Guanjie Zheng

Zhiyin Yu, Yuchen Mou, Juncheng Yan et al. · pku

Reinforcement learning (RL) has emerged as a powerful post-training paradigm for enhancing the reasoning capabilities of large language models (LLMs). However, reinforcement learning for LLMs faces substantial data scarcity challenges, including the limited availability of high-quality external supervision and the constrained volume of model-generated experience. These limitations make data-efficient reinforcement learning a critical research direction. In this survey, we present the first systematic review of reinforcement learning for LLMs under data scarcity. We propose a bottom-up hierarchical framework built around three complementary perspectives: the data-centric perspective, the training-centric perspective, and the framework-centric perspective. We develop a taxonomy of existing methods, summarize representative approaches in each category, and analyze their strengths and limitations. Our taxonomy aims to provide a clear conceptual foundation for understanding the design space of data-efficient RL for LLMs and to guide researchers working in this emerging area. We hope this survey offers a comprehensive roadmap for future research and inspires new directions toward more efficient and scalable reinforcement learning post-training for LLMs.

Hao Mei, Junxian Li, Zhiming Liang et al.

Traffic prediction is a crucial topic because of its broad scope of applications in the transportation domain. Recently, various studies have achieved promising results. However, most studies assume the prediction locations have complete or at least partial historical records and cannot be extended to non-historical recorded locations. In real-life scenarios, the deployment of sensors could be limited due to budget limitations and installation availability, which makes most current models not applicable. Though few pieces of literature tried to impute traffic states at the missing locations, these methods need the data simultaneously observed at the locations with sensors, making them not applicable to prediction tasks. Another drawback is the lack of measurement of uncertainty in prediction, making prior works unsuitable for risk-sensitive tasks or involving decision-making. To fill the gap, inspired by the previous inductive graph neural network, this work proposed an uncertainty-aware framework with the ability to 1) extend prediction to missing locations with no historical records and significantly extend spatial coverage of prediction locations while reducing deployment of sensors and 2) generate probabilistic prediction with uncertainty quantification to help the management of risk and decision making in the down-stream tasks. Through extensive experiments on real-life datasets, the result shows our method achieved promising results on prediction tasks, and the uncertainty quantification gives consistent results which highly correlated with the locations with and without historical data. We also show that our model could help support sensor deployment tasks in the transportation field to achieve higher accuracy with a limited sensor deployment budget.

Yude Zou, Junji Gong, Xing Gao et al.

Human-object-scene interactions (HOSI) generation has broad applications in embodied AI, simulation, and animation. Unlike human-object interaction (HOI) and human-scene interaction (HSI), HOSI generation requires reasoning over dynamic object-scene changes, yet suffers from limited annotated data. To address these issues, we propose a coarse-to-fine instruction-conditioned interaction generation framework that is explicitly aligned with the iterative denoising process of a consistency model. In particular, we adopt a dynamic perception strategy that leverages trajectories from the preceding refinement to update scene context and condition subsequent refinement at each denoising step of consistency model, yielding consistent interactions. To further reduce physical artifacts, we introduce a bump-aware guidance that mitigates collisions and penetrations during sampling without requiring fine-grained scene geometry, enabling real-time generation. To overcome data scarcity, we design a hybrid training startegy that synthesizes pseudo-HOSI samples by injecting voxelized scene occupancy into HOI datasets and jointly trains with high-fidelity HSI data, allowing interaction learning while preserving realistic scene awareness. Extensive experiments demonstrate that our method achieves state-of-the-art performance in both HOSI and HOI generation, and strong generalization to unseen scenes. Project page: https://yudezou.github.io/InfBaGel-page/

