Author name cluster

Pengfei Li

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

24 papers

2 author rows

IROS Conference 2025 Conference Paper

Bench4Merge: A Comprehensive Benchmark for Merging in Realistic Dense Traffic with Micro-Interactive Vehicles

Zhengming Wang
Junli Wang
Pengfei Li
Zhaohan Li
Chunyang Liu
Bo Zhang
Peng Li
Yilun Chen

While the capabilities of autonomous driving have advanced rapidly, merging into dense traffic remains a significant challenge, many motion planning methods for this scenario have been proposed but it is hard to evaluate them. Most existing closed-loop simulators rely on rule-based controls for other vehicles, which results in a lack of diversity and randomness, thus failing to accurately assess the motion planning capabilities in highly interactive scenarios. Moreover, traditional evaluation metrics are insufficient for comprehensively evaluating the performance of merging in dense traffic. In response, we proposed a closed-loop evaluation benchmark for assessing motion planning capabilities in merging scenarios. Our approach involves other vehicles trained in large scale datasets with micro-behavioral characteristics that significantly enhance the complexity and diversity. Additionally, we have restructured the evaluation mechanism by leveraging Large Language Models (LLMs) to assess each autonomous vehicle merging onto the main lane. Extensive experiments and test-vehicle deployment have demonstrated the progressiveness of this benchmark. Through this benchmark, we have obtained an evaluation of existing methods and identified common issues. The simulation environment and evaluation process can be accessed at https://github.com/WZM5853/Bench4Merge.

Details

NeurIPS Conference 2025 Conference Paper

Bohdi: Heterogeneous LLM Fusion with Automatic Data Exploration

俊琪高
Zhichang Guo
Dazhi Zhang
Dong Li
Runze Liu
Pengfei Li
Kai Tian
Biqing Qi

Heterogeneous Large Language Model (LLM) fusion integrates the strengths of multiple source LLMs with different architectures into a target LLM with low computational overhead. While promising, existing methods suffer from two major limitations: 1) reliance on real data from limited domain for knowledge fusion, preventing the target LLM from fully acquiring knowledge across diverse domains, and 2) fixed data allocation proportions across domains, failing to dynamically adjust according to the target LLM's varying capabilities across domains, leading to a capability imbalance. To overcome these limitations, we propose Bohdi, a synthetic-data-only heterogeneous LLM fusion framework. Through the organization of knowledge domains into a hierarchical tree structure, Bohdi enables automatic domain exploration and multi-domain data generation through multi-model collaboration, thereby comprehensively extracting knowledge from source LLMs. By formalizing domain expansion and data sampling proportion allocation on the knowledge tree as a Hierarchical Multi-Armed Bandit problem, Bohdi leverages the designed DynaBranches mechanism to adaptively adjust sampling proportions based on the target LLM's performance feedback across domains. Integrated with our proposed Introspection-Rebirth (IR) mechanism, DynaBranches dynamically tracks capability shifts during target LLM's updates via Sliding Window Binomial Likelihood Ratio Testing (SWBLRT), further enhancing its online adaptation capability. Comparative experimental results on a comprehensive suite of benchmarks demonstrate that Bohdi significantly outperforms existing baselines on multiple target LLMs, exhibits higher data efficiency, and virtually eliminates the imbalance in the target LLM's capabilities.

PDF Details

JBHI Journal 2025 Journal Article

Bridging Brain and Machine: A Neurocognitive Framework for Early Alzheimer's Detection Through Hierarchical Speech Processing

