Na Zhang

Jiarong Ye, Yin-Ting Yeh, Yuan Xue et al.

Rapid identification of newly emerging or circulating viruses is an important first step toward managing the public health response to potential outbreaks. A portable virus capture device coupled with label-free Raman Spectroscopy holds the promise of fast detection by rapidly obtaining the Raman signature of a virus followed by a machine learning approach applied to recognize the virus based on its Raman spectrum, which is used as a fingerprint. We present such a machine learning approach for analyzing Raman spectra of human and avian viruses. A Convolutional Neural Network (CNN) classifier specifically designed for spectral data achieves very high accuracy for a variety of virus type or subtype identification tasks. In particular, it achieves 99% accuracy for classifying influenza virus type A vs. type B, 96% accuracy for classifying four subtypes of influenza A, 95% accuracy for differentiating enveloped and non-enveloped viruses, and 99% accuracy for differentiating avian coronavirus (infectious bronchitis virus, IBV) from other avian viruses. Furthermore, interpretation of neural net responses in the trained CNN model using a full-gradient algorithm highlights Raman spectral ranges that are most important to virus identification. By correlating ML-selected salient Raman ranges with the signature ranges of known biomolecules and chemical functional groups (for example, amide, amino acid, carboxylic acid), we verify that our ML model effectively recognizes the Raman signatures of proteins, lipids and other vital functional groups present in different viruses and uses a weighted combination of these signatures to identify viruses.

Yalchin Efendiev, Seong Lee, Guanglian Li et al.

In this paper, we develop a multiscale finite element method for solving flows in fractured media. Our approach is based on Generalized Multiscale Finite Element Method (GMsFEM), where we represent the fracture effects on a coarse grid via multiscale basis functions. These multiscale basis functions are constructed in the offline stage via local spectral problems following GMsFEM. To represent the fractures on the fine grid, we consider two approaches (1) Discrete Fracture Model (DFM) (2) Embedded Fracture Model (EFM) and their combination. In DFM, the fractures are resolved via the fine grid, while in EFM the fracture and the fine grid block interaction is represented as a source term. In the proposed multiscale method, additional multiscale basis functions are used to represent the long fractures, while short-size fractures are collectively represented by a single basis functions. The procedure is automatically done via local spectral problems. In this regard, our approach shares common concepts with several approaches proposed in the literature as we discuss. Numerical results are presented where we demonstrate how one can adaptively add basis functions in the regions of interest based on error indicators. We also discuss the use of randomized snapshots (\cite{randomized2014}) which reduces the offline computational cost.

Na Zhang, Shan Jia, Siwei Lyu et al.

The vulnerability of face recognition systems to morphing attacks has posed a serious security threat due to the wide adoption of face biometrics in the real world. Most existing morphing attack detection (MAD) methods require a large amount of training data and have only been tested on a few predefined attack models. The lack of good generalization properties, especially in view of the growing interest in developing novel morphing attacks, is a critical limitation with existing MAD research. To address this issue, we propose to extend MAD from supervised learning to few-shot learning and from binary detection to multiclass fingerprinting in this paper. Our technical contributions include: 1) We propose a fusion-based few-shot learning (FSL) method to learn discriminative features that can generalize to unseen morphing attack types from predefined presentation attacks; 2) The proposed FSL based on the fusion of the PRNU model and Noiseprint network is extended from binary MAD to multiclass morphing attack fingerprinting (MAF). 3) We have collected a large-scale database, which contains five face datasets and eight different morphing algorithms, to benchmark the proposed few-shot MAF (FS-MAF) method. Extensive experimental results show the outstanding performance of our fusion-based FS-MAF. The code and data will be publicly available at https://github.com/nz0001na/mad maf.

Na Zhang, Xudong Liu, Xin Li et al.

Semantic face image manipulation has received increasing attention in recent years. StyleGAN-based approaches to face morphing are among the leading techniques; however, they often suffer from noticeable blurring and artifacts as a result of the uniform attention in the latent feature space. In this paper, we propose to develop a transformer-based alternative to face morphing and demonstrate its superiority to StyleGAN-based methods. Our contributions are threefold. First, inspired by GANformer, we introduce a bipartite structure to exploit long-range interactions in face images for iterative propagation of information from latent variables to salient facial features. Special loss functions are designed to support the optimization of face morphing. Second, we extend the study of transformer-based face morphing to demorphing by presenting an effective defense strategy with access to a reference image using the same generator of MorphGANFormer. Such demorphing is conceptually similar to unmixing of hyperspectral images but operates in the latent (instead of pixel) space. Third, for the first time, we address a fundamental issue of vulnerability-detectability trade-off for face morphing studies. It is argued that neither doppelganger norrandom pair selection is optimal, and a Lagrangian multiplier-based approach should be used to achieve an improved trade-off between recognition vulnerability and attack detectability.