Zhanyu Liu, Guanjie Zheng, Yanwei Yu

Traffic forecasting is crucial for intelligent transportation systems (ITS), aiding in efficient resource allocation and effective traffic control. However, its effectiveness often relies heavily on abundant traffic data, while many cities lack sufficient data due to limited device support, posing a significant challenge for traffic forecasting. Recognizing this challenge, we have made a noteworthy observation: traffic patterns exhibit similarities across diverse cities. Building on this key insight, we propose a solution for the cross-city few-shot traffic forecasting problem called Multi-scale Traffic Pattern Bank (MTPB). Primarily, MTPB initiates its learning process by leveraging data-rich source cities, effectively acquiring comprehensive traffic knowledge through a spatial-temporal-aware pre-training process. Subsequently, the framework employs advanced clustering techniques to systematically generate a multi-scale traffic pattern bank derived from the learned knowledge. Next, the traffic data of the data-scarce target city could query the traffic pattern bank, facilitating the aggregation of meta-knowledge. This meta-knowledge, in turn, assumes a pivotal role as a robust guide in subsequent processes involving graph reconstruction and forecasting. Empirical assessments conducted on real-world traffic datasets affirm the superior performance of MTPB, surpassing existing methods across various categories and exhibiting numerous attributes conducive to the advancement of cross-city few-shot forecasting methodologies. The code is available in https://github.com/zhyliu00/MTPB.

Yuhao Zhou, Yiheng Wang, Xuming He et al.

Scientific discoveries increasingly rely on complex multimodal reasoning based on information-intensive scientific data and domain-specific expertise. Empowered by expert-level scientific benchmarks, scientific Multimodal Large Language Models (MLLMs) hold the potential to significantly enhance this discovery process in realistic workflows. However, current scientific benchmarks mostly focus on evaluating the knowledge understanding capabilities of MLLMs, leading to an inadequate assessment of their perception and reasoning abilities. To address this gap, we present the Scientists' First Exam (SFE) benchmark, designed to evaluate the scientific cognitive capacities of MLLMs through three interconnected levels: scientific signal perception, scientific attribute understanding, scientific comparative reasoning. Specifically, SFE comprises 830 expert-verified VQA pairs across three question types, spanning 66 multimodal tasks across five high-value disciplines. Extensive experiments reveal that current state-of-the-art GPT-o3 and InternVL-3 achieve only 34.08% and 26.52% on SFE, highlighting significant room for MLLMs to improve in scientific realms. We hope the insights obtained in SFE will facilitate further developments in AI-enhanced scientific discoveries.

Zhanyu Liu, Ke Hao, Guanjie Zheng et al.

Time series data has been demonstrated to be crucial in various research fields. The management of large quantities of time series data presents challenges in terms of deep learning tasks, particularly for training a deep neural network. Recently, a technique named \textit{Dataset Condensation} has emerged as a solution to this problem. This technique generates a smaller synthetic dataset that has comparable performance to the full real dataset in downstream tasks such as classification. However, previous methods are primarily designed for image and graph datasets, and directly adapting them to the time series dataset leads to suboptimal performance due to their inability to effectively leverage the rich information inherent in time series data, particularly in the frequency domain. In this paper, we propose a novel framework named Dataset \textit{\textbf{Cond}}ensation for \textit{\textbf{T}}ime \textit{\textbf{S}}eries \textit{\textbf{C}}lassification via Dual Domain Matching (\textbf{CondTSC}) which focuses on the time series classification dataset condensation task. Different from previous methods, our proposed framework aims to generate a condensed dataset that matches the surrogate objectives in both the time and frequency domains. Specifically, CondTSC incorporates multi-view data augmentation, dual domain training, and dual surrogate objectives to enhance the dataset condensation process in the time and frequency domains. Through extensive experiments, we demonstrate the effectiveness of our proposed framework, which outperforms other baselines and learns a condensed synthetic dataset that exhibits desirable characteristics such as conforming to the distribution of the original data.

Xiang Li, Jianpeng Qi, Zhongying Zhao et al.