Xiaohong Lyu
Jiayuan Bai
Pengfei Li
Fan Zhang
Xianglin Zhang
Yun Gu
Ning Liu

Alzheimer's disease progressively disrupts temporal-frontal language networks, manifesting through degraded speech production detectable years before clinical dementia diagnosis. Traditional automated detection systems rely on handcrafted acoustic features, including mel-frequency cepstral coefficients, that may discard diagnostically relevant temporal dynamics through fixed spectral transformations. Drawing inspiration from cognitive neuroscience models of hierarchical speech processing, we propose the neuro residual acoustic learning framework, an end-to-end system operating directly on raw waveforms while incorporating architectural principles mirroring biological auditory pathways. The framework implements acoustic-symbolic transformation bridging continuous signals and discrete phonetic-acoustic categories analogous to primary auditory cortex processing, hierarchical temporal encoding through residual blocks with dilated convolutions capturing multi-scale patterns across phoneme to utterance timescales, and squeeze-and-excitation attention mechanisms modeling prefrontal modulation of temporal representations. Evaluation on NCMMSC2021 Mandarin speech demonstrates 86. 55% accuracy on extended utterances and 81. 35% on brief segments, surpassing contemporary methods, including multimodal architectures and large language model hybrids, by 2. 88% and 2. 43% respectively. Cross-linguistic validation on INTERSPEECH2020 ADReSS English speech achieves 70. 83% accuracy, exceeding baselines by 2. 08%. Confusion matrix analysis reveals 93. 3% healthy control recall and asymmetric Alzheimer's disease versus mild cognitive impairment misclassification patterns consistent with prodromal acoustic biomarkers.

Details DOI

NeurIPS Conference 2025 Conference Paper

Fairness-Regularized Online Optimization with Switching Costs

Pengfei Li
Yuelin Han
Adam Wierman
Shaolei Ren

Fairness and action smoothness are two crucial considerations in many online optimization problems, but they have yet to be addressed simultaneously. In this paper, we study a new and challenging setting of fairness-regularized smoothed online convex optimization with switching costs. First, to highlight the fundamental challenges introduced by the long-term fairness regularizer evaluated based on the entire sequence of actions, we prove that even without switching costs, no online algorithms can possibly achieve a sublinear regret or finite competitive ratio compared to the offline optimal algorithm as the problem episode length $T$ increases. Then, we propose **FairOBD** (Fairness-regularized Online Balanced Descent), which reconciles the tension between minimizing the hitting cost, switching cost, and fairness cost. Concretely, **FairOBD** decomposes the long-term fairness cost into a sequence of online costs by introducing an auxiliary variable and then leverages the auxiliary variable to regularize the online actions for fair outcomes. Based on a new approach to account for switching costs, we prove that **FairOBD** offers a worst-case asymptotic competitive ratio against a novel benchmark---the optimal offline algorithm with parameterized constraints---by considering $T\to\infty$. Finally, we run trace-driven experiments of dynamic computing resource provisioning for socially responsible AI inference to empirically evaluate **FairOBD**, showing that **FairOBD** can effectively reduce the total fairness-regularized cost and better promote fair outcomes compared to existing baseline solutions.

PDF Details

JBHI Journal 2025 Journal Article

Fall Detection Method Based on a Human Electrostatic Field and VMD-ECANet Architecture

Xi Chen
Jiaao Yan
Sichao Qin
Pengfei Li
Shuangqian Ning
Yuting Liu

Falls are one of the most serious health risks faced by older adults worldwide, and they can have a significant impact on their physical and mental well-being as well as their quality of life. Detecting falls promptly and accurately and providing assistance can effectively reduce the harm caused by falls to older adults. This paper proposed a noncontact fall detection method based on the human electrostatic field and a VMD-ECANet framework. An electrostatic measurement system was used to measure the electrostatic signals of four types of falling postures and five types of daily actions. The signals were randomly divided in proportion and by individuals to construct a training set and test set. A fall detection model based on the VMD-ECA network was proposed that decomposes electrostatic signals into modal component signals using the variational mode decomposition (VMD) technique. These signals were then fed into a multichannel convolutional neural network for feature extraction. Information fusion was achieved through the efficient channel attention network (ECANet) module. Finally, the extracted features were input into a classifier to obtain the output results. The constructed model achieved an accuracy of 96. 44%. The proposed fall detection solution has several advantages, including being noncontact, cost-effective, and privacy friendly. It is suitable for detecting indoor falls by older individuals living alone and helps to reduce the harm caused by falls.