Zhengyong Huang, Sijuan Zou, Guoshuai Wang et al.

Achieving accurate and automated tumor segmentation plays an important role in both clinical practice and radiomics research. Segmentation in medicine is now often performed manually by experts, which is a laborious, expensive and error-prone task. Manual annotation relies heavily on the experience and knowledge of these experts. In addition, there is much intra- and interobserver variation. Therefore, it is of great significance to develop a method that can automatically segment tumor target regions. In this paper, we propose a deep learning segmentation method based on multimodal positron emission tomography-computed tomography (PET-CT), which combines the high sensitivity of PET and the precise anatomical information of CT. We design an improved spatial attention network(ISA-Net) to increase the accuracy of PET or CT in detecting tumors, which uses multi-scale convolution operation to extract feature information and can highlight the tumor region location information and suppress the non-tumor region location information. In addition, our network uses dual-channel inputs in the coding stage and fuses them in the decoding stage, which can take advantage of the differences and complementarities between PET and CT. We validated the proposed ISA-Net method on two clinical datasets, a soft tissue sarcoma(STS) and a head and neck tumor(HECKTOR) dataset, and compared with other attention methods for tumor segmentation. The DSC score of 0.8378 on STS dataset and 0.8076 on HECKTOR dataset show that ISA-Net method achieves better segmentation performance and has better generalization. Conclusions: The method proposed in this paper is based on multi-modal medical image tumor segmentation, which can effectively utilize the difference and complementarity of different modes. The method can also be applied to other multi-modal data or single-modal data by proper adjustment.

Shuyue Stella Li, Bhargavi Paranjape, Kerem Oktar et al.

User preferences evolve across months of interaction, and tracking them requires inferring when a stated preference has been changed by a subsequent life event. We define this problem as long-horizon personalization and observe that progress on it is limited by data availability and measurement, with no existing resource providing both naturalistic long-horizon interactions and the ground-truth provenance needed to diagnose why models fail. We introduce a data generator that produces conversations from a structured mental state graph, yielding ground-truth provenance for every preference change across 6-month timelines, and from it construct HorizonBench, a benchmark of 4,245 items from 360 simulated users with 6-month conversation histories averaging ~4,300 turns and ~163K tokens. HorizonBench provides a testbed for long-context modeling, memory-augmented architectures, theory-of-mind reasoning, and user modeling. Across 25 frontier models, the best model reaches 52.8% and most score at or below the 20% chance baseline. When these models err on evolved preferences, over a third of the time they select the user's originally stated value without tracking the updated user state. This belief-update failure persists across context lengths and expression explicitness levels, identifying state-tracking capability as the primary bottleneck for long-horizon personalization.

Mindi Ruan, Xiangxu Yu, Na Zhang et al.

How can we teach a computer to recognize 10,000 different actions? Deep learning has evolved from supervised and unsupervised to self-supervised approaches. In this paper, we present a new contrastive learning-based framework for decision tree-based classification of actions, including human-human interactions (HHI) and human-object interactions (HOI). The key idea is to translate the original multi-class action recognition into a series of binary classification tasks on a pre-constructed decision tree. Under the new framework of contrastive learning, we present the design of an interaction adjacent matrix (IAM) with skeleton graphs as the backbone for modeling various action-related attributes such as periodicity and symmetry. Through the construction of various pretext tasks, we obtain a series of binary classification nodes on the decision tree that can be combined to support higher-level recognition tasks. Experimental justification for the potential of our approach in real-world applications ranges from interaction recognition to symmetry detection. In particular, we have demonstrated the promising performance of video-based autism spectrum disorder (ASD) diagnosis on the CalTech interview video database.

Hui Zeng, Jingyuan Xue, Meng Hao et al.

