Author name cluster

Xun Zhou

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.

30 papers

2 author rows

AAAI Conference 2026 Conference Paper

IndexTTS2: A Breakthrough in Emotionally Expressive and Duration-Controlled Auto-Regressive Zero-Shot Text-to-Speech

Siyi Zhou
Yiquan Zhou
Yi He
Xun Zhou
Jinchao Wang
Wei Deng
Jingchen Shu

Existing autoregressive large-scale text-to-speech (TTS) models have advantages in speech naturalness, but their token-by-token generation mechanism makes it difficult to precisely control the duration of synthesized speech. This becomes a significant limitation in applications requiring strict audio-visual synchronization, such as video dubbing. This paper introduces IndexTTS2, which proposes a novel, general, and autoregressive model-friendly method for speech duration control. The method supports two generation modes: one explicitly specifies the number of generated tokens to precisely control speech duration; the other freely generates speech in an autoregressive manner without specifying the number of tokens, while faithfully reproducing the prosodic features of the input prompt. Furthermore, IndexTTS2 achieves disentanglement between emotional expression and speaker identity, enabling independent control over timbre and emotion. In the zero-shot setting, the model can accurately reconstruct the target timbre (from the timbre prompt) while perfectly reproducing the specified emotional tone (from the style prompt). To enhance speech clarity in highly emotional expressions, we incorporate GPT latent representations and design a novel three-stage training paradigm to improve the stability of the generated speech. Additionally, to lower the barrier for emotional control, we designed a soft instruction mechanism based on text descriptions by fine-tuning Qwen3, effectively guiding the generation of speech with the desired emotional orientation. Finally, experimental results on multiple datasets show that IndexTTS2 outperforms state-of-the-art zero-shot TTS models in terms of word error rate, speaker similarity, and emotional fidelity.

PDF Details DOI

AAAI Conference 2026 Conference Paper

Sparse Poisson Gamma Belief Networks for High-Dimensional Sparse Count Data

Rui Huang
Dian Meng
Xun Zhou
Sikun Yang

Bayesian networks play a crucial role in various domains for unsupervised feature extraction and data interpretation. The Poisson gamma belief networks (PGBNs), as a type of Bayesian networks, have shown promise in analyzing high-dimensional count data. However, PGBNs encounter significant challenges when applied to sparse data, particularly in achieving accurate feature extraction and avoiding overfitting during missing value prediction. In this paper, we propose the sparse Poisson gamma belief networks (SPGBNs), a Bayesian network model designed to address these limitations. By incorporating sparse graph-structured priors over the weight matrices between adjacent layers, the proposed SPGBNs effectively capture the inherent sparsity and graph structures of latent features. Meanwhile, SPGBNs demonstrate superior generalization on missing data prediction and enable more stable extraction of meaningful latent features compared to existing approaches. Additionally, we develop an efficient Gibbs sampling algorithm that significantly improves the training stability and computational efficiency of SPGBNs. Extensive experiments on real-world datasets are conducted to validate the effectiveness of our approach.

PDF Details DOI

AAAI Conference 2025 Conference Paper

Design Principle Transfer in Neural Architecture Search via Large Language Models

Xun Zhou
Xingyu Wu
Liang Feng
Zhichao Lu
Kay Chen Tan

Transferable neural architecture search (TNAS) has been introduced to design efficient neural architectures for multiple tasks, to enhance the practical applicability of NAS in real-world scenarios. In TNAS, architectural knowledge accumulated in previous search processes is reused to warm up the architecture search for new tasks. However, existing TNAS methods still search in an extensive search space, necessitating the evaluation of numerous architectures. To overcome this challenge, this work proposes a novel transfer paradigm, i.e., design principle transfer. In this work, the linguistic description of various structural components' effects on architectural performance is termed design principles. They are learned from established architectures and then can be reused to reduce the search space tasks by discarding unpromising architectures. Searching in the refined search space can boost both the search performance and efficiency for new NAS tasks. To this end, a large language model (LLM)-assisted design principle transfer (LAPT) framework is devised. In LAPT, LLM is applied to automatically reason the design principles from a set of given architectures, and then a principle adaptation method is applied to refine these principles progressively based on the search results. Experimental results demonstrate that LAPT can beat the state-of-the-art TNAS methods on most tasks and achieve comparable performance on the remainder.

PDF Details DOI

ICML Conference 2025 Conference Paper

Expert Race: A Flexible Routing Strategy for Scaling Diffusion Transformer with Mixture of Experts

Yike Yuan
Ziyu Wang
Zihao Huang
Defa Zhu
Xun Zhou
Jingyi Yu 0001
Qiyang Min

Diffusion models have emerged as mainstream framework in visual generation. Building upon this success, the integration of Mixture of Experts (MoE) methods has shown promise in enhancing model scalability and performance. In this paper, we introduce Race-DiT, a novel MoE model for diffusion transformers with a flexible routing strategy, Expert Race. By allowing tokens and experts to compete together and select the top candidates, the model learns to dynamically assign experts to critical tokens. Additionally, we propose per-layer regularization to address challenges in shallow layer learning, and router similarity loss to prevent mode collapse, ensuring better expert utilization. Extensive experiments on ImageNet validate the effectiveness of our approach, showcasing significant performance gains while promising scaling properties.