Details DOI

AAAI Conference 2025 Conference Paper

Fast and Slow Gradient Approximation for Binary Neural Network Optimization

Xinquan Chen
Junqi Gao
Biqing Qi
Dong Li
Yiang Luo
Fangyuan Li
Pengfei Li

Binary Neural Networks (BNNs) have garnered significant attention due to their immense potential for deployment on edge devices. However, the non-differentiability of the quantization function poses a challenge for the optimization of BNNs, as its derivative cannot be backpropagated. To address this issue, hypernetwork based methods, which utilize neural networks to learn the gradients of non-differentiable quantization functions, have emerged as a promising approach due to their adaptive learning capabilities to reduce estimation errors. However, existing hypernetwork based methods typically rely solely on current gradient information, neglecting the influence of historical gradients. This oversight can lead to accumulated gradient errors when calculating gradient momentum during optimization. To incorporate historical gradient information, we design a Historical Gradient Storage (HGS) module, which models the historical gradient sequence to generate the first-order momentum required for optimization. To further enhance gradient generation in hypernetworks, we propose a Fast and Slow Gradient Generation (FSG) method. Additionally, to produce more precise gradients, we introduce Layer Recognition Embeddings (LRE) into the hypernetwork, facilitating the generation of layer-specific fine gradients. Extensive comparative experiments on the CIFAR-10 and CIFAR-100 datasets demonstrate that our method achieves faster convergence and lower loss values, outperforming existing baselines.

PDF Details DOI

EAAI Journal 2025 Journal Article

Identification of surface subsidence risk in deep foundation pits using a Mamba fusion model

Chenhe Ge
Pengfei Li
Mingju Zhang
Meng Yang

This study proposes a novel artificial intelligence–driven neural network model, CNN-Mamba-LSTM-SA, for predicting surface settlement induced by deep foundation pit excavation. The model integrates Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), a Self-Attention (SA) mechanism, and the Mamba architecture to capture both spatial and long-range temporal dependencies in multi-source monitoring data. Bayesian optimization is employed for hyperparameter tuning, and Variational Mode Decomposition (VMD) is used for data denoising, resulting in improved prediction accuracy. To enhance model interpretability, Shapley Additive Explanations (SHAP) are applied to identify key deformation drivers, revealing groundwater level and building settlement as the most influential factors. Model performance is validated using monitoring data from the Nanjing Gemini excavation project, where it achieves superior results compared to conventional models. The CNN-Mamba-LSTM-SA model reduces Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE) by up to 69. 77 %, 62. 45 %, and 79. 45 %, respectively. Further analysis through ablation experiments confirms the contribution of each module. Interestingly, the combined removal of CNN and SA results in greater performance degradation than the sum of their individual effects. Finally, a risk warning framework is developed to enable the dynamic transformation of predicted and observed settlement values into actionable risk levels. This integrated artificial intelligence approach offers a robust and interpretable tool for managing excavation-induced risks in urban geotechnical engineering.

Details DOI

AAAI Conference 2025 Conference Paper

LiON: Learning Point-Wise Abstaining Penalty for LiDAR Outlier DetectioN Using Diverse Synthetic Data

Shaocong Xu
Pengfei Li
Qianpu Sun
Xinyu Liu
Yang Li
Shihui Guo
Zhen Wang
Bo Jiang

LiDAR-based semantic scene understanding is an important module in the modern autonomous driving perception stack. However, identifying outlier points in a LiDAR point cloud is challenging as LiDAR point clouds lack semantically-rich information. While former SOTA methods adopt heuristic architectures, we revisit this problem from the perspective of Selective Classification, which introduces a selective function into the standard closed-set classification setup. Our solution is built upon the basic idea of abstaining from choosing any inlier categories but learns a point-wise abstaining penalty with a margin-based loss. Apart from learning paradigms, synthesizing outliers to approximate unlimited real outliers is also critical, so we propose a strong synthesis pipeline that generates outliers originated from various factors: object categories, sampling patterns and sizes. We demonstrate that learning different abstaining penalties, apart from point-wise penalty, for different types of (synthesized) outliers can further improve the performance. We benchmark our method on SemanticKITTI and nuScenes and achieve SOTA results.