This paper unveils CG-Eval, the first-ever comprehensive and automated evaluation framework designed for assessing the generative capabilities of large Chinese language models across a spectrum of academic disciplines. CG-Eval stands out for its automated process, which critically assesses models based on their proficiency in generating precise and contextually relevant responses to a diverse array of questions within six key domains: Science and Engineering, Humanities and Social Sciences, Mathematical Calculations, Medical Practitioner Qualification Examination, Judicial Examination, and Certified Public Accountant Examination. Alongside this, we introduce Gscore, an innovative composite index developed from a weighted sum of multiple metrics. Gscore uniquely automates the quality measurement of a model's text generation against reference standards, providing a detailed and nuanced assessment of model performance. This automation not only enhances the efficiency and scalability of the evaluation process but also ensures objective and consistent assessment across various models. The detailed test data and results, highlighting the robust capabilities and comparative performance of the evaluated models, are accessible at http://cgeval.besteasy.com/.

Zhenglin Huang, Xiaoan Bao, Na Zhang et al.

Data augmentation has been proven effective for training high-accuracy convolutional neural network classifiers by preventing overfitting. However, building deep neural networks in real-world scenarios requires not only high accuracy on clean data but also robustness when data distributions shift. While prior methods have proposed that there is a trade-off between accuracy and robustness, we propose IPMix, a simple data augmentation approach to improve robustness without hurting clean accuracy. IPMix integrates three levels of data augmentation (image-level, patch-level, and pixel-level) into a coherent and label-preserving technique to increase the diversity of training data with limited computational overhead. To further improve the robustness, IPMix introduces structural complexity at different levels to generate more diverse images and adopts the random mixing method for multi-scale information fusion. Experiments demonstrate that IPMix outperforms state-of-the-art corruption robustness on CIFAR-C and ImageNet-C. In addition, we show that IPMix also significantly improves the other safety measures, including robustness to adversarial perturbations, calibration, prediction consistency, and anomaly detection, achieving state-of-the-art or comparable results on several benchmarks, including ImageNet-R, ImageNet-A, and ImageNet-O.

Zhihan Liu, Lin Guan, Yixin Nie et al.

Generalist LLM agents are often post-trained on a narrow set of environments but deployed across far broader, unseen domains. In this work, we investigate the challenge of agentic post-training when the eventual test domains are unknown. Specifically, we analyze which properties of reinforcement learning (RL) environments and modeling choices have the greatest influence on out-of-domain performance. First, we identify two environment axes that strongly correlate with cross-domain generalization: (i) state information richness, i.e., the amount of information for the agent to process from the state, and (ii) planning complexity, estimated via goal reachability and trajectory length under a base policy. Notably, domain realism and text-level similarity are not the primary factors; for instance, the simple grid-world domain Sokoban leads to even stronger generalization in SciWorld than the more realistic ALFWorld. Motivated by these findings, we further show that increasing state information richness alone can already effectively improve cross-domain robustness. We propose a randomization technique, which is low-overhead and broadly applicable: add small amounts of distractive goal-irrelevant features to the state to make it richer without altering the task. Beyond environment-side properties, we also examine several modeling choices: (a) SFT warmup or mid-training helps prevent catastrophic forgetting during RL but undermines generalization to domains that are not included in the mid-training datamix; and (b) turning on step-by-step thinking during RL, while not always improving in-domain performance, plays a crucial role in preserving generalization.

Yixin Nie, Lin Guan, Zhongyao Ma et al.

This report presents CharacterFlywheel, an iterative flywheel process for improving large language models (LLMs) in production social chat applications across Instagram, WhatsApp, and Messenger. Starting from LLaMA 3.1, we refined models across 15 generations using data from both internal and external real-user traffic. Through continuous deployments from July 2024 to April 2025, we conducted controlled 7-day A/B tests showing consistent engagement improvements: 7 of 8 newly deployed models demonstrated positive lift over the baseline, with the strongest performers achieving up to 8.8% improvement in engagement breadth and 19.4% in engagement depth. We also observed substantial gains in steerability, with instruction following increasing from 59.2% to 84.8% and instruction violations decreasing from 26.6% to 5.8%. We detail the CharacterFlywheel process which integrates data curation, reward modeling to estimate and interpolate the landscape of engagement metrics, supervised fine-tuning (SFT), reinforcement learning (RL), and both offline and online evaluation to ensure reliable progress at each optimization step. We also discuss our methods for overfitting prevention and navigating production dynamics at scale. These contributions advance the scientific rigor and understanding of LLMs in social applications serving millions of users.

Rongzhe Wei, Ge Shi, Min Cheng et al.