Details

ICLR Conference 2025 Conference Paper

FlexPrefill: A Context-Aware Sparse Attention Mechanism for Efficient Long-Sequence Inference

Xunhao Lai
Jianqiao Lu
Yao Luo
Yiyuan Ma
Xun Zhou

Large language models (LLMs) encounter computational challenges during long-sequence inference, especially in the attention pre-filling phase, where the complexity grows quadratically with the prompt length. Previous efforts to mitigate these challenges have relied on fixed sparse attention patterns or identifying sparse attention patterns based on limited cases. However, these methods lacked the flexibility to efficiently adapt to varying input demands. In this paper, we introduce FlexPrefill, a Flexible sparse Pre-filling mechanism that dynamically adjusts sparse attention patterns and computational budget in real-time to meet the specific requirements of each input and attention head. The flexibility of our method is demonstrated through two key innovations: 1) Query-Aware Sparse Pattern Determination: By measuring Jensen-Shannon divergence, this component adaptively switches between query-specific diverse attention patterns and predefined attention patterns. 2) Cumulative-Attention Based Index Selection: This component dynamically selects query-key indexes to be computed based on different attention patterns, ensuring the sum of attention scores meets a predefined threshold. FlexPrefill adaptively optimizes the sparse pattern and sparse ratio of each attention head based on the prompt, enhancing efficiency in long-sequence inference tasks. Experimental results show significant improvements in both speed and accuracy over prior methods, providing a more flexible and efficient solution for LLM inference.

Details

AAAI Conference 2025 Conference Paper

GeoPro-Net: Learning Interpretable Spatiotemporal Prediction Models Through Statistically-Guided Geo-Prototyping

Bang An
Xun Zhou
Zirui Zhou
Ronilo Ragodos
Zenglin Xu
Jun Luo

The problem of forecasting spatiotemporal events such as crimes and accidents is crucial to public safety and city management. Besides accuracy, interpretability is also a key requirement for spatiotemporal forecasting models to justify the decisions. Merely presenting predicted scores fails to convince the public and does not contribute to future urban planning. Interpretation of the spatiotemporal forecasting mechanism is, however, challenging due to the complexity of multi-source spatiotemporal features, the non-intuitive nature of spatiotemporal patterns for non-expert users, and the presence of spatial heterogeneity in the data. Currently, no existing deep learning model intrinsically interprets the complex predictive process learned from multi-source spatiotemporal features. To bridge the gap, we propose GeoPro-Net, an intrinsically interpretable spatiotemporal model for spatiotemporal event forecasting problems. GeoPro-Net introduces a novel Geo-concept convolution operation, which employs statistical tests to extract predictive patterns in the input as "Geo-concepts'', and condenses the "Geo-concept-encoded'' input through interpretable channel fusion and geographic-based pooling. In addition, GeoPro-Net learns different sets of prototypes of concepts inherently, and projects them to real-world cases for interpretation. Comprehensive experiments and case studies on four real-world datasets demonstrate that GeoPro-Net provides better interpretability while still achieving competitive prediction performance compared with state-of-the-art baselines.

PDF Details DOI

ICLR Conference 2025 Conference Paper

Hyper-Connections

Defa Zhu
Hongzhi Huang
Zihao Huang
Yutao Zeng
Yunyao Mao
Banggu Wu
Qiyang Min
Xun Zhou

We present hyper-connections, a simple yet effective method that can serve as an alternative to residual connections. This approach specifically addresses common drawbacks observed in residual connection variants, such as the seesaw effect between gradient vanishing and representation collapse. Theoretically, hyper-connections allow the network to adjust the strength of connections between features at different depths and dynamically rearrange layers. We conduct experiments focusing on the pre-training of large language models, including dense and sparse models, where hyper-connections show significant performance improvements over residual connections. Additional experiments conducted on vision tasks also demonstrate similar improvements. We anticipate that this method will be broadly applicable and beneficial across a wide range of AI problems.

Details

ICML Conference 2025 Conference Paper

MARS: Unleashing the Power of Variance Reduction for Training Large Models

Huizhuo Yuan
Yifeng Liu 0004
Shuang Wu
Xun Zhou
Quanquan Gu

Training deep neural networks–and more recently, large models–demands efficient and scalable optimizers. Adaptive gradient algorithms like Adam, AdamW, and their variants have been central to this task. Despite the development of numerous variance reduction algorithms in the past decade aimed at accelerating stochastic optimization in both convex and nonconvex settings, variance reduction has not found widespread success in training deep neural networks or large language models. Consequently, it has remained a less favored approach in modern AI. In this paper, to unleash the power of variance reduction for efficient training of large models, we propose a unified optimization framework, MARS ( M ake v A riance R eduction S hine), which reconciles preconditioned gradient methods with variance reduction via a scaled stochastic recursive momentum technique. Within our framework, we introduce three instances of MARS that leverage preconditioned gradient updates based on AdamW, Lion, and Shampoo, respectively. We also draw a connection between our algorithms and existing optimizers. Experimental results on training GPT-2 models indicate that MARS consistently outperforms AdamW by a large margin.