PDF Details DOI

NeurIPS Conference 2025 Conference Paper

LLM-PySC2: Starcraft II learning environment for Large Language Models

Zongyuan Li
Yanan Ni
Runnan Qi
Chang Lu
Lumin Jiang
Xu Xiaojie
Xiangbei Liu
Pengfei Li

The tremendous potential has been demonstrated by large language models (LLMs) in intelligent decision-making problems, with unprecedented capabilities shown across diverse applications ranging from gaming AI systems to complex strategic planning frameworks. However, the StarCraft II platform, which has been widely adopted for validating decision-making algorithms in the past decade, has not yet provided substantial support for this emerging domain. To address issues that LLMs cannot interface with the hundreds of actions of the pysc2 backend and the lack of native support for multi-agent (MA) collaboration, we propose the LLM-PySC2 environment. This is the first environment that offers LLMs the complete pysc2 action space with sufficient multi-modal information and game Wiki knowledge. With an asynchronous query architecture, the environment efficiently interacts with LLMs that maintain a constant latency regardless of the scale of the agents' population. In the experiments, we evaluated LLMs' decision-making performance in both the macro-decision and micro-operation scenarios, with traditional StarCraft II Multi-Agent Challenge (SMAC) tasks and a series of new proposed. Results indicate that LLMs possess the potential to achieve victories in complex scenarios but cannot constantly generate correct decisions, especially in the recovered pysc2 action space and MA settings. Without task-relevant instructions, the pre-trained models suffer from issues such as hallucinations and inefficient collaboration. Our findings suggest that StarCraft II still challenges in the era of large models, revealing that there is a lot to do to develop an advanced LLM decision-making system, and the proposed LLM-PySC2 environment will support future development of LLM-based decision-making solutions.

PDF Details

NeurIPS Conference 2024 Conference Paper

Exploring Adversarial Robustness of Deep State Space Models

Biqing Qi
Yiang Luo
Junqi Gao
Pengfei Li
Kai Tian
Zhiyuan Ma
Bowen Zhou

Deep State Space Models (SSMs) have proven effective in numerous task scenarios but face significant security challenges due to Adversarial Perturbations (APs) in real-world deployments. Adversarial Training (AT) is a mainstream approach to enhancing Adversarial Robustness (AR) and has been validated on various traditional DNN architectures. However, its effectiveness in improving the AR of SSMs remains unclear. While many enhancements in SSM components, such as integrating Attention mechanisms and expanding to data-dependent SSM parameterizations, have brought significant gains in Standard Training (ST) settings, their potential benefits in AT remain unexplored. To investigate this, we evaluate existing structural variants of SSMs with AT to assess their AR performance. We observe that pure SSM structures struggle to benefit from AT, whereas incorporating Attention yields a markedly better trade-off between robustness and generalization for SSMs in AT compared to other components. Nonetheless, the integration of Attention also leads to Robust Overfitting (RO) issues. To understand these phenomena, we empirically and theoretically analyze the output error of SSMs under AP. We find that fixed-parameterized SSMs have output error bounds strictly related to their parameters, limiting their AT benefits, while input-dependent SSMs may face the problem of error explosion. Furthermore, we show that the Attention component effectively scales the output error of SSMs during training, enabling them to benefit more from AT, but at the cost of introducing RO due to its high model complexity. Inspired by this, we propose a simple and effective Adaptive Scaling (AdS) mechanism that brings AT performance close to Attention-integrated SSMs without introducing the issue of RO.

PDF Details DOI

AAAI Conference 2024 Conference Paper

ModWaveMLP: MLP-Based Mode Decomposition and Wavelet Denoising Model to Defeat Complex Structures in Traffic Forecasting

