Lei Lin

Tianwen Wei, Liang Zhao, Lichang Zhang et al.

In this technical report, we present Skywork-13B, a family of large language models (LLMs) trained on a corpus of over 3.2 trillion tokens drawn from both English and Chinese texts. This bilingual foundation model is the most extensively trained and openly published LLMs of comparable size to date. We introduce a two-stage training methodology using a segmented corpus, targeting general purpose training and then domain-specific enhancement training, respectively. We show that our model not only excels on popular benchmarks, but also achieves \emph{state of the art} performance in Chinese language modeling on diverse domains. Furthermore, we propose a novel leakage detection method, demonstrating that test data contamination is a pressing issue warranting further investigation by the LLM community. To spur future research, we release Skywork-13B along with checkpoints obtained during intermediate stages of the training process. We are also releasing part of our SkyPile corpus, a collection of over 150 billion tokens of web text, which is the largest high quality open Chinese pre-training corpus to date. We hope Skywork-13B and our open corpus will serve as a valuable open-source resource to democratize access to high-quality LLMs.

Boyang Zhang, SuPing Wu, Hu Cao et al.

In this paper, we present a spatio-temporal tendency reasoning (STR) network for recovering human body pose and shape from videos. Previous approaches have focused on how to extend 3D human datasets and temporal-based learning to promote accuracy and temporal smoothing. Different from them, our STR aims to learn accurate and natural motion sequences in an unconstrained environment through temporal and spatial tendency and to fully excavate the spatio-temporal features of existing video data. To this end, our STR learns the representation of features in the temporal and spatial dimensions respectively, to concentrate on a more robust representation of spatio-temporal features. More specifically, for efficient temporal modeling, we first propose a temporal tendency reasoning (TTR) module. TTR constructs a time-dimensional hierarchical residual connection representation within a video sequence to effectively reason temporal sequences' tendencies and retain effective dissemination of human information. Meanwhile, for enhancing the spatial representation, we design a spatial tendency enhancing (STE) module to further learns to excite spatially time-frequency domain sensitive features in human motion information representations. Finally, we introduce integration strategies to integrate and refine the spatio-temporal feature representations. Extensive experimental findings on large-scale publically available datasets reveal that our STR remains competitive with the state-of-the-art on three datasets. Our code are available at https://github.com/Changboyang/STR.git.

Yan Gong, Xinyu Zhang, Hao Liu et al.

Multi-modal fusion is increasingly being used for autonomous driving tasks, as different modalities provide unique information for feature extraction. However, the existing two-stream networks are only fused at a specific network layer, which requires a lot of manual attempts to set up. As the CNN goes deeper, the two modal features become more and more advanced and abstract, and the fusion occurs at the feature level with a large gap, which can easily hurt the performance. To reduce the loss of height and depth information during the process of projecting point clouds into 2D space, we utilize calibration parameters to project the point cloud into Altitude Difference Images (ADIs), which exhibit more distinct road features. In this study, we propose a novel fusion architecture called Skip-cross Networks (SkipcrossNets), which combine adaptively ADIs and camera images without being bound to a certain fusion epoch. Specifically, skip-cross fusion strategy connects each layer to each layer in a feed-forward manner, and for each layer, the feature maps of all previous layers are used as input and its own feature maps are used as input to all subsequent layers for the other modality, enhancing feature propagation and multi-modal features fusion. This strategy facilitates selection of the most similar feature layers from two modalities, enhancing feature reuse and providing complementary effects for sparse point cloud features. The advantages of skip-cross fusion strategy is demonstrated through application to the KITTI and A2D2 datasets, achieving a MaxF score of 96.85% on KITTI and an F1 score of 84.84% on A2D2. The model parameters require only 2.33 MB of memory at a speed of 68.24 FPS, which can be viable for mobile terminals and embedded devices.

Lei Lin, Jiayi Fu, Pengli Liu et al.

