we introduce a novel hashing algorithm called Harmonious Hashing which aims at learning hash functions with low information loss. Specifically, we learn a set of optimized projections to preserve the maximum cumulative energy and meet the constraint of equivalent variance on each dimension as much as possible. In this way, we could minimize the information loss after binarization. Despite the extreme simplicity, our method outperforms superiorly to many state-of-the-art hashing methods in large-scale and high-dimensional nearest neighbor search experiments.

AAAI Conference 2012 Conference Paper

A Bregman Divergence Optimization Framework for Ranking on Data Manifold and Its New Extensions

• Bin Xu
• Jiajun Bu
• Chun Chen
• Deng Cai

Recently, graph-based ranking algorithms have received considerable interests in machine learning, computer vision and information retrieval communities. Ranking on data manifold (or manifold ranking, MR) is one of the representative approaches. One of the limitations of manifold ranking is its high computational complexity (O(n3 ), where n is the number of samples in database). In this paper, we cast the manifold ranking into a Bregman divergence optimization framework under which we transform the original MR to an equivalent optimal kernel matrix learning problem. With this new formulation, two effective and efficient extensions are proposed to enhance the ranking performance. Extensive experimental results on two real world image databases show the effectiveness of the proposed approach.

AAAI Conference 2012 Conference Paper

Document Summarization Based on Data Reconstruction

• Zhanying He
• Chun Chen
• Jiajun Bu
• Can Wang
• Lijun Zhang
• Deng Cai
• Xiaofei He

Document summarization is of great value to many real world applications, such as snippets generation for search results and news headlines generation. Traditionally, document summarization is implemented by extracting sentences that cover the main topics of a document with a minimum redundancy. In this paper, we take a different perspective from data reconstruction and propose a novel framework named Document Summarization based on Data Reconstruction (DSDR). Specifically, our approach generates a summary which consist of those sentences that can best reconstruct the original document. To model the relationship among sentences, we introduce two objective functions: (1) linear reconstruction, which approximates the document by linear combinations of the selected sentences; (2) nonnegative linear reconstruction, which allows only additive, not subtractive, linear combinations. In this framework, the reconstruction error becomes a natural criterion for measuring the quality of the summary. For each objective function, we develop an efficient algorithm to solve the corresponding optimization problem. Extensive experiments on summarization benchmark data sets DUC 2006 and DUC 2007 demonstrate the effectiveness of our proposed approach.

AAAI Conference 2012 Conference Paper

Random Projection with Filtering for Nearly Duplicate Search

• Yue Lin
• Rong Jin
• Deng Cai
• Xiaofei He

High dimensional nearest neighbor search is a fundamental problem and has found applications in many domains. Although many hashing based approaches have been proposed for approximate nearest neighbor search in high dimensional space, one main drawback is that they often return many false positives that need to be filtered out by a post procedure. We propose a novel method to address this limitation in this paper. The key idea is to introduce a filtering procedure within the search algorithm, based on the compressed sensing theory, that effectively removes the false positive answers. We first obtain a sparse representation for each data point by the landmark based approach, after which we solve the nearly duplicate search that the difference between the query and its nearest neighbors forms a sparse vector living in a small `p ball, where p ≤ 1. Our empirical study on real-world datasets demonstrates the effectiveness of the proposed approach compared to the state-of-the-art hashing methods.

AAAI Conference 2011 Conference Paper

Large Scale Spectral Clustering with Landmark-Based Representation

• Xinlei Chen
• Deng Cai

Spectral clustering is one of the most popular clustering approaches. Despite its good performance, it is limited in its applicability to large-scale problems due to its high computational complexity. Recently, many approaches have been proposed to accelerate the spectral clustering. Unfortunately, these methods usually sacri- fice quite a lot information of the original data, thus result in a degradation of performance. In this paper, we propose a novel approach, called Landmark-based Spectral Clustering (LSC), for large scale clustering problems. Specifically, we select p ( n) representative data points as the landmarks and represent the original data points as the linear combinations of these landmarks. The spectral embedding of the data can then be efficiently computed with the landmark-based representation. The proposed algorithm scales linearly with the problem size. Extensive experiments show the effectiveness and efficiency of our approach comparing to the state-of-the-art methods.