Ke Sun
Pei Liu
Pengfei Li
Zhifang Liao

Traffic prediction is the core issue of Intelligent Transportation Systems. Recently, researchers have tended to use complex structures, such as transformer-based structures, for tasks such as traffic prediction. Notably, traffic data is simpler to process compared to text and images, which raises questions about the necessity of these structures. Additionally, when handling traffic data, researchers tend to manually design the model structure based on the data features, which makes the structure of traffic prediction redundant and the model generalizability limited. To address the above, we introduce the ‘ModWaveMLP’—A multilayer perceptron (MLP) based model designed according to mode decomposition and wavelet noise reduction information learning concepts. The model is based on simple MLP structure, which achieves the separation and prediction of different traffic modes and does not depend on additional features introduced such as the topology of the traffic network. By performing experiments on real-world datasets METR-LA and PEMS-BAY, our model achieves SOTA, outperforms GNN and transformer-based models, and outperforms those that introduce additional feature data with better generalizability, and we further demonstrate the effectiveness of the various parts of the model through ablation experiments. This offers new insights to subsequent researchers involved in traffic model design. The code is available at: https://github.com/Kqingzheng/ModWaveMLP.

PDF Details DOI

NeurIPS Conference 2024 Conference Paper

Online Budgeted Matching with General Bids

Jianyi Yang
Pengfei Li
Adam Wierman
Shaolei Ren

Online Budgeted Matching (OBM) is a classic problem with important applications in online advertising, online service matching, revenue management, and beyond. Traditional online algorithms typically assume a small bid setting, where the maximum bid-to-budget ratio ($\kappa$) is infinitesimally small. While recent algorithms have tried to address scenarios with non-small or general bids, they often rely on the Fractional Last Matching (FLM) assumption, which allows for accepting partial bids when the remaining budget is insufficient. This assumption, however, does not hold for many applications with indivisible bids. In this paper, we remove the FLM assumption and tackle the open problem of OBM with general bids. We first establish an upper bound of $1-\kappa$ on the competitive ratio for any deterministic online algorithm. We then propose a novel meta algorithm, called MetaAd, which reduces to different algorithms with first known provable competitive ratios parameterized by the maximum bid-to-budget ratio $\kappa\in [0, 1]$. As a by-product, we extend MetaAd to the FLM setting and get provable competitive algorithms. Finally, we apply our competitive analysis to the design learning- augmented algorithms.

PDF Details DOI

EAAI Journal 2024 Journal Article

Signal detection and material identification method for loose particles inside sealed relays based on fusion classification model

Zhigang Sun
Guotao Wang
Guofu Zhai
Pengfei Li
Qi Liang
Min Zhang

In this study, the authors combined the research on loose particle signal and component signal identification with the research on loose particle material identification for the first time, providing comprehensive and reliable loose particle detection results. Based on this, a signal detection and material identification method for loose particles inside sealed relays based on fusion classification model is proposed. Due to the limitations of technical means and confidentiality management, the authors made a real sealed relay sample, and took it as the research object. Through the steps of data acquisition, signal processing, feature engineering, and model training, the dedicated component identification feature library and material identification feature library was constructed, respectively, the component identification model and material identification model based on parameter-optimized SVM with linear kernel and XGBoost was trained, respectively. For the seal relay to be tested, through the steps of data acquisition, signal processing and feature engineering, the data set to be tested was created. The component identification model was used to identify component signals with loose particle signals, and the material identification model was used to identify the materials of loose particles. The majority voting process was used to convert the classification results into identification results, resulting in loose particle detection and material identification results. In addition, the general procedure steps of the proposed method for physical testing were given, and the identification accuracy for device-level loose particle detection was newly proposed. The loose particle testing event containing thirty-seven identification tasks shows that, the achieved identification accuracy was 97. 30%, and 92. 16% of the average classification accuracy was achieved by the component identification model, 80. 41% of the average classification accuracy was achieved by the material identification model. This effectively demonstrates the feasibility and practicality of the proposed method in this paper. It is an important supplement to the loose particle detection research, and provides references for signal detection in similar fields.

Details DOI

NeurIPS Conference 2023 Conference Paper

Anytime-Competitive Reinforcement Learning with Policy Prior

Jianyi Yang
Pengfei Li
Tongxin Li
Adam Wierman
Shaolei Ren

This paper studies the problem of Anytime-Competitive Markov Decision Process (A-CMDP). Existing works on Constrained Markov Decision Processes (CMDPs) aim to optimize the expected reward while constraining the expected cost over random dynamics, but the cost in a specific episode can still be unsatisfactorily high. In contrast, the goal of A-CMDP is to optimize the expected reward while guaranteeing a bounded cost in each round of any episode against a policy prior. We propose a new algorithm, called Anytime-Competitive Reinforcement Learning (ACRL), which provably guarantees the anytime cost constraints. The regret analysis shows the policy asymptotically matches the optimal reward achievable under the anytime competitive constraints. Experiments on the application of carbon-intelligent computing verify the reward performance and cost constraint guarantee of ACRL.