Graph anomaly detection (GAD) is a critical task in graph machine learning, with the primary objective of identifying anomalous nodes that deviate significantly from the majority. This task is widely applied in various real-world scenarios, including fraud detection and social network analysis. However, existing GAD methods still face two major challenges: (1) They are often limited to detecting anomalies in single-type interaction graphs and struggle with multiple interaction types in multiplex heterogeneous graphs. (2) In unsupervised scenarios, selecting appropriate anomaly score thresholds remains a significant challenge for accurate anomaly detection. To address the above challenges, we propose a novel Unsupervised Multiplex Graph Anomaly Detection method, named UMGAD. We first learn multi-relational correlations among nodes in multiplex heterogeneous graphs and capture anomaly information during node attribute and structure reconstruction through graph-masked autoencoder (GMAE). Then, to further extract abnormal information, we generate attribute-level and subgraph-level augmented-view graphs, respectively, and perform attribute and structure reconstruction through GMAE. Finally, we learn to optimize node attributes and structural features through contrastive learning between original-view and augmented-view graphs to improve the model's ability to capture anomalies. Meanwhile, we propose a new anomaly score threshold selection strategy, which allows the model to be independent of ground truth information in real unsupervised scenarios. Extensive experiments on six datasets show that our UMGAD significantly outperforms state-of-the-art methods, achieving average improvements of 12.25% in AUC and 11.29% in Macro-F1 across all datasets.

Chaolv Zeng, Zhanyu Liu, Guanjie Zheng et al.

Recent advancements in multivariate time series forecasting have been propelled by Linear-based, Transformer-based, and Convolution-based models, with Transformer-based architectures gaining prominence for their efficacy in temporal and cross-channel mixing. More recently, Mamba, a state space model, has emerged with robust sequence and feature mixing capabilities. However, the suitability of the vanilla Mamba design for time series forecasting remains an open question, particularly due to its inadequate handling of cross-channel dependencies. Capturing cross-channel dependencies is critical in enhancing the performance of multivariate time series prediction. Recent findings show that self-attention excels in capturing cross-channel dependencies, whereas other simpler mechanisms, such as MLP, may degrade model performance. This is counterintuitive, as MLP, being a learnable architecture, should theoretically capture both correlations and irrelevances, potentially leading to neutral or improved performance. Diving into the self-attention mechanism, we attribute the observed degradation in MLP performance to its lack of data dependence and global receptive field, which result in MLP's lack of generalization ability. Based on the above insights, we introduce a refined Mamba variant tailored for time series forecasting. Our proposed model, \textbf{CMamba}, incorporates a modified Mamba (M-Mamba) module for temporal dependencies modeling, a global data-dependent MLP (GDD-MLP) to effectively capture cross-channel dependencies, and a Channel Mixup mechanism to mitigate overfitting. Comprehensive experiments conducted on seven real-world datasets demonstrate the efficacy of our model in improving forecasting performance.

Jianrong Ding, Zhanyu Liu, Guanjie Zheng et al.

Dataset condensation is a newborn technique that generates a small dataset that can be used in training deep neural networks to lower training costs. The objective of dataset condensation is to ensure that the model trained with the synthetic dataset can perform comparably to the model trained with full datasets. However, existing methods predominantly concentrate on classification tasks, posing challenges in their adaptation to time series forecasting (TS-forecasting). This challenge arises from disparities in the evaluation of synthetic data. In classification, the synthetic data is considered well-distilled if the model trained with the full dataset and the model trained with the synthetic dataset yield identical labels for the same input, regardless of variations in output logits distribution. Conversely, in TS-forecasting, the effectiveness of synthetic data distillation is determined by the distance between predictions of the two models. The synthetic data is deemed well-distilled only when all data points within the predictions are similar. Consequently, TS-forecasting has a more rigorous evaluation methodology compared to classification. To mitigate this gap, we theoretically analyze the optimization objective of dataset condensation for TS-forecasting and propose a new one-line plugin of dataset condensation designated as Dataset Condensation for Time Series Forecasting (CondTSF) based on our analysis. Plugging CondTSF into previous dataset condensation methods facilitates a reduction in the distance between the predictions of the model trained with the full dataset and the model trained with the synthetic dataset, thereby enhancing performance. We conduct extensive experiments on eight commonly used time series datasets. CondTSF consistently improves the performance of all previous dataset condensation methods across all datasets, particularly at low condensing ratios.