TIST Journal 2011 Journal Article

Variational inference with graph regularization for image annotation

• Yuanlong Shao
• Yuan Zhou
• Deng Cai

Image annotation is a typical area where there are multiple types of attributes associated with each individual image. In order to achieve better performance, it is important to develop effective modeling by utilizing prior knowledge. In this article, we extend the graph regularization approaches to a more general case where the regularization is imposed on the factorized variational distributions, instead of posterior distributions implicitly involved in EM-like algorithms. In this way, the problem modeling can be more flexible, and we can choose any factor in the problem domain to impose graph regularization wherever there are similarity constraints among the instances. We formulate the problem formally and show its geometrical background in manifold learning. We also design two practically effective algorithms and analyze their properties such as the convergence. Finally, we apply our approach to image annotation and show the performance improvement of our algorithm.

AAAI Conference 2010 Conference Paper

Gaussian Mixture Model with Local Consistency

• Jialu Liu
• Deng Cai
• Xiaofei He

Gaussian Mixture Model (GMM) is one of the most popular data clustering methods which can be viewed as a linear combination of different Gaussian components. In GMM, each cluster obeys Gaussian distribution and the task of clustering is to group observations into different components through estimating each cluster’s own parameters. The Expectation- Maximization algorithm is always involved in such estimation problem. However, many previous studies have shown naturally occurring data may reside on or close to an underlying submanifold. In this paper, we consider the case where the probability distribution is supported on a submanifold of the ambient space. We take into account the smoothness of the conditional probability distribution along the geodesics of data manifold. That is, if two observations are “close” in intrinsic geometry, their distributions over different Gaussian components are similar. Simply speaking, we introduce a novel method based on manifold structure for data clustering, called Locally Consistent Gaussian Mixture Model (LCGMM). Specifically, we construct a nearest neighbor graph and adopt Kullback-Leibler Divergence as the “distance” measurement to regularize the objective function of GMM. Experiments on several data sets demonstrate the effectiveness of such regularization.

IJCAI Conference 2009 Conference Paper

• Deng Cai
• Xiaofei He
• Xuanhui Wang
• Hujun Bao
• Jiawei Han

Matrix factorization techniques have been frequently applied in information processing tasks. Among them, Non-negative Matrix Factorization (NMF) have received considerable attentions due to its psychological and physiological interpretation of naturally occurring data whose representation may be parts-based in human brain. On the other hand, from geometric perspective the data is usually sampled from a low dimensional manifold embedded in high dimensional ambient space. One hopes then to find a compact representation which uncovers the hidden topics and simultaneously respects the intrinsic geometric structure. In this paper, we propose a novel algorithm, called Locality Preserving Non-negative Matrix Factorization (LPNMF), for this purpose. For two data points, we use KL-divergence to evaluate their similarity on the hidden topics. The optimal maps are obtained such that the feature values on hidden topics are restricted to be non-negative and vary smoothly along the geodesics of the data manifold. Our empirical study shows the encouraging results of the proposed algorithm in comparisons to the state-ofthe-art algorithms on two large high-dimensional databases.

AAAI Conference 2008 Conference Paper

Sparse Projections over Graph

• Deng Cai

Recent study has shown that canonical algorithms such as Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) can be obtained from graph based dimensionality reduction framework. However, these algorithms yield projective maps which are linear combination of all the original features. The results are difficult to be interpreted psychologically and physiologically. This paper presents a novel technique for learning a sparse projection over graphs. The data in the reduced subspace is represented as a linear combination of a subset of the most relevant features. Comparing to PCA and LDA, the results obtained by sparse projection are often easier to be interpreted. Our algorithm is based on a graph embedding model, which encodes the discriminating and geometrical structure in terms of the data affinity. Once the embedding results are obtained, we then apply regularized regression for learning a set of sparse basis functions. Specifically, by using a L1-norm regularizer (e. g. lasso), the sparse projections can be efficiently computed. Experimental results on two document databases demonstrate the effectiveness of our method.