Details

ICML Conference 2025 Conference Paper

Over-Tokenized Transformer: Vocabulary is Generally Worth Scaling

Hongzhi Huang
Defa Zhu
Banggu Wu
Yutao Zeng
Ya Wang
Qiyang Min
Xun Zhou

Tokenization is a fundamental component of large language models (LLMs), yet its influence on model scaling and performance is not fully explored. In this paper, we introduce Over-Tokenized Transformers, a novel framework that decouples input and output vocabularies to improve language modeling performance. Specifically, our approach scales up input vocabularies to leverage multi-gram tokens. Through extensive experiments, we uncover a log-linear relationship between input vocabulary size and training loss, demonstrating that larger input vocabularies consistently enhance model performance, regardless of model size. Using a large input vocabulary, we achieve performance comparable to double-sized baselines with no additional cost. Our findings highlight the importance of tokenization in scaling laws and provide practical insight for tokenizer design, paving the way for more efficient and powerful LLMs.

Details

ICLR Conference 2025 Conference Paper

Polynomial Composition Activations: Unleashing the Dynamics of Large Language Models

Zhijian Zhuo 0001
Ya Wang 0002
Yutao Zeng
Xiaoqing Li
Xun Zhou
Jinwen Ma

Transformers have found extensive applications across various domains due to their powerful fitting capabilities. This success can be partially attributed to their inherent nonlinearity. Thus, in addition to the ReLU function employed in the original transformer architecture, researchers have explored alternative modules such as GeLU and SwishGLU to enhance nonlinearity and thereby augment representational capacity. In this paper, we propose a novel category of polynomial composition activations (PolyCom), designed to optimize the dynamics of transformers. Theoretically, we provide a comprehensive mathematical analysis of PolyCom, highlighting its enhanced expressivity and efficacy relative to other activation functions. Notably, we demonstrate that networks incorporating PolyCom achieve the **optimal approximation rate**, indicating that PolyCom networks require minimal parameters to approximate general smooth functions in Sobolev spaces. We conduct empirical experiments on the pre-training configurations of large language models (LLMs), including both dense and sparse architectures. By substituting conventional activation functions with PolyCom, we enable LLMs to capture higher-order interactions within the data, thus improving performance metrics in terms of accuracy and convergence rates. Extensive experimental results demonstrate the effectiveness of our method, showing substantial improvements over other activation functions. Code is available at https://github.com/BryceZhuo/PolyCom.

Details

ICLR Conference 2025 Conference Paper

Ultra-Sparse Memory Network

Zihao Huang
Qiyang Min
Hongzhi Huang
Yutao Zeng
Defa Zhu
Ran Guo
Xun Zhou

It is widely acknowledged that the performance of Transformer models is logarithmically related to their number of parameters and computational complexity. While approaches like Mixture of Experts (MoE) decouple parameter count from computational complexity, they still face challenges in inference due to high memory access costs. This work introduces UltraMem, incorporating large-scale, ultra-sparse memory layer to address these limitations. Our approach significantly reduces inference latency while maintaining model performance. We also investigate the scaling laws of this new architecture, demonstrating that it not only exhibits favorable scaling properties but outperforms MoE. In experiments, the largest UltraMem we train has \textbf{20 million} memory slots. The results show that our method achieves state-of-the-art inference speed and model performance within a given computational budget, paving the way for billions of slots or experts.

Details

NeurIPS Conference 2025 Conference Paper

UniTraj: Learning a Universal Trajectory Foundation Model from Billion-Scale Worldwide Traces

Yuanshao Zhu
James Yu
Xiangyu Zhao
Xun Zhou
Liang Han
Xuetao Wei
Yuxuan Liang

Building a universal trajectory foundation model is a promising solution to address the limitations of existing trajectory modeling approaches, such as task specificity, regional dependency, and data sensitivity. Despite its potential, data preparation, pre-training strategy development, and architectural design present significant challenges in constructing this model. Therefore, we introduce UniTraj, a Universal Trajectory foundation model that aims to address these limitations through three key innovations. First, we construct WorldTrace, an unprecedented dataset of 2. 45 million trajectories with billions of GPS points spanning 70 countries, providing the diverse geographic coverage essential for region-independent modeling. Second, we develop novel pre-training strategies--Adaptive Trajectory Resampling and Self-supervised Trajectory Masking--that enable robust learning from heterogeneous trajectory data with varying sampling rates and quality. Finally, we tailor a flexible model architecture to accommodate a variety of trajectory tasks, effectively capturing complex movement patterns to support broad applicability. Extensive experiments across multiple tasks and real-world datasets demonstrate that UniTraj consistently outperforms existing methods, exhibiting superior scalability, adaptability, and generalization, with WorldTrace serving as an ideal yet non-exclusive training resource. The implementation codes and full dataset are available at https: //github. com/Yasoz/UniTraj.