Although chain-of-thought (CoT) prompting combined with language models has achieved encouraging results on complex reasoning tasks, the naive greedy decoding used in CoT prompting usually causes the repetitiveness and local optimality. To address this shortcoming, ensemble-optimization tries to obtain multiple reasoning paths to get the final answer assembly. However, current ensemble-optimization methods either simply employ rule-based post-processing such as \textit{self-consistency}, or train an additional model based on several task-related human annotations to select the best one among multiple reasoning paths, yet fail to generalize to realistic settings where the type of input questions is unknown or the answer format of reasoning paths is unknown. To avoid their limitations, we propose \textbf{Self-Agreement}, a generalizable ensemble-optimization method applying in almost all scenarios where the type of input questions and the answer format of reasoning paths may be known or unknown. Self-agreement firstly samples from language model's decoder to generate a \textit{diverse} set of reasoning paths, and subsequently prompts the language model \textit{one more time} to determine the optimal answer by selecting the most \textit{agreed} answer among the sampled reasoning paths. Self-agreement simultaneously achieves remarkable performance on six public reasoning benchmarks and superior generalization capabilities.

Yafang Zheng, Lei Lin, Shuangtao Li et al.

Existing neural models are demonstrated to struggle with compositional generalization (CG), i.e., the ability to systematically generalize to unseen compositions of seen components. A key reason for failure on CG is that the syntactic and semantic representations of sequences in both the uppermost layer of the encoder and decoder are entangled. However, previous work concentrates on separating the learning of syntax and semantics instead of exploring the reasons behind the representation entanglement (RE) problem to solve it. We explain why it exists by analyzing the representation evolving mechanism from the bottom to the top of the Transformer layers. We find that the ``shallow'' residual connections within each layer fail to fuse previous layers' information effectively, leading to information forgetting between layers and further the RE problems. Inspired by this, we propose LRF, a novel \textbf{L}ayer-wise \textbf{R}epresentation \textbf{F}usion framework for CG, which learns to fuse previous layers' information back into the encoding and decoding process effectively through introducing a \emph{fuse-attention module} at each encoder and decoder layer. LRF achieves promising results on two realistic benchmarks, empirically demonstrating the effectiveness of our proposal.

Jiayi Fu, Lei Lin, Xiaoyang Gao et al.

Recent advancements in large language models (LLMs) have demonstrated remarkable abilities in handling a variety of natural language processing (NLP) downstream tasks, even on mathematical tasks requiring multi-step reasoning. In this report, we introduce the KwaiYiiMath which enhances the mathematical reasoning abilities of KwaiYiiBase1, by applying Supervised Fine-Tuning (SFT) and Reinforced Learning from Human Feedback (RLHF), including on both English and Chinese mathematical tasks. Meanwhile, we also constructed a small-scale Chinese primary school mathematics test set (named KMath), consisting of 188 examples to evaluate the correctness of the problem-solving process generated by the models. Empirical studies demonstrate that KwaiYiiMath can achieve state-of-the-art (SOTA) performance on GSM8k, CMath, and KMath compared with the similar size models, respectively.

Lei Lin, Shuangtao Li, Xiaodong Shi

Simultaneous machine translation, which aims at a real-time translation, is useful in many live scenarios but very challenging due to the trade-off between accuracy and latency. To achieve the balance for both, the model needs to wait for appropriate streaming text (READ policy) and then generates its translation (WRITE policy). However, WRITE policies of previous work either are specific to the method itself due to the end-to-end training or suffer from the input mismatch between training and decoding for the non-end-to-end training. Therefore, it is essential to learn a generic and better WRITE policy for simultaneous machine translation. Inspired by strategies utilized by human interpreters and "wait" policies, we propose a novel adaptive prefix-to-prefix training policy called LEAPT, which allows our machine translation model to learn how to translate source sentence prefixes and make use of the future context. Experiments show that our proposed methods greatly outperform competitive baselines and achieve promising results.

Lei Lin, Jizhao Zhu, Qirui Tang et al.