Large Language Models (LLMs) have significantly advanced tool-augmented agents, enabling autonomous reasoning via API interactions. However, executing multi-step tasks within massive tool libraries remains challenging due to two critical bottlenecks: (1) the absence of rigorous, plan-level evaluation frameworks and (2) the computational demand of exploring vast decision spaces stemming from large toolsets and long-horizon planning. To bridge these gaps, we first introduce SLATE (Synthetic Large-scale API Toolkit for E-commerce), a large-scale context-aware benchmark designed for the automated assessment of tool-integrated agents. Unlike static metrics, SLATE accommodates diverse yet functionally valid execution trajectories, revealing that current agents struggle with self-correction and search efficiency. Motivated by these findings, we next propose Entropy-Guided Branching (EGB), an uncertainty-aware search algorithm that dynamically expands decision branches where predictive entropy is high. EGB optimizes the exploration-exploitation trade-off, significantly enhancing both task success rates and computational efficiency. Extensive experiments on SLATE demonstrate that our dual contribution provides a robust foundation for developing reliable and scalable LLM agents in tool-rich environments.

Iqbal Nouyed, Na Zhang

Unconstrained face recognition is an active research area among computer vision and biometric researchers for many years now. Still the problem of face recognition in low quality photos has not been well-studied so far. In this paper, we explore the face recognition performance on low quality photos, and we try to improve the accuracy in dealing with low quality face images. We assemble a large database with low quality photos, and examine the performance of face recognition algorithms for three different quality sets. Using state-of-the-art facial image enhancement approaches, we explore the face recognition performance for the enhanced face images. To perform this without experimental bias, we have developed a new protocol for recognition with low quality face photos and validate the performance experimentally. Our designed protocol for face recognition with low quality face images can be useful to other researchers. Moreover, experiment results show some of the challenging aspects of this problem.

Na Zhang, Moran Li, Chengming Xu et al.

Realistic hair strand generation is crucial for applications like computer graphics and virtual reality. While diffusion models can generate hairstyles from text or images, these inputs lack precision and user-friendliness. Instead, we propose the first sketch-based strand generation model, which offers finer control while remaining user-friendly. Our framework tackles key challenges, such as modeling complex strand interactions and diverse sketch patterns, through two main innovations: a learnable strand upsampling strategy that encodes 3D strands into multi-scale latent spaces, and a multi-scale adaptive conditioning mechanism using a transformer with diffusion heads to ensure consistency across granularity levels. Experiments on several benchmark datasets show our method outperforms existing approaches in realism and precision. Qualitative results further confirm its effectiveness. Code will be released at [GitHub](https://github.com/fighting-Zhang/StrandDesigner).

Na Zhang

Deep learning has received increasing interests in face recognition recently. Large quantities of deep learning methods have been proposed to handle various problems appeared in face recognition. Quite a lot deep methods claimed that they have gained or even surpassed human-level face verification performance in certain databases. As we know, face image quality poses a great challenge to traditional face recognition methods, e.g. model-driven methods with hand-crafted features. However, a little research focus on the impact of face image quality on deep learning methods, and even human performance. Therefore, we raise a question: Is face image quality still one of the challenges for deep learning based face recognition, especially in unconstrained condition. Based on this, we further investigate this problem on human level. In this paper, we partition face images into three different quality sets to evaluate the performance of deep learning methods on cross-quality face images in the wild, and then design a human face verification experiment on these cross-quality data. The result indicates that quality issue still needs to be studied thoroughly in deep learning, human own better capability in building the relations between different face images with large quality gaps, and saying deep learning method surpasses human-level is too optimistic.

Jin Wang, Jianxiang Lu, Comi Chen et al.

Generating high-quality 3D characters from single images remains a significant challenge in digital content creation, particularly due to complex body poses and self-occlusion. In this paper, we present RCM (Rotate your Character Model), an advanced image-to-video diffusion framework tailored for high-quality novel view synthesis (NVS) and 3D character generation. Compared to existing diffusion-based approaches, RCM offers several key advantages: (1) transferring characters with any complex poses into a canonical pose, enabling consistent novel view synthesis across the entire viewing orbit, (2) high-resolution orbital video generation at 1024x1024 resolution, (3) controllable observation positions given different initial camera poses, and (4) multi-view conditioning supporting up to 4 input images, accommodating diverse user scenarios. Extensive experiments demonstrate that RCM outperforms state-of-the-art methods in both novel view synthesis and 3D generation quality.