PDF Details

AAAI Conference 2024 Conference Paper

FT-GAN: Fine-Grained Tune Modeling for Chinese Opera Synthesis

Meizhen Zheng
Peng Bai
Xiaodong Shi
Xun Zhou
Yiting Yan

Although singing voice synthesis (SVS) has made significant progress recently, with its unique styles and various genres, Chinese opera synthesis requires greater attention but is rarely studied for lack of training data and high expressiveness. In this work, we build a high-quality Gezi Opera (a type of Chinese opera popular in Fujian and Taiwan) audio-text alignment dataset and formulate specific data annotation methods applicable to Chinese operas. We propose FT-GAN, an acoustic model for fine-grained tune modeling in Chinese opera synthesis based on the empirical analysis of the differences between Chinese operas and pop songs. To further improve the quality of the synthesized opera, we propose a speech pre-training strategy for additional knowledge injection. The experimental results show that FT-GAN outperforms the strong baselines in SVS on the Gezi Opera synthesis task. Extensive experiments further verify that FT-GAN performs well on synthesis tasks of other operas such as Peking Opera. Audio samples, the dataset, and the codes are available at https://zhengmidon.github.io/FTGAN.github.io/.

PDF Details DOI

NeurIPS Conference 2024 Conference Paper

MoTE: Reconciling Generalization with Specialization for Visual-Language to Video Knowledge Transfer

Minghao Zhu
Zhengpu Wang
Mengxian Hu
Ronghao Dang
Xiao Lin
Xun Zhou
Chengju Liu
Qijun Chen

Transferring visual-language knowledge from large-scale foundation models for video recognition has proved to be effective. To bridge the domain gap, additional parametric modules are added to capture the temporal information. However, zero-shot generalization diminishes with the increase in the number of specialized parameters, making existing works a trade-off between zero-shot and close-set performance. In this paper, we present MoTE, a novel framework that enables generalization and specialization to be balanced in one unified model. Our approach tunes a mixture of temporal experts to learn multiple task views with various degrees of data fitting. To maximally preserve the knowledge of each expert, we propose Weight Merging Regularization, which regularizes the merging process of experts in weight space. Additionally with temporal feature modulation to regularize the contribution of temporal feature during test. We achieve a sound balance between zero-shot and close-set video recognition tasks and obtain state-of-the-art or competitive results on various datasets, including Kinetics-400 & 600, UCF, and HMDB. Code is available at https: //github. com/ZMHH-H/MoTE.

PDF Details DOI

AAAI Conference 2024 Conference Paper

Referee-Meta-Learning for Fast Adaptation of Locational Fairness

Weiye Chen
Yiqun Xie
Xiaowei Jia
Erhu He
Han Bao
Bang An
Xun Zhou

When dealing with data from distinct locations, machine learning algorithms tend to demonstrate an implicit preference of some locations over the others, which constitutes biases that sabotage the spatial fairness of the algorithm. This unfairness can easily introduce biases in subsequent decision-making given broad adoptions of learning-based solutions in practice. However, locational biases in AI are largely understudied. To mitigate biases over locations, we propose a locational meta-referee (Meta-Ref) to oversee the few-shot meta-training and meta-testing of a deep neural network. Meta-Ref dynamically adjusts the learning rates for training samples of given locations to advocate a fair performance across locations, through an explicit consideration of locational biases and the characteristics of input data. We present a three-phase training framework to learn both a meta-learning-based predictor and an integrated Meta-Ref that governs the fairness of the model. Once trained with a distribution of spatial tasks, Meta-Ref is applied to samples from new spatial tasks (i.e., regions outside the training area) to promote fairness during the fine-tune step. We carried out experiments with two case studies on crop monitoring and transportation safety, which show Meta-Ref can improve locational fairness while keeping the overall prediction quality at a similar level.

PDF Details DOI

NeurIPS Conference 2024 Conference Paper

Seeing the Image: Prioritizing Visual Correlation by Contrastive Alignment

Xin Xiao
Bohong Wu
Jiacong Wang
Chunyuan Li
Xun Zhou
Haoyuan Guo

Existing image-text modality alignment in Vision Language Models (VLMs) treats each text token equally in an autoregressive manner. Despite being simple and effective, this method results in sub-optimal cross-modal alignment by over-emphasizing the text tokens that are less correlated with or even contradictory with the input images. In this paper, we advocate for distinct contributions for each text token based on its visual correlation. Specifically, we present by contrasting image inputs, the difference in prediction logits on each text token provides strong guidance of visual correlation. We therefore introduce Contrastive Alignment (CAL), a simple yet effective re-weighting strategy that prioritizes training visually correlated tokens. Our experimental results demonstrate that CAL consistently improves different types of VLMs across different resolutions and model sizes on various benchmark datasets. Importantly, our method incurs minimal additional computational overhead, rendering it highly efficient compared to alternative data scaling strategies.