In social media networks, users produce a large amount of text content anytime, providing researchers with an invaluable approach to digging for personality-related information. Personality detection based on user-generated text is a method with broad application prospects, such as for constructing user portraits. The presence of significant noise in social media texts hinders personality detection. However, previous studies have not delved deeper into addressing this challenge. Inspired by the scanning reading technique, we propose an attention-based information extraction mechanism (AIEM) for long texts, which is applied to quickly locate valuable pieces of text, and fully integrate beneficial semantic information. Then, we provide a novel attention-based denoising framework (ADF) for personality detection tasks and achieve state-of-the-art performance on two commonly used datasets. Notably, we obtain an average accuracy improvement of 10.2% on the gold standard Twitter-Myers-Briggs Type Indicator (Twitter-MBTI) dataset. We made our code publicly available on GitHub\footnote{https://github.com/Once2gain/PersonalityDetection}. We shed light on how AIEM works to magnify personality-related signals through a case study.

Lei Lin, Jizhao Zhu, Yong Liu et al.

This paper addresses two limitations of large language models (LLMs) in solving complex problems: (1) their reasoning processes exhibit Bayesian-like stochastic generation, where each token is sampled from a context-dependent probability distribution, leading to inherently random decision trajectories rather than deterministic planning; (2) the reasoning and decision-making mechanisms are statically decoupled, meaning dynamically retrieved domain knowledge fails to dynamically adjust the underlying reasoning strategy. These dual deficiencies result in initial decisions lacking strategic anchoring and reasoning chains often failing to converge on correct solutions, as stochastic generation lacks mechanisms for trajectory correction or knowledge-guided optimization during sequential reasoning. To resolve these issues, we propose a problem-solving method integrated into the LLM's generation process to guide reasoning. This method, compatible with numerous LLMs and featuring reusable solutions, is grounded in a novel Heuristic-Classification-of-Thoughts prompting schema (HCoT). HCoT synergizes the LLM's reasoning ability with a structured problem space via a heuristic classification model that controls the reasoning process and provides reusable abstract solutions. Evaluated on two complex inductive reasoning tasks with ill-defined search spaces, HCoT outperforms existing approaches (e.g., Tree-of-Thoughts and Chain-of-Thoughts prompting) in performance. On the well-structured 24 Game task, HCoT demonstrates significantly higher token efficiency compared to the state-of-the-art Tree-of-Thoughts-Breadth-First-Search. In terms of both accuracy and token usage, HCoT achieves a Pareto frontier balance, offering a strong trade-off between performance and computational cost.

Jiakang Wang, Runze Liu, Lei Lin et al.

Recent Large Language Model (LLM) post-training methods rely on token-level clipping mechanisms during Reinforcement Learning (RL). However, we identify a fundamental flaw in this Outcome-Supervised RL (OSRL) paradigm: the Importance Sampling (IS) ratios of positive-advantage tokens are mismatched, leading to unbalanced token weighting for positive and negative tokens. This mismatch suppresses the update of low-probability tokens while over-amplifying already high-probability ones. To address this, we propose Asymmetric Importance Sampling Policy Optimization (ASPO), which uses a simple yet effective strategy that flips the IS ratios of positive-advantage tokens, aligning their update direction with the learning dynamics of negative ones. AIS further incorporates a soft dual-clipping mechanism to stabilize extreme updates while maintaining gradient flow. Comprehensive experiments on coding and mathematical reasoning benchmarks demonstrate that ASPO significantly mitigates premature convergence, improves training stability, and enhances final performance over strong GRPO-based baselines. Our analysis provides new insights into the role of token-level weighting in OSRL and highlights the critical importance of correcting IS in LLM RL. The code and models of ASPO are available at https://github.com/wizard-III/Archer2.0.

Liu Yang, Haihua Yang, Wenjun Cheng et al.

Large language models (LLMs) have shown great potential to solve varieties of natural language processing (NLP) tasks, including mathematical reasoning. In this work, we present SkyMath, a large language model for mathematics with 13 billion parameters. By applying self-compare fine-tuning, we have enhanced mathematical reasoning abilities of Skywork-13B-Base remarkably. On GSM8K, SkyMath outperforms all known open-source models of similar size and has established a new SOTA performance.

Lei Lin, Shuangtao Li, Yafang Zheng et al.