Na Zhang

MOBIO is a bi-modal database that was captured almost exclusively on mobile phones. It aims to improve research into deploying biometric techniques to mobile devices. Research has been shown that face and speaker recognition can be performed in a mobile environment. Facial landmark localization aims at finding the coordinates of a set of pre-defined key points for 2D face images. A facial landmark usually has specific semantic meaning, e.g. nose tip or eye centre, which provides rich geometric information for other face analysis tasks such as face recognition, emotion estimation and 3D face reconstruction. Pretty much facial landmark detection methods adopt still face databases, such as 300W, AFW, AFLW, or COFW, for evaluation, but seldomly use mobile data. Our work is first to perform facial landmark detection evaluation on the mobile still data, i.e., face images from MOBIO database. About 20,600 face images have been extracted from this audio-visual database and manually labeled with 22 landmarks as the groundtruth. Several state-of-the-art facial landmark detection methods are adopted to evaluate their performance on these data. The result shows that the data from MOBIO database is pretty challenging. This database can be a new challenging one for facial landmark detection evaluation.

Junxi Wang, Jize liu, Na Zhang et al.

This paper focuses to detect the fake news on the short video platforms. While significant research efforts have been devoted to this task with notable progress in recent years, current detection accuracy remains suboptimal due to the rapid evolution of content manipulation and generation technologies. Existing approaches typically employ a cross-modal fusion strategy that directly combines raw video data with metadata inputs before applying a classification layer. However, our empirical observations reveal a critical oversight: manipulated content frequently exhibits inter-modal inconsistencies that could serve as valuable discriminative features, yet remain underutilized in contemporary detection frameworks. Motivated by this insight, we propose a novel detection paradigm that explicitly identifies and leverages cross-modal contradictions as discriminative cues. Our approach consists of two core modules: Cross-modal Consistency Learning (CMCL) and Multi-modal Collaborative Diagnosis (MMCD). CMCL includes Pseudo-label Generation (PLG) and Cross-modal Consistency Diagnosis (CMCD). In PLG, a Multimodal Large Language Model is used to generate pseudo-labels for evaluating cross-modal semantic consistency. Then, CMCD extracts [CLS] tokens and computes cosine loss to quantify cross-modal inconsistencies. MMCD further integrates multimodal features through Multimodal Feature Fusion (MFF) and Probability Scores Fusion (PSF). MFF employs a co-attention mechanism to enhance semantic interactions across different modalities, while a Transformer is utilized for comprehensive feature fusion. Meanwhile, PSF further integrates the fake news probability scores obtained in the previous step. Extensive experiments on established benchmarks (FakeSV and FakeTT) demonstrate our model exhibits outstanding performance in Fake videos detection.

Natalie Brownlowe, Christopher R. Cornwell, Ethan Montes et al.

The choice of architecture of a neural network influences which functions will be realizable by that neural network and, as a result, studying the expressiveness of a chosen architecture has received much attention. In ReLU neural networks, the presence of stably unactivated neurons can reduce the network's expressiveness. In this work, we investigate the probability of a neuron in the second hidden layer of such neural networks being stably unactivated when the weights and biases are initialized from symmetric probability distributions. For networks with input dimension $n_0$, we prove that if the first hidden layer has $n_0+1$ neurons then this probability is exactly $\frac{2^{n_0}+1}{4^{n_0+1}}$, and if the first hidden layer has $n_1$ neurons, $n_1 \le n_0$, then the probability is $\frac{1}{2^{n_1+1}}$. Finally, for the case when the first hidden layer has more neurons than $n_0+1$, a conjecture is proposed along with the rationale. Computational evidence is presented to support the conjecture.

Yihao Feng, Ziyang Tang, Na Zhang et al.

Off-policy evaluation (OPE) is the task of estimating the expected reward of a given policy based on offline data previously collected under different policies. Therefore, OPE is a key step in applying reinforcement learning to real-world domains such as medical treatment, where interactive data collection is expensive or even unsafe. As the observed data tends to be noisy and limited, it is essential to provide rigorous uncertainty quantification, not just a point estimation, when applying OPE to make high stakes decisions. This work considers the problem of constructing non-asymptotic confidence intervals in infinite-horizon off-policy evaluation, which remains a challenging open question. We develop a practical algorithm through a primal-dual optimization-based approach, which leverages the kernel Bellman loss (KBL) of Feng et al.(2019) and a new martingale concentration inequality of KBL applicable to time-dependent data with unknown mixing conditions. Our algorithm makes minimum assumptions on the data and the function class of the Q-function, and works for the behavior-agnostic settings where the data is collected under a mix of arbitrary unknown behavior policies. We present empirical results that clearly demonstrate the advantages of our approach over existing methods.