Zhanyu Liu, Jianrong Ding, Guanjie Zheng

The field of Intelligent Transportation Systems (ITS) relies on accurate traffic forecasting to enable various downstream applications. However, developing cities often face challenges in collecting sufficient training traffic data due to limited resources and outdated infrastructure. Recognizing this obstacle, the concept of cross-city few-shot forecasting has emerged as a viable approach. While previous cross-city few-shot forecasting methods ignore the frequency similarity between cities, we have made an observation that the traffic data is more similar in the frequency domain between cities. Based on this fact, we propose a \textbf{F}requency \textbf{E}nhanced \textbf{P}re-training Framework for \textbf{Cross}-city Few-shot Forecasting (\textbf{FEPCross}). FEPCross has a pre-training stage and a fine-tuning stage. In the pre-training stage, we propose a novel Cross-Domain Spatial-Temporal Encoder that incorporates the information of the time and frequency domain and trains it with self-supervised tasks encompassing reconstruction and contrastive objectives. In the fine-tuning stage, we design modules to enrich training samples and maintain a momentum-updated graph structure, thereby mitigating the risk of overfitting to the few-shot training data. Empirical evaluations performed on real-world traffic datasets validate the exceptional efficacy of FEPCross, outperforming existing approaches of diverse categories and demonstrating characteristics that foster the progress of cross-city few-shot forecasting.

Muchao Ye, Junyu Luo, Guanjie Zheng et al.

Deep neural networks (DNNs) have been broadly adopted in health risk prediction to provide healthcare diagnoses and treatments. To evaluate their robustness, existing research conducts adversarial attacks in the white/gray-box setting where model parameters are accessible. However, a more realistic black-box adversarial attack is ignored even though most real-world models are trained with private data and released as black-box services on the cloud. To fill this gap, we propose the first black-box adversarial attack method against health risk prediction models named MedAttacker to investigate their vulnerability. MedAttacker addresses the challenges brought by EHR data via two steps: hierarchical position selection which selects the attacked positions in a reinforcement learning (RL) framework and substitute selection which identifies substitute with a score-based principle. Particularly, by considering the temporal context inside EHRs, it initializes its RL position selection policy by using the contribution score of each visit and the saliency score of each code, which can be well integrated with the deterministic substitute selection process decided by the score changes. In experiments, MedAttacker consistently achieves the highest average success rate and even outperforms a recent white-box EHR adversarial attack technique in certain cases when attacking three advanced health risk prediction models in the black-box setting across multiple real-world datasets. In addition, based on the experiment results we include a discussion on defending EHR adversarial attacks.

Guanjie Zheng, Hanyang Liu, Kai Xu et al.

Traffic simulators act as an essential component in the operating and planning of transportation systems. Conventional traffic simulators usually employ a calibrated physical car-following model to describe vehicles' behaviors and their interactions with traffic environment. However, there is no universal physical model that can accurately predict the pattern of vehicle's behaviors in different situations. A fixed physical model tends to be less effective in a complicated environment given the non-stationary nature of traffic dynamics. In this paper, we formulate traffic simulation as an inverse reinforcement learning problem, and propose a parameter sharing adversarial inverse reinforcement learning model for dynamics-robust simulation learning. Our proposed model is able to imitate a vehicle's trajectories in the real world while simultaneously recovering the reward function that reveals the vehicle's true objective which is invariant to different dynamics. Extensive experiments on synthetic and real-world datasets show the superior performance of our approach compared to state-of-the-art methods and its robustness to variant dynamics of traffic.