PDF Details DOI

IROS Conference 2024 Conference Paper

SNF-Feat: Semantic-Guided Negative-Sample-Free Representation Learning for Local Feature Extraction

Xun Zhou
Qingqing Yan
Minghao Zhu
Mengxian Hu
Chengju Liu
Qijun Chen

Local feature extraction constitutes a foundational module crucial for numerous downstream tasks of computer vision. Its primary challenge lies in the generation of discriminative feature representations. Prior methodologies have employed contrastive learning within their pipelines, yet have encountered limitations stemming from inherent conflicts within their training data, including the ambiguity of negative samples and the distortion of positive samples. In this study, we propose a semantic-guided negative-sample-free method for local feature learning, denoted as SNF-Feat. Our framework entails dense patch-level representation learning without reliance on negative samples, aiming to ensure that descriptors derived from transformed views of the same local area exhibit predictive capability towards each other. To assess the impact of positive sample distortion, we harness high-level semantic information to derive point-wise loss weights. Furthermore, we establish a self-supervised feature learning paradigm that extends our utilization of datasets. Experimental results demonstrate the superior performance of our method across a range of typical datasets and tasks in comparison to state-of-the-art approaches.

Details

AAAI Conference 2023 Conference Paper

Auto-CM: Unsupervised Deep Learning for Satellite Imagery Composition and Cloud Masking Using Spatio-Temporal Dynamics

Yiqun Xie
Zhili Li
Han Bao
Xiaowei Jia
Dongkuan Xu
Xun Zhou
Sergii Skakun

Cloud masking is both a fundamental and a critical task in the vast majority of Earth observation problems across social sectors, including agriculture, energy, water, etc. The sheer volume of satellite imagery to be processed has fast-climbed to a scale (e.g., >10 PBs/year) that is prohibitive for manual processing. Meanwhile, generating reliable cloud masks and image composite is increasingly challenging due to the continued distribution-shifts in the imagery collected by existing sensors and the ever-growing variety of sensors and platforms. Moreover, labeled samples are scarce and geographically limited compared to the needs in real large-scale applications. In related work, traditional remote sensing methods are often physics-based and rely on special spectral signatures from multi- or hyper-spectral bands, which are often not available in data collected by many -- and especially more recent -- high-resolution platforms. Machine learning and deep learning based methods, on the other hand, often require large volumes of up-to-date training data to be reliable and generalizable over space. We propose an autonomous image composition and masking (Auto-CM) framework to learn to solve the fundamental tasks in a label-free manner, by leveraging different dynamics of events in both geographic domains and time-series. Our experiments show that Auto-CM outperforms existing methods on a wide-range of data with different satellite platforms, geographic regions and bands.

PDF Details DOI

NeurIPS Conference 2023 Conference Paper

SpatialRank: Urban Event Ranking with NDCG Optimization on Spatiotemporal Data

Bang An
Xun Zhou
YONGJIAN ZHONG
Tianbao Yang

The problem of urban event ranking aims at predicting the top-$k$ most risky locations of future events such as traffic accidents and crimes. This problem is of fundamental importance to public safety and urban administration especially when limited resources are available. The problem is, however, challenging due to complex and dynamic spatio-temporal correlations between locations, uneven distribution of urban events in space, and the difficulty to correctly rank nearby locations with similar features. Prior works on event forecasting mostly aim at accurately predicting the actual risk score or counts of events for all the locations. Rankings obtained as such usually have low quality due to prediction errors. Learning-to-rank methods directly optimize measures such as Normalized Discounted Cumulative Gain (NDCG), but cannot handle the spatiotemporal autocorrelation existing among locations. Due to the common assumption that items are independent. In this paper, we bridge the gap by proposing a novel spatial event ranking approach named SpatialRank. SpatialRank features adaptive graph convolution layers that dynamically learn the spatiotemporal dependencies across locations from data. In addition, the model optimizes through surrogates a hybrid NDCG loss with a spatial component to better rank neighboring spatial locations. We design an importance-sampling with a spatial filtering algorithm to effectively evaluate the loss during training. Comprehensive experiments on three real-world datasets demonstrate that SpatialRank can effectively identify the top riskiest locations of crimes and traffic accidents and outperform state-of-art methods in terms of NDCG by up to 12. 7%.

PDF Details

TIST Journal 2022 Journal Article

COVID-GAN+: Estimating Human Mobility Responses to COVID-19 through Spatio-temporal Generative Adversarial Networks with Enhanced Features

Han Bao
Xun Zhou
Yiqun Xie
Yingxue Zhang
Yanhua Li

Estimating human mobility responses to the large-scale spreading of the COVID-19 pandemic is crucial, since its significance guides policymakers to give Non-pharmaceutical Interventions, such as closure or reopening of businesses. It is challenging to model due to complex social contexts and limited training data. Recently, we proposed a conditional generative adversarial network (COVID-GAN) to estimate human mobility response under a set of social and policy conditions integrated from multiple data sources. Although COVID-GAN achieves a good average estimation accuracy under real-world conditions, it produces higher errors in certain regions due to the presence of spatial heterogeneity and outliers. To address these issues, in this article, we extend our prior work by introducing a new spatio-temporal deep generative model, namely, COVID-GAN+. COVID-GAN+ deals with the spatial heterogeneity issue by introducing a new spatial feature layer that utilizes the local Moran statistic to model the spatial heterogeneity strength in the data. In addition, we redesign the training objective to learn the estimated mobility changes from historical average levels to mitigate the effects of spatial outliers. We perform comprehensive evaluations using urban mobility data derived from cell phone records and census data. Results show that COVID-GAN+ can better approximate real-world human mobility responses than prior methods, including COVID-GAN.