PDF Details

EAAI Journal 2023 Journal Article

Method of Locating Loose Particles Inside Aerospace Equipment Based on Parameter-optimized XGBoost

Zhigang Sun
Guotao Wang
Guofu Zhai
Pengfei Li
Qi Liang
Min Zhang

The complex internal structure and heterogeneous composition material of aerospace equipment make it difficult to apply traditional acoustic emission source localization methods to the loose particle localization research. In previous studies, the authors transformed the loose particle localization problem into the multi-classification problem in machine learning, trained the loose particle localization model, but there were problems of low classification accuracy and low practicality. In this paper, the authors first introduced the ensemble learning idea into the loose particle detection field, designed a complete loose particle localization scheme, analyzed the above problems from multiple perspectives, and proposed new algorithms or strategies to enhance the superiority and practicability of the loose particle localization model. Specifically, in view of the low classification accuracy, the authors carried out the research on pulse preprocessing, feature engineering and model training from the four perspectives of signal, feature, data set and classifier, respectively. The zero-pulse-filling pulse matching algorithm and the channel-weighting-based feature selection method was newly proposed, Mel-Frequency Cepstral Coefficients features was newly extracted, feature optimization scheme was designed, and XGBoost ensemble classifier was trained. Thus, high-quality loose particle signals, high-quality localization data sets and high-performance loose particle localization models were obtained, respectively. Test results show that, the classification accuracy achieved by the new loose particle localization model was 96. 80%, which was a significant improvement compared to the 83. 53% achieved in previous studies. In view of the low practicality, the authors built the loose particle localization experimental system, gave the specific implementation steps of the loose particle localization method, as well as the general procedures for applying the loose particle localization model for physical testing. Meanwhile, taking into account the requirements of aerospace engineering applications, the authors added the majority voting strategy to convert classification results into localization results, thus newly proposed the definition of equipment-level loose particle localization accuracy. Several physical testing results show that, the loose particle localization model achieved a localization accuracy of 90. 91%, which effectively verified the feasibility and stability of the proposed method. This study is an important supplement to the loose particle detection research, and of great significance in improving the reliability of aerospace systems.

Details DOI

NeurIPS Conference 2023 Conference Paper

Robust Learning for Smoothed Online Convex Optimization with Feedback Delay

Pengfei Li
Jianyi Yang
Adam Wierman
Shaolei Ren

We study a general form of Smoothed Online Convex Optimization, a. k. a. SOCO, including multi-step switching costs and feedback delay. We propose a novel machine learning (ML) augmented online algorithm, Robustness-Constrained Learning (RCL), which combines untrusted ML predictions with a trusted expert online algorithm via constrained projection to robustify the ML prediction. Specifically, we prove that RCL is able to guarantee $(1+\lambda)$-competitiveness against any given expert for any $\lambda>0$, while also explicitly training the ML model in a robustification-aware manner to improve the average-case performance. Importantly, RCL is the first ML-augmented algorithm with a provable robustness guarantee in the case of multi-step switching cost and feedback delay. We demonstrate the improvement of RCL in both robustness and average performance using battery management as a case study.