Chang Liu, Guanjie Zheng, Zhenhui Li

The heavy traffic and related issues have always been concerns for modern cities. With the help of deep learning and reinforcement learning, people have proposed various policies to solve these traffic-related problems, such as smart traffic signal control systems and taxi dispatching systems. People usually validate these policies in a city simulator, since directly applying them in the real city introduces real cost. However, these policies validated in the city simulator may fail in the real city if the simulator is significantly different from the real world. To tackle this problem, we need to build a real-like traffic simulation system. Therefore, in this paper, we propose to learn the human routing model, which is one of the most essential part in the traffic simulator. This problem has two major challenges. First, human routing decisions are determined by multiple factors, besides the common time and distance factor. Second, current historical routes data usually covers just a small portion of vehicles, due to privacy and device availability issues. To address these problems, we propose a theory-guided residual network model, where the theoretical part can emphasize the general principles for human routing decisions (e.g., fastest route), and the residual part can capture drivable condition preferences (e.g., local road or highway). Since the theoretical part is composed of traditional shortest path algorithms that do not need data to train, our residual network can learn human routing models from limited data. We have conducted extensive experiments on multiple real-world datasets to show the superior performance of our model, especially with small data. Besides, we have also illustrated why our model is better at recovering real routes through case studies.

Hua Wei, Chacha Chen, Chang Liu et al.

Simulation of the real-world traffic can be used to help validate the transportation policies. A good simulator means the simulated traffic is similar to real-world traffic, which often requires dense traffic trajectories (i.e., with a high sampling rate) to cover dynamic situations in the real world. However, in most cases, the real-world trajectories are sparse, which makes simulation challenging. In this paper, we present a novel framework ImInGAIL to address the problem of learning to simulate the driving behavior from sparse real-world data. The proposed architecture incorporates data interpolation with the behavior learning process of imitation learning. To the best of our knowledge, we are the first to tackle the data sparsity issue for behavior learning problems. We investigate our framework on both synthetic and real-world trajectory datasets of driving vehicles, showing that our method outperforms various baselines and state-of-the-art methods.

Chang Liu, Huichu Zhang, Weinan Zhang et al.

The heavy traffic congestion problem has always been a concern for modern cities. To alleviate traffic congestion, researchers use reinforcement learning (RL) to develop better traffic signal control (TSC) algorithms in recent years. However, most RL models are trained and tested in the same traffic flow environment, which results in a serious overfitting problem. Since the traffic flow environment in the real world keeps varying, these models can hardly be applied due to the lack of generalization ability. Besides, the limited number of accessible traffic flow data brings extra difficulty in testing the generalization ability of the models. In this paper, we design a novel traffic flow generator based on Wasserstein generative adversarial network to generate sufficient diverse and quality traffic flows and use them to build proper training and testing environments. Then we propose a meta-RL TSC framework GeneraLight to improve the generalization ability of TSC models. GeneraLight boosts the generalization performance by combining the idea of flow clustering and model-agnostic meta-learning. We conduct extensive experiments on multiple real-world datasets to show the superior performance of GeneraLight on generalizing to different traffic flows.

Guanjie Zheng, Mengqi Liu, Tao Wen et al.

In the face of growing needs for water and energy, a fundamental understanding of the environmental impacts of human activities becomes critical for managing water and energy resources, remedying water pollution, and making regulatory policy wisely. Among activities that impact the environment, oil and gas production, wastewater transport, and urbanization are included. In addition to the occurrence of anthropogenic contamination, the presence of some contaminants (e.g., methane, salt, and sulfate) of natural origin is not uncommon. Therefore, scientists sometimes find it difficult to identify the sources of contaminants in the coupled natural and human systems. In this paper, we propose a technique to simultaneously conduct source detection and prediction, which outperforms other approaches in the interdisciplinary case study of the identification of potential groundwater contamination within a region of high-density shale gas development.

Guanjie Zheng, Yuanhao Xiong, Xinshi Zang et al.