Recent studies have shown that sequence-to-sequence (seq2seq) models struggle with compositional generalization (CG), i.e., the ability to systematically generalize to unseen compositions of seen components. There is mounting evidence that one of the reasons hindering CG is the representation of the encoder uppermost layer is entangled, i.e., the syntactic and semantic representations of sequences are entangled. However, we consider that the previously identified representation entanglement problem is not comprehensive enough. Additionally, we hypothesize that the source keys and values representations passing into different decoder layers are also entangled. Starting from this intuition, we propose \textsc{CompoSition} (\textbf{Compo}se \textbf{S}yntactic and Semant\textbf{i}c Representa\textbf{tion}s), an extension to seq2seq models which learns to compose representations of different encoder layers dynamically for different tasks, since recent studies reveal that the bottom layers of the Transformer encoder contain more syntactic information and the top ones contain more semantic information. Specifically, we introduce a \textit{composed layer} between the encoder and decoder to compose different encoder layers' representations to generate specific keys and values passing into different decoder layers. \textsc{CompoSition} achieves competitive results on two comprehensive and realistic benchmarks, which empirically demonstrates the effectiveness of our proposal. Codes are available at~\url{https://github.com/thinkaboutzero/COMPOSITION}.

Lei Lin, Weizi Li, Lei Zhu

Accurate prediction of network-wide traffic conditions is essential for intelligent transportation systems. In the last decade, machine learning techniques have been widely used for this task, resulting in state-of-the-art performance. We propose a novel deep learning model, Graph Convolutional Gated Recurrent Neural Network (GCGRNN), to predict network-wide, multi-step traffic volume. GCGRNN can automatically capture spatial correlations between traffic sensors and temporal dependencies in historical traffic data. We have evaluated our model using two traffic datasets extracted from 150 sensors in Los Angeles, California, at the time resolutions one hour and 15 minutes, respectively. The results show that our model outperforms the other five benchmark models in terms of prediction accuracy. For instance, our model reduces MAE by 25.3%, RMSE by 29.2%, and MAPE by 20.2%, compared to the state-of-the-art Diffusion Convolutional Recurrent Neural Network (DCRNN) model using the hourly dataset. Our model also achieves faster training than DCRNN by up to 52%. The data and implementation of GCGRNN can be found at https://github.com/leilin-research/GCGRNN.

Chongxuan Huang, Lei Lin, Xiaodong Shi et al.

Reinforcement Learning with Verifiable Rewards (RLVR) has demonstrated promising gains in enhancing the reasoning capabilities of large language models. However, its dependence on domain-specific verifiers significantly restricts its applicability to open and general domains. Recent efforts such as RLPR have extended RLVR to general domains, enabling training on broader datasets and achieving improvements over RLVR. However, a notable limitation of these methods is their tendency to overfit to reference answers, which constrains the model's ability to generate diverse outputs. This limitation is particularly pronounced in open-ended tasks such as writing, where multiple plausible answers exist. To address this, we propose DARL, a simple yet effective reinforcement learning framework that encourages the generation of diverse answers within a controlled deviation range from the reference while preserving alignment with it. Our framework is fully compatible with existing general reinforcement learning methods and can be seamlessly integrated without additional verifiers. Extensive experiments on thirteen benchmarks demonstrate consistent improvements in reasoning performance. Notably, DARL surpasses RLPR, achieving average gains of 1.3 points on six reasoning benchmarks and 9.5 points on seven general benchmarks, highlighting its effectiveness in improving both reasoning accuracy and output diversity.

Runze Liu, Jiakang Wang, Yuling Shi et al.

Reinforcement Learning (RL) has shown remarkable success in enhancing the reasoning capabilities of Large Language Models (LLMs). Process-Supervised RL (PSRL) has emerged as a more effective paradigm compared to outcome-based RL. However, existing PSRL approaches suffer from limited exploration efficiency, both in terms of branching positions and sampling. In this paper, we introduce a novel PSRL framework (AttnRL), which enables efficient exploration for reasoning models. Motivated by preliminary observations that steps exhibiting high attention scores correlate with reasoning behaviors, we propose to branch from positions with high values. Furthermore, we develop an adaptive sampling strategy that accounts for problem difficulty and historical batch size, ensuring that the whole training batch maintains non-zero advantage values. To further improve sampling efficiency, we design a one-step off-policy training pipeline for PSRL. Extensive experiments on multiple challenging mathematical reasoning benchmarks demonstrate that our method consistently outperforms prior approaches in terms of performance and sampling and training efficiency.