PDF Details

NeurIPS Conference 2022 Conference Paper

Constrained Update Projection Approach to Safe Policy Optimization

Long Yang
Jiaming Ji
Juntao Dai
Linrui Zhang
Binbin Zhou
Pengfei Li
Yaodong Yang
Gang Pan

Safe reinforcement learning (RL) studies problems where an intelligent agent has to not only maximize reward but also avoid exploring unsafe areas. In this study, we propose CUP, a novel policy optimization method based on Constrained Update Projection framework that enjoys rigorous safety guarantee. Central to our CUP development is the newly proposed surrogate functions along with the performance bound. Compared to previous safe reinforcement learning meth- ods, CUP enjoys the benefits of 1) CUP generalizes the surrogate functions to generalized advantage estimator (GAE), leading to strong empirical performance. 2) CUP unifies performance bounds, providing a better understanding and in- terpretability for some existing algorithms; 3) CUP provides a non-convex im- plementation via only first-order optimizers, which does not require any strong approximation on the convexity of the objectives. To validate our CUP method, we compared CUP against a comprehensive list of safe RL baselines on a wide range of tasks. Experiments show the effectiveness of CUP both in terms of reward and safety constraint satisfaction. We have opened the source code of CUP at https: //github. com/zmsn-2077/CUP-safe-rl.

PDF Details

IJCAI Conference 2022 Conference Paper

Graph Masked Autoencoder Enhanced Predictor for Neural Architecture Search

Kun Jing
Jungang Xu
Pengfei Li

Performance estimation of neural architecture is a crucial component of neural architecture search (NAS). Meanwhile, neural predictor is a current mainstream performance estimation method. However, it is a challenging task to train the predictor with few architecture evaluations for efficient NAS. In this paper, we propose a graph masked autoencoder (GMAE) enhanced predictor, which can reduce the dependence on supervision data by self-supervised pre-training with untrained architectures. We compare our GMAE-enhanced predictor with existing predictors in different search spaces, and experimental results show that our predictor has high query utilization. Moreover, GMAE-enhanced predictor with different search strategies can discover competitive architectures in different search spaces. Code and supplementary materials are available at https: //github. com/kunjing96/GMAENAS. git.

PDF Details DOI

AAAI Conference 2022 Conference Paper

Policy Optimization with Stochastic Mirror Descent

Long Yang
Yu Zhang
Gang Zheng
Qian Zheng
Pengfei Li
Jianhang Huang
Gang Pan

Improving sample efficiency has been a longstanding goal in reinforcement learning. This paper proposes VRMPO algorithm: a sample efficient policy gradient method with stochastic mirror descent. In VRMPO, a novel variance-reduced policy gradient estimator is presented to improve sample efficiency. We prove that the proposed VRMPO needs only O( −3 ) sample trajectories to achieve an -approximate first-order stationary point, which matches the best sample complexity for policy optimization. Extensive empirical results demonstrate that VRMPO outperforms the state-of-the-art policy gradient methods in various settings.

PDF Details

NeurIPS Conference 2022 Conference Paper

TOIST: Task Oriented Instance Segmentation Transformer with Noun-Pronoun Distillation

Pengfei Li
Beiwen Tian
Yongliang Shi
Xiaoxue Chen
Hao Zhao
Guyue Zhou
Ya-Qin Zhang

Current referring expression comprehension algorithms can effectively detect or segment objects indicated by nouns, but how to understand verb reference is still under-explored. As such, we study the challenging problem of task oriented detection, which aims to find objects that best afford an action indicated by verbs like sit comfortably on. Towards a finer localization that better serves downstream applications like robot interaction, we extend the problem into task oriented instance segmentation. A unique requirement of this task is to select preferred candidates among possible alternatives. Thus we resort to the transformer architecture which naturally models pair-wise query relationships with attention, leading to the TOIST method. In order to leverage pre-trained noun referring expression comprehension models and the fact that we can access privileged noun ground truth during training, a novel noun-pronoun distillation framework is proposed. Noun prototypes are generated in an unsupervised manner and contextual pronoun features are trained to select prototypes. As such, the network remains noun-agnostic during inference. We evaluate TOIST on the large-scale task oriented dataset COCO-Tasks and achieve +10. 7% higher $\rm{mAP^{box}}$ than the best-reported results. The proposed noun-pronoun distillation can boost $\rm{mAP^{box}}$ and $\rm{mAP^{mask}}$ by +2. 6% and +3. 6%. Codes and models are publicly available.