Increasingly available city data and advanced learning techniques have empowered people to improve the efficiency of our city functions. Among them, improving the urban transportation efficiency is one of the most prominent topics. Recent studies have proposed to use reinforcement learning (RL) for traffic signal control. Different from traditional transportation approaches which rely heavily on prior knowledge, RL can learn directly from the feedback. On the other side, without a careful model design, existing RL methods typically take a long time to converge and the learned models may not be able to adapt to new scenarios. For example, a model that is trained well for morning traffic may not work for the afternoon traffic because the traffic flow could be reversed, resulting in a very different state representation. In this paper, we propose a novel design called FRAP, which is based on the intuitive principle of phase competition in traffic signal control: when two traffic signals conflict, priority should be given to one with larger traffic movement (i.e., higher demand). Through the phase competition modeling, our model achieves invariance to symmetrical cases such as flipping and rotation in traffic flow. By conducting comprehensive experiments, we demonstrate that our model finds better solutions than existing RL methods in the complicated all-phase selection problem, converges much faster during training, and achieves superior generalizability for different road structures and traffic conditions.

Guanjie Zheng, Xinshi Zang, Nan Xu et al.

With the increasing availability of traffic data and advance of deep reinforcement learning techniques, there is an emerging trend of employing reinforcement learning (RL) for traffic signal control. A key question for applying RL to traffic signal control is how to define the reward and state. The ultimate objective in traffic signal control is to minimize the travel time, which is difficult to reach directly. Hence, existing studies often define reward as an ad-hoc weighted linear combination of several traffic measures. However, there is no guarantee that the travel time will be optimized with the reward. In addition, recent RL approaches use more complicated state (e.g., image) in order to describe the full traffic situation. However, none of the existing studies has discussed whether such a complex state representation is necessary. This extra complexity may lead to significantly slower learning process but may not necessarily bring significant performance gain. In this paper, we propose to re-examine the RL approaches through the lens of classic transportation theory. We ask the following questions: (1) How should we design the reward so that one can guarantee to minimize the travel time? (2) How to design a state representation which is concise yet sufficient to obtain the optimal solution? Our proposed method LIT is theoretically supported by the classic traffic signal control methods in transportation field. LIT has a very simple state and reward design, thus can serve as a building block for future RL approaches to traffic signal control. Extensive experiments on both synthetic and real datasets show that our method significantly outperforms the state-of-the-art traffic signal control methods.

Hua Wei, Guanjie Zheng, Vikash Gayah et al.

Traffic signal control is an important and challenging real-world problem, which aims to minimize the travel time of vehicles by coordinating their movements at the road intersections. Current traffic signal control systems in use still rely heavily on oversimplified information and rule-based methods, although we now have richer data, more computing power and advanced methods to drive the development of intelligent transportation. With the growing interest in intelligent transportation using machine learning methods like reinforcement learning, this survey covers the widely acknowledged transportation approaches and a comprehensive list of recent literature on reinforcement for traffic signal control. We hope this survey can foster interdisciplinary research on this important topic.

Huaxiu Yao, Xianfeng Tang, Hua Wei et al.

Traffic prediction has drawn increasing attention in AI research field due to the increasing availability of large-scale traffic data and its importance in the real world. For example, an accurate taxi demand prediction can assist taxi companies in pre-allocating taxis. The key challenge of traffic prediction lies in how to model the complex spatial dependencies and temporal dynamics. Although both factors have been considered in modeling, existing works make strong assumptions about spatial dependence and temporal dynamics, i.e., spatial dependence is stationary in time, and temporal dynamics is strictly periodical. However, in practice, the spatial dependence could be dynamic (i.e., changing from time to time), and the temporal dynamics could have some perturbation from one period to another period. In this paper, we make two important observations: (1) the spatial dependencies between locations are dynamic; and (2) the temporal dependency follows daily and weekly pattern but it is not strictly periodic for its dynamic temporal shifting. To address these two issues, we propose a novel Spatial-Temporal Dynamic Network (STDN), in which a flow gating mechanism is introduced to learn the dynamic similarity between locations, and a periodically shifted attention mechanism is designed to handle long-term periodic temporal shifting. To the best of our knowledge, this is the first work that tackles both issues in a unified framework. Our experimental results on real-world traffic datasets verify the effectiveness of the proposed method.