Zhongkai Shangguan, Zihe Zheng, Lei Lin

Nowadays social media platforms such as Twitter provide a great opportunity to understand public opinion of climate change compared to traditional survey methods. In this paper, we constructed a massive climate change Twitter dataset and conducted comprehensive analysis using machine learning. By conducting topic modeling and natural language processing, we show the relationship between the number of tweets about climate change and major climate events; the common topics people discuss climate change; and the trend of sentiment. Our dataset was published on Kaggle (\url{https://www.kaggle.com/leonshangguan/climate-change-tweets-ids-until-aug-2021}) and can be used in further research.

Han Sun, Lei Lin, Ningzhong Liu et al.

Recently, in order to address the unsupervised domain adaptation (UDA) problem, extensive studies have been proposed to achieve transferrable models. Among them, the most prevalent method is adversarial domain adaptation, which can shorten the distance between the source domain and the target domain. Although adversarial learning is very effective, it still leads to the instability of the network and the drawbacks of confusing category information. In this paper, we propose a Robust Ensembling Network (REN) for UDA, which applies a robust time ensembling teacher network to learn global information for domain transfer. Specifically, REN mainly includes a teacher network and a student network, which performs standard domain adaptation training and updates weights of the teacher network. In addition, we also propose a dual-network conditional adversarial loss to improve the ability of the discriminator. Finally, for the purpose of improving the basic ability of the student network, we utilize the consistency constraint to balance the error between the student network and the teacher network. Extensive experimental results on several UDA datasets have demonstrated the effectiveness of our model by comparing with other state-of-the-art UDA algorithms.

Lei Lin

It is expensive to evaluate the results of Machine Translation(MT), which usually requires manual translation as a reference. Machine Translation Quality Estimation (QE) is a task of predicting the quality of machine translations without relying on any reference. Recently, the emergence of predictor-estimator framework which trains the predictor as a feature extractor and estimator as a QE predictor, and pre-trained language models(PLM) have achieved promising QE performance. However, we argue that there are still gaps between the predictor and the estimator in both data quality and training objectives, which preclude QE models from benefiting from a large number of parallel corpora more directly. Based on previous related work that have alleviated gaps to some extent, we propose a novel framework that provides a more accurate direct pretraining for QE tasks. In this framework, a generator is trained to produce pseudo data that is closer to the real QE data, and a estimator is pretrained on these data with novel objectives that are the same as the QE task. Experiments on widely used benchmarks show that our proposed framework outperforms existing methods, without using any pretraining models such as BERT.

Zhongkai Shangguan, Lei Lin, Wencheng Wu et al.

There are a large number of optimization problems in physical models where the relationships between model parameters and outputs are unknown or hard to track. These models are named as black-box models in general because they can only be viewed in terms of inputs and outputs, without knowledge of the internal workings. Optimizing the black-box model parameters has become increasingly expensive and time consuming as they have become more complex. Hence, developing effective and efficient black-box model optimization algorithms has become an important task. One powerful algorithm to solve such problem is Bayesian optimization, which can effectively estimates the model parameters that lead to the best performance, and Gaussian Process (GP) has been one of the most widely used surrogate model in Bayesian optimization. However, the time complexity of GP scales cubically with respect to the number of observed model outputs, and GP does not scale well with large parameter dimension either. Consequently, it has been challenging for GP to optimize black-box models that need to query many observations and/or have many parameters. To overcome the drawbacks of GP, in this study, we propose a general Bayesian optimization algorithm that employs a Neural Process (NP) as the surrogate model to perform black-box model optimization, namely, Neural Process for Bayesian Optimization (NPBO). In order to validate the benefits of NPBO, we compare NPBO with four benchmark approaches on a power system parameter optimization problem and a series of seven benchmark Bayesian optimization problems. The results show that the proposed NPBO performs better than the other four benchmark approaches on the power system parameter optimization problem and competitively on the seven benchmark problems.