IJCAI Conference 2007 Conference Paper

• Deng Cai
• Xiaofei He
• Kun Zhou
• Jiawei Han
• Hujun Bao

Linear Discriminant Analysis (LDA) is a popular data-analytic tool for studying the class relationship between data points. A major disadvantage of LDA is that it fails to discover the local geometrical structure of the data manifold. In this paper, we introduce a novel linear algorithm for discriminant analysis, called {\bf Locality Sensitive Discriminant Analysis} (LSDA). When there is no sufficient training samples, local structure is generally more important than global structure for discriminant analysis. By discovering the local manifold structure, LSDA finds a projection which maximizes the margin between data points from different classes at each local area. Specifically, the data points are mapped into a subspace in which the nearby points with the same label are close to each other while the nearby points with different labels are far apart. Experiments carried out on several standard face databases show a clear improvement over the results of LDA-based recognition.

AAAI Conference 2007 Conference Paper

Isometric Projection

• Deng Cai

Recently the problem of dimensionality reduction has received a lot of interests in many fields of information processing. We consider the case where data is sampled from a low dimensional manifold which is embedded in high dimensional Euclidean space. The most popular manifold learning algorithms include Locally Linear Embedding, ISOMAP, and Laplacian Eigenmap. However, these algorithms are nonlinear and only provide the embedding results of training samples. In this paper, we propose a novel linear dimensionality reduction algorithm, called Isometric Projection. Isometric Projection constructs a weighted data graph where the weights are discrete approximations of the geodesic distances on the data manifold. A linear subspace is then obtained by preserving the pairwise distances. In this way, Isometric Projection can be defined everywhere. Comparing to Principal Component Analysis (PCA) which is widely used in data processing, our algorithm is more capable of discovering the intrinsic geometrical structure. Specially, PCA is optimal only when the data space is linear, while our algorithm has no such assumption and therefore can handle more complex data space. Experimental results on two real life data sets illustrate the effectiveness of the proposed method.

NeurIPS Conference 2005 Conference Paper

Laplacian Score for Feature Selection

• Xiaofei He
• Deng Cai
• Partha Niyogi

In supervised learning scenarios, feature selection has been studied widely in the literature. Selecting features in unsupervised learning scenarios is a much harder problem, due to the absence of class labels that would guide the search for relevant information. And, almost all of previous unsupervised feature selection methods are "wrapper" techniques that require a learning algorithm to evaluate the candidate feature subsets. In this paper, we propose a "filter" method for feature selection which is independent of any learning algorithm. Our method can be performed in either supervised or unsupervised fashion. The proposed method is based on the observation that, in many real world classification problems, data from the same class are often close to each other. The importance of a feature is evaluated by its power of locality preserving, or, Laplacian Score. We compare our method with data variance (unsupervised) and Fisher score (supervised) on two data sets. Experimental results demonstrate the effectiveness and efficiency of our algorithm.

NeurIPS Conference 2005 Conference Paper

Tensor Subspace Analysis

• Xiaofei He
• Deng Cai
• Partha Niyogi

Previous work has demonstrated that the image variations of many ob- jects (human faces in particular) under variable lighting can be effec- tively modeled by low dimensional linear spaces. The typical linear sub- space learning algorithms include Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Locality Preserving Projec- tion (LPP). All of these methods consider an n1 × n2 image as a high dimensional vector in Rn1×n2, while an image represented in the plane is intrinsically a matrix. In this paper, we propose a new algorithm called Tensor Subspace Analysis (TSA). TSA considers an image as the sec- ond order tensor in Rn1 ⊗ Rn2, where Rn1 and Rn2 are two vector spaces. The relationship between the column vectors of the image ma- trix and that between the row vectors can be naturally characterized by TSA. TSA detects the intrinsic local geometrical structure of the tensor space by learning a lower dimensional tensor subspace. We compare our proposed approach with PCA, LDA and LPP methods on two standard databases. Experimental results demonstrate that TSA achieves better recognition rate, while being much more efficient.