Xingchao Liu, Xing Han, Na Zhang et al.

Learning monotonic models with respect to a subset of the inputs is a desirable feature to effectively address the fairness, interpretability, and generalization issues in practice. Existing methods for learning monotonic neural networks either require specifically designed model structures to ensure monotonicity, which can be too restrictive/complicated, or enforce monotonicity by adjusting the learning process, which cannot provably guarantee the learned model is monotonic on selected features. In this work, we propose to certify the monotonicity of the general piece-wise linear neural networks by solving a mixed integer linear programming problem.This provides a new general approach for learning monotonic neural networks with arbitrary model structures. Our method allows us to train neural networks with heuristic monotonicity regularizations, and we can gradually increase the regularization magnitude until the learned network is certified monotonic. Compared to prior works, our approach does not require human-designed constraints on the weight space and also yields more accurate approximation. Empirical studies on various datasets demonstrate the efficiency of our approach over the state-of-the-art methods, such as Deep Lattice Networks.

Ziyang Tang, Yihao Feng, Na Zhang et al.

Off-policy evaluation provides an essential tool for evaluating the effects of different policies or treatments using only observed data. When applied to high-stakes scenarios such as medical diagnosis or financial decision-making, it is crucial to provide provably correct upper and lower bounds of the expected reward, not just a classical single point estimate, to the end-users, as executing a poor policy can be very costly. In this work, we propose a provably correct method for obtaining interval bounds for off-policy evaluation in a general continuous setting. The idea is to search for the maximum and minimum values of the expected reward among all the Lipschitz Q-functions that are consistent with the observations, which amounts to solving a constrained optimization problem on a Lipschitz function space. We go on to introduce a Lipschitz value iteration method to monotonically tighten the interval, which is simple yet efficient and provably convergent. We demonstrate the practical efficiency of our method on a range of benchmarks.

Xinglei Wang, Xuefeng Guan, Jun Cao et al.

Obtaining accurate information about future traffic flows of all links in a traffic network is of great importance for traffic management and control applications. This research studies two particular problems in traffic forecasting: (1) capture the dynamic and non-local spatial correlation between traffic links and (2) model the dynamics of temporal dependency for accurate multiple steps ahead predictions. To address these issues, we propose a deep learning framework named Spatial-Temporal Sequence to Sequence model (STSeq2Seq). This model builds on sequence to sequence (seq2seq) architecture to capture temporal feature and relies on graph convolution for aggregating spatial information. Moreover, STSeq2Seq defines and constructs pattern-aware adjacency matrices (PAMs) based on pair-wise similarity of the recent traffic patterns on traffic links and integrate it into graph convolution operation. It also deploys a novel seq2sesq architecture which couples a convolutional encoder and a recurrent decoder with attention mechanism for dynamic modeling of long-range dependence between different time steps. We conduct extensive experiments using two publicly-available large-scale traffic datasets and compare STSeq2Seq with other baseline models. The numerical results demonstrate that the proposed model achieves state-of-the-art forecasting performance in terms of various error measures. The ablation study verifies the effectiveness of PAMs in capturing dynamic non-local spatial correlation and the superiority of proposed seq2seq architecture in modeling non-stationary temporal dependency for multiple steps ahead prediction. Furthermore, qualitative analysis is conducted on PAMs as well as the attention weights for model interpretation.

Na Zhang, Xuefeng Guan, Jun Cao et al.

Traffic forecasting is crucial for urban traffic management and guidance. However, existing methods rarely exploit the time-frequency properties of traffic speed observations, and often neglect the propagation of traffic flows from upstream to downstream road segments. In this paper, we propose a hybrid approach that learns the spatio-temporal dependency in traffic flows and predicts short-term traffic speeds on a road network. Specifically, we employ wavelet transform to decompose raw traffic data into several components with different frequency sub-bands. A Motif-based Graph Convolutional Recurrent Neural Network (Motif-GCRNN) and Auto-Regressive Moving Average (ARMA) are used to train and predict low-frequency components and high-frequency components, respectively. In the Motif-GCRNN framework, we integrate Graph Convolutional Networks (GCNs) with local sub-graph structures - Motifs - to capture the spatial correlations among road segments, and apply Long Short-Term Memory (LSTM) to extract the short-term and periodic patterns in traffic speeds. Experiments on a traffic dataset collected in Chengdu, China, demonstrate that the proposed hybrid method outperforms six state-of-art prediction methods.