Trevor W. Richardson, Wencheng Wu, Lei Lin et al.

We consider the topic of data imputation, a foundational task in machine learning that addresses issues with missing data. To that end, we propose MCFlow, a deep framework for imputation that leverages normalizing flow generative models and Monte Carlo sampling. We address the causality dilemma that arises when training models with incomplete data by introducing an iterative learning scheme which alternately updates the density estimate and the values of the missing entries in the training data. We provide extensive empirical validation of the effectiveness of the proposed method on standard multivariate and image datasets, and benchmark its performance against state-of-the-art alternatives. We demonstrate that MCFlow is superior to competing methods in terms of the quality of the imputed data, as well as with regards to its ability to preserve the semantic structure of the data.

Tao Li, Lei Lin

With billions of personal images being generated from social media and cameras of all sorts on a daily basis, security and privacy are unprecedentedly challenged. Although extensive attempts have been made, existing face image de-identification techniques are either insufficient in photo-reality or incapable of balancing privacy and usability qualitatively and quantitatively, i.e., they fail to answer counterfactual questions such as "is it private now?", "how private is it?", and "can it be more private?" In this paper, we propose a novel framework called AnonymousNet, with an effort to address these issues systematically, balance usability, and enhance privacy in a natural and measurable manner. The framework encompasses four stages: facial attribute estimation, privacy-metric-oriented face obfuscation, directed natural image synthesis, and adversarial perturbation. Not only do we achieve the state-of-the-arts in terms of image quality and attribute prediction accuracy, we are also the first to show that facial privacy is measurable, can be factorized, and accordingly be manipulated in a photo-realistic fashion to fulfill different requirements and application scenarios. Experiments further demonstrate the effectiveness of the proposed framework.

Lei Lin, Beilei Xu, Wencheng Wu et al.

Myotonia, which refers to delayed muscle relaxation after contraction, is the main symptom of myotonic dystrophy patients. We propose a hierarchical attention-based temporal convolutional network (HA-TCN) for myotonic dystrohpy diagnosis from handgrip time series data, and introduce mechanisms that enable model explainability. We compare the performance of the HA-TCN model against that of benchmark TCN models, LSTM models with and without attention mechanisms, and SVM approaches with handcrafted features. In terms of classification accuracy and F1 score, we found all deep learning models have similar levels of performance, and they all outperform SVM. Further, the HA-TCN model outperforms its TCN counterpart with regards to computational efficiency regardless of network depth, and in terms of performance particularly when the number of hidden layers is small. Lastly, HA-TCN models can consistently identify relevant time series segments in the relaxation phase of the handgrip time series, and exhibit increased robustness to noise when compared to attention-based LSTM models.

Tao Li, Minsoo Choi, Kaiming Fu et al.

Recommendation systems that automatically generate personalized music playlists for users have attracted tremendous attention in recent years. Nowadays, most music recommendation systems rely on item-based or user-based collaborative filtering or content-based approaches. In this paper, we propose a novel mixture hidden Markov model (HMM) for music play sequence prediction. We compare the mixture model with state-of-the-art methods and evaluate the predictions quantitatively and qualitatively on a large-scale real-world dataset in a Kaggle competition. Results show that our model significantly outperforms traditional methods as well as other competitors. We conclude by envisioning a next-generation music recommendation system that integrates our model with recent advances in deep learning, computer vision, and speech techniques, and has promising potential in both academia and industry.

Tao Li, Lei Lin, Minsoo Choi et al.

With one billion monthly viewers, and millions of users discussing and sharing opinions, comments below YouTube videos are rich sources of data for opinion mining and sentiment analysis. We introduce the YouTube AV 50K dataset, a freely-available collections of more than 50,000 YouTube comments and metadata below autonomous vehicle (AV)-related videos. We describe its creation process, its content and data format, and discuss its possible usages. Especially, we do a case study of the first self-driving car fatality to evaluate the dataset, and show how we can use this dataset to better understand public attitudes toward self-driving cars and public reactions to the accident. Future developments of the dataset are also discussed.