Details DOI

NeurIPS Conference 2022 Conference Paper

ProtoX: Explaining a Reinforcement Learning Agent via Prototyping

Ronilo Ragodos
Tong Wang
Qihang Lin
Xun Zhou

While deep reinforcement learning has proven to be successful in solving control tasks, the ``black-box'' nature of an agent has received increasing concerns. We propose a prototype-based post-hoc \emph{policy explainer}, ProtoX, that explains a black-box agent by prototyping the agent's behaviors into scenarios, each represented by a prototypical state. When learning prototypes, ProtoX considers both visual similarity and scenario similarity. The latter is unique to the reinforcement learning context since it explains why the same action is taken in visually different states. To teach ProtoX about visual similarity, we pre-train an encoder using contrastive learning via self-supervised learning to recognize states as similar if they occur close together in time and receive the same action from the black-box agent. We then add an isometry layer to allow ProtoX to adapt scenario similarity to the downstream task. ProtoX is trained via imitation learning using behavior cloning, and thus requires no access to the environment or agent. In addition to explanation fidelity, we design different prototype shaping terms in the objective function to encourage better interpretability. We conduct various experiments to test ProtoX. Results show that ProtoX achieved high fidelity to the original black-box agent while providing meaningful and understandable explanations.

PDF Details

IJCAI Conference 2022 Conference Paper

Statistically-Guided Deep Network Transformation to Harness Heterogeneity in Space (Extended Abstract)

Yiqun Xie
Erhu He
Xiaowei Jia
Han Bao
Xun Zhou
Rahul Ghosh
Praveen Ravirathinam

Spatial data are ubiquitous and have transformed decision-making in many critical domains, including public health, agriculture, transportation, etc. While recent advances in machine learning offer promising ways to harness massive spatial datasets (e. g. , satellite imagery), spatial heterogeneity -- a fundamental property of spatial data -- poses a major challenge as data distributions or generative processes often vary over space. Recent studies targeting this difficult problem either require a known space-partitioning as the input, or can only support limited special cases (e. g. , binary classification). Moreover, heterogeneity-pattern learned by these methods are locked to the locations of the training samples, and cannot be applied to new locations. We propose a statistically-guided framework to adaptively partition data in space during training using distribution-driven optimization and transform a deep learning model (of user's choice) into a heterogeneity-aware architecture. We also propose a spatial moderator to generalize learned patterns to new test regions. Experiment results on real-world datasets show that the framework can effectively capture footprints of heterogeneity and substantially improve prediction performances.

PDF Details DOI

TIST Journal 2022 Journal Article

Urban Traffic Dynamics Prediction—A Continuous Spatial-temporal Meta-learning Approach

Yingxue Zhang
Yanhua Li
Xun Zhou
Jun Luo
Zhi-Li Zhang

Urban traffic status (e.g., traffic speed and volume) is highly dynamic in nature, namely, varying across space and evolving over time. Thus, predicting such traffic dynamics is of great importance to urban development and transportation management. However, it is very challenging to solve this problem due to spatial-temporal dependencies and traffic uncertainties. In this article, we solve the traffic dynamics prediction problem from Bayesian meta-learning perspective and propose a novel continuous spatial-temporal meta-learner (cST-ML), which is trained on a distribution of traffic prediction tasks segmented by historical traffic data with the goal of learning a strategy that can be quickly adapted to related but unseen traffic prediction tasks. cST-ML tackles the traffic dynamics prediction challenges by advancing the Bayesian black-box meta-learning framework through the following new points: (1) cST-ML captures the dynamics of traffic prediction tasks using variational inference, and to better capture the temporal uncertainties within tasks, cST-ML performs as a rolling window within each task; (2) cST-ML has novel designs in architecture, where CNN and LSTM are embedded to capture the spatial-temporal dependencies between traffic status and traffic-related features; (3) novel training and testing algorithms for cST-ML are designed. We also conduct experiments on two real-world traffic datasets (taxi inflow and traffic speed) to evaluate our proposed cST-ML. The experimental results verify that cST-ML can significantly improve the urban traffic prediction performance and outperform all baseline models especially when obvious traffic dynamics and temporal uncertainties are presented.