PDF Details

AAAI Conference 2021 Conference Paper

ACT: an Attentive Convolutional Transformer for Efficient Text Classification

Pengfei Li
Peixiang Zhong
Kezhi Mao
Dongzhe Wang
Xuefeng Yang
Yunfeng Liu
Jianxiong Yin
Simon See

Recently, Transformer has been demonstrating promising performance in many NLP tasks and showing a trend of replacing Recurrent Neural Network (RNN). Meanwhile, less attention is drawn to Convolutional Neural Network (CNN) due to its weak ability in capturing sequential and longdistance dependencies, although it has excellent local feature extraction capability. In this paper, we introduce an Attentive Convolutional Transformer (ACT) that takes the advantages of both Transformer and CNN for efficient text classification. Specifically, we propose a novel attentive convolution mechanism that utilizes the semantic meaning of convolutional filters attentively to transform text from complex word space to a more informative convolutional filter space where important n-grams are captured. ACT is able to capture both local and global dependencies effectively while preserving sequential information. Experiments on various text classification tasks and detailed analyses show that ACT is a lightweight, fast, and effective universal text classifier, outperforming CNNs, RNNs, and attentive models including Transformer.

PDF Details

AAAI Conference 2021 Conference Paper

CARE: Commonsense-Aware Emotional Response Generation with Latent Concepts

Peixiang Zhong
Di Wang
Pengfei Li
Chen Zhang
Hao Wang
Chunyan Miao

Rationality and emotion are two fundamental elements of humans. Endowing agents with rationality and emotion has been one of the major milestones in AI. However, in the field of conversational AI, most existing models only specialize in one aspect and neglect the other, which often leads to dull or unrelated responses. In this paper, we hypothesize that combining rationality and emotion into conversational agents can improve response quality. To test the hypothesis, we focus on one fundamental aspect of rationality, i. e. , commonsense, and propose CARE, a novel model for commonsense-aware emotional response generation. Specifically, we first propose a framework to learn and construct commonsense-aware emotional latent concepts of the response given an input message and a desired emotion. We then propose three methods to collaboratively incorporate the latent concepts into response generation. Experimental results on two large-scale datasets support our hypothesis and show that our model can produce more accurate and commonsense-aware emotional responses and achieve better human ratings than state-of-the-art models that only specialize in one aspect.

PDF Details

AAAI Conference 2021 Conference Paper

On Convergence of Gradient Expected Sarsa(λ)

Long Yang
Gang Zheng
Yu Zhang
Qian Zheng
Pengfei Li
Gang Pan

We study the convergence of Expected Sarsa(λ) with function approximation. We show that with off-line estimate (multi-step bootstrapping) to Expected Sarsa(λ) is unstable for off-policy learning. Furthermore, based on convex-concave saddle-point framework, we propose a convergent Gradient Expected Sarsa(λ) (GES(λ)) algorithm. The theoretical analysis shows that the proposed GES(λ) converges to the optimal solution at a linear convergence rate under true gradient setting. Furthermore, we develop a Lyapunov function technique to investigate how the stepsize influences finite-time performance of GES(λ). Additionally, such a technique of Lyapunov function can be potentially generalized to other gradient temporal difference algorithms. Finally, our experiments verify the effectiveness of our GES(λ). For the details of proof, please refer to https: //arxiv. org/pdf/2012. 07199. pdf.

PDF Details

IJCAI Conference 2020 Conference Paper

DIDFuse: Deep Image Decomposition for Infrared and Visible Image Fusion

Zixiang Zhao
Shuang Xu
Chunxia Zhang
Junmin Liu
Jiangshe Zhang
Pengfei Li

Infrared and visible image fusion, a hot topic in the field of image processing, aims at obtaining fused images keeping the advantages of source images. This paper proposes a novel auto-encoder (AE) based fusion network. The core idea is that the encoder decomposes an image into background and detail feature maps with low- and high-frequency information, respectively, and that the decoder recovers the original image. To this end, the loss function makes the background/detail feature maps of source images similar/dissimilar. In the test phase, background and detail feature maps are respectively merged via a fusion module, and the fused image is recovered by the decoder. Qualitative and quantitative results illustrate that our method can generate fusion images containing highlighted targets and abundant detail texture information with strong reproducibility and meanwhile surpass state-of-the-art (SOTA) approaches.

PDF Details DOI