Details DOI

TIST Journal 2021 Journal Article

DILSA+: Predicting Urban Dispersal Events through Deep Survival Analysis with Enhanced Urban Features

Amin Vahedian Khezerlou
Xun Zhou
Xinyi Li
W. Nick Street
Yanhua Li

Urban dispersal events occur when an unexpectedly large number of people leave an area in a relatively short period of time. It is beneficial for the city authorities, such as law enforcement and city management, to have an advance knowledge of such events, as it can help them mitigate the safety risks and handle important challenges such as managing traffic, and so forth. Predicting dispersal events is also beneficial to Taxi drivers and/or ride-sharing services, as it will help them respond to an unexpected demand and gain competitive advantage. Large urban datasets such as detailed trip records and point of interest ( POI ) data make such predictions achievable. The related literature mainly focused on taxi demand prediction. The pattern of the demand was assumed to be repetitive and proposed methods aimed at capturing those patterns. However, dispersal events are, by definition, violations of those patterns and are, understandably, missed by the methods in the literature. We proposed a different approach in our prior work [32]. We showed that dispersal events can be predicted by learning the complex patterns of arrival and other features that precede them in time. We proposed a survival analysis formulation of this problem and proposed a two-stage framework (DILSA), where a deep learning model predicted the survival function at each point in time in the future. We used that prediction to determine the time of the dispersal event in the future, or its non-occurrence. However, DILSA is subject to a few limitations. First, based on evidence from the data, mobility patterns can vary through time at a given location. DILSA does not distinguish between different mobility patterns through time. Second, mobility patterns are also different for different locations. DILSA does not have the capability to directly distinguish between different locations based on their mobility patterns. In this article, we address these limitations by proposing a method to capture the interaction between POIs and mobility patterns and we create vector representations of locations based on their mobility patterns. We call our new method DILSA+. We conduct extensive case studies and experiments on the NYC Yellow taxi dataset from 2014 to 2016. Results show that DILSA+ can predict events in the next 5 hours with an F1-score of 0.66. It is significantly better than DILSA and the state-of-the-art deep learning approaches for taxi demand prediction.

Details DOI

TIST Journal 2021 Journal Article

Significant DBSCAN+: Statistically Robust Density-based Clustering

Yiqun Xie
Xiaowei Jia
Shashi Shekhar
Han Bao
Xun Zhou

Cluster detection is important and widely used in a variety of applications, including public health, public safety, transportation, and so on. Given a collection of data points, we aim to detect density-connected spatial clusters with varying geometric shapes and densities, under the constraint that the clusters are statistically significant. The problem is challenging, because many societal applications and domain science studies have low tolerance for spurious results, and clusters may have arbitrary shapes and varying densities. As a classical topic in data mining and learning, a myriad of techniques have been developed to detect clusters with both varying shapes and densities (e.g., density-based, hierarchical, spectral, or deep clustering methods). However, the vast majority of these techniques do not consider statistical rigor and are susceptible to detecting spurious clusters formed as a result of natural randomness. On the other hand, scan statistic approaches explicitly control the rate of spurious results, but they typically assume a single “hotspot” of over-density and many rely on further assumptions such as a tessellated input space. To unite the strengths of both lines of work, we propose a statistically robust formulation of a multi-scale DBSCAN, namely Significant DBSCAN+, to identify significant clusters that are density connected. As we will show, incorporation of statistical rigor is a powerful mechanism that allows the new Significant DBSCAN+ to outperform state-of-the-art clustering techniques in various scenarios. We also propose computational enhancements to speed-up the proposed approach. Experiment results show that Significant DBSCAN+ can simultaneously improve the success rate of true cluster detection (e.g., 10–20% increases in absolute F1 scores) and substantially reduce the rate of spurious results (e.g., from thousands/hundreds of spurious detections to none or just a few across 100 datasets), and the acceleration methods can improve the efficiency for both clustered and non-clustered data.

Details DOI

TIST Journal 2020 Journal Article

DHPA

Menghai Pan
Weixiao Huang
Yanhua Li
Xun Zhou
Zhenming Liu
Rui Song
Hui Lu
Zhihong Tian

Many real-world human behaviors can be modeled and characterized as sequential decision-making processes, such as a taxi driver’s choices of working regions and times. Each driver possesses unique preferences on the sequential choices over time and improves the driver’s working efficiency. Understanding the dynamics of such preferences helps accelerate the learning process of taxi drivers. Prior works on taxi operation management mostly focus on finding optimal driving strategies or routes, lacking in-depth analysis on what the drivers learned during the process and how they affect the performance of the driver. In this work, we make the first attempt to establish Dynamic Human Preference Analytics. We inversely learn the taxi drivers’ preferences from data and characterize the dynamics of such preferences over time. We extract two types of features (i.e., profile features and habit features) to model the decision space of drivers. Then through inverse reinforcement learning, we learn the preferences of drivers with respect to these features. The results illustrate that self-improving drivers tend to keep adjusting their preferences to habit features to increase their earning efficiency while keeping the preferences to profile features invariant. However, experienced drivers have stable preferences over time. The exploring drivers tend to randomly adjust the preferences over time.

Details DOI

TIST Journal 2020 Journal Article

Discovering Interesting Subpaths with Statistical Significance from Spatiotemporal Datasets

Yiqun Xie
Xun Zhou
Shashi Shekhar

Given a path in a spatial or temporal framework, we aim to find all contiguous subpaths that are both interesting (e.g., abrupt changes) and statistically significant (i.e., persistent trends rather than local fluctuations). Discovering interesting subpaths can provide meaningful information for a variety of domains including Earth science, environmental science, urban planning, and the like. Existing methods are limited to detecting individual points of interest along an input path but cannot find interesting subpaths. Our preliminary work provided a Subpath Enumeration and Pruning (SEP) algorithm to detect interesting subpaths of arbitrary length. However, SEP is not effective in avoiding detections that are random variations rather than meaningful trends, which hampers clear and proper interpretations of the results. In this article, we extend our previous work by proposing a significance testing framework to eliminate these random variations. To compute the statistical significance, we first show a baseline Monte-Carlo method based on our previous work and then propose a Dynamic Search-and-Prune (D-SAP) algorithm to improve its computational efficiency. Our experiments show that the significance testing can greatly suppress the noisy detections in the output and D-SAP can greatly reduce the execution time.

Details DOI

AAAI Conference 2019 Conference Paper

Predicting Urban Dispersal Events: A Two-Stage Framework through Deep Survival Analysis on Mobility Data

Amin Vahedian
Xun Zhou
Ling Tong
W. Nick Street
Yanhua Li

Urban dispersal events are processes where an unusually large number of people leave the same area in a short period. Early prediction of dispersal events is important in mitigating congestion and safety risks and making better dispatching decisions for taxi and ride-sharing fleets. Existing work mostly focuses on predicting taxi demand in the near future by learning patterns from historical data. However, they fail in case of abnormality because dispersal events with abnormally high demand are non-repetitive and violate common assumptions such as smoothness in demand change over time. Instead, in this paper we argue that dispersal events follow a complex pattern of trips and other related features in the past, which can be used to predict such events. Therefore, we formulate the dispersal event prediction problem as a survival analysis problem. We propose a two-stage framework (DILSA), where a deep learning model combined with survival analysis is developed to predict the probability of a dispersal event and its demand volume. We conduct extensive case studies and experiments on the NYC Yellow taxi dataset from 2014- 2016. Results show that DILSA can predict events in the next 5 hours with F1-score of 0. 7 and with average time error of 18 minutes. It is orders of magnitude better than the state-ofthe-art deep learning approaches for taxi demand prediction.

PDF Details

NeurIPS Conference 2018 Conference Paper

Faster Online Learning of Optimal Threshold for Consistent F-measure Optimization

Xiaoxuan Zhang
Mingrui Liu
Xun Zhou
Tianbao Yang

In this paper, we consider online F-measure optimization (OFO). Unlike traditional performance metrics (e. g. , classification error rate), F-measure is non-decomposable over training examples and is a non-convex function of model parameters, making it much more difficult to be optimized in an online fashion. Most existing results of OFO usually suffer from high memory/computational costs and/or lack statistical consistency guarantee for optimizing F-measure at the population level. To advance OFO, we propose an efficient online algorithm based on simultaneously learning a posterior probability of class and learning an optimal threshold by minimizing a stochastic strongly convex function with unknown strong convexity parameter. A key component of the proposed method is a novel stochastic algorithm with low memory and computational costs, which can enjoy a convergence rate of $\widetilde O(1/\sqrt{n})$ for learning the optimal threshold under a mild condition on the convergence of the posterior probability, where $n$ is the number of processed examples. It is provably faster than its predecessor based on a heuristic for updating the threshold. The experiments verify the efficiency of the proposed algorithm in comparison with state-of-the-art OFO algorithms.

PDF Details

TIST Journal 2017 Journal Article

A Traffic Flow Approach to Early Detection of Gathering Events

Amin Vahedian Khezerlou
Xun Zhou
Lufan Li
Zubair Shafiq
Alex X. Liu
Fan Zhang

Given a spatial field and the traffic flow between neighboring locations, the early detection of gathering events ( edge ) problem aims to discover and localize a set of most likely gathering events. It is important for city planners to identify emerging gathering events that might cause public safety or sustainability concerns. However, it is challenging to solve the edge problem due to numerous candidate gathering footprints in a spatial field and the nontrivial task of balancing pattern quality and computational efficiency. Prior solutions to model the edge problem lack the ability to describe the dynamic flow of traffic and the potential gathering destinations because they rely on static or undirected footprints. In our recent work, we modeled the footprint of a gathering event as a Gathering Graph (G-Graph), where the root of the directed acyclic G-Graph is the potential destination and the directed edges represent the most likely paths traffic takes to move toward the destination. We also proposed an efficient algorithm called SmartEdge to discover the most likely nonoverlapping G-Graphs in the given spatial field. However, it is challenging to perform a systematic performance study of the proposed algorithm, due to unavailability of the ground truth of gathering events. In this article, we introduce an event simulation mechanism, which makes it possible to conduct a comprehensive performance study of the SmartEdge algorithm. We measure the quality of the detected patterns, in a systematic way, in terms of timeliness and location accuracy. The results show that, on average, the SmartEdge algorithm is able to detect patterns within a grid cell away (less than 500 meters) of the simulated events and detect patterns of the simulated events as early as 10 minutes prior to the first arrival to the gathering event.

Details DOI