Lin Liu, Chunyuan Li, Yongfu Li et al.

Vehicle-to-vehicle communications can change the driving behavior of drivers significantly by providing them rich information on downstream traffic flow conditions. This study seeks to model the varying car-following behaviors involving connected vehicles and human-driving vehicles in mixed traffic flow. A revised car-following model is developed using an intelligent driver model (IDM) to capture drivers' perceptions of their preceding traffic conditions through vehicle-to-vehicle communications. Stability analysis of the mixed traffic flow is conducted for a specific case. Numerical results show that the stable region is apparently enlarged compared with the IDM.

Lei Lin, Zhengbing He, Srinivas Peeta

This study proposes a novel Graph Convolutional Neural Network with Data-driven Graph Filter (GCNN-DDGF) model that can learn hidden heterogeneous pairwise correlations between stations to predict station-level hourly demand in a large-scale bike-sharing network. Two architectures of the GCNN-DDGF model are explored; GCNNreg-DDGF is a regular GCNN-DDGF model which contains the convolution and feedforward blocks, and GCNNrec-DDGF additionally contains a recurrent block from the Long Short-term Memory neural network architecture to capture temporal dependencies in the bike-sharing demand series. Furthermore, four types of GCNN models are proposed whose adjacency matrices are based on various bike-sharing system data, including Spatial Distance matrix (SD), Demand matrix (DE), Average Trip Duration matrix (ATD), and Demand Correlation matrix (DC). These six types of GCNN models and seven other benchmark models are built and compared on a Citi Bike dataset from New York City which includes 272 stations and over 28 million transactions from 2013 to 2016. Results show that the GCNNrec-DDGF performs the best in terms of the Root Mean Square Error, the Mean Absolute Error and the coefficient of determination (R2), followed by the GCNNreg-DDGF. They outperform the other models. Through a more detailed graph network analysis based on the learned DDGF, insights are obtained on the black box of the GCNN-DDGF model. It is found to capture some information similar to details embedded in the SD, DE and DC matrices. More importantly, it also uncovers hidden heterogeneous pairwise correlations between stations that are not revealed by any of those matrices.

Lei Lin, John Handley, Adel Sadek

Short-term traffic volume prediction models have been extensively studied in the past few decades. However, most of the previous studies only focus on single-value prediction. Considering the uncertain and chaotic nature of the transportation system, an accurate and reliable prediction interval with upper and lower bounds may be better than a single point value for transportation management. In this paper, we introduce a neural network model called Extreme Learning Machine (ELM) for interval prediction of short-term traffic volume and improve it with the heuristic particle swarm optimization algorithm (PSO). The hybrid PSO-ELM model can generate the prediction intervals under different confidence levels and guarantee the quality by minimizing a multi-objective function which considers two criteria reliability and interval sharpness. The PSO-ELM models are built based on an hourly traffic dataset and compared with ARMA and Kalman Filter models. The results show that ARMA models are the worst for all confidence levels, and the PSO-ELM models are comparable with Kalman Filter from the aspects of reliability and narrowness of the intervals, although the parameters of PSO-ELM are fixed once the training is done while Kalman Filter is updated in an online approach. Additionally, only the PSO-ELMs are able to produce intervals with coverage probabilities higher than or equal to the confidence levels. For the points outside of the prediction levels given by PSO-ELMs, they lie very close to the bounds.

Zhenhua Zhang, Lei Lin

Border crossing delays cause problems like huge economics loss and heavy environmental pollutions. To understand more about the nature of border crossing delay, this study applies a dictionary-based compression algorithm to process the historical Niagara Frontier border wait times data. It can identify the abnormal spatial-temporal patterns for both passenger vehicles and trucks at three bridges connecting US and Canada. Furthermore, it provides a quantitate anomaly score to rank the wait times patterns across the three bridges for each vehicle type and each direction. By analyzing the top three most abnormal patterns, we find that there are at least two factors contributing the anomaly of the patterns. The weekends and holidays may cause unusual heave congestions at the three bridges at the same time, and the freight transportation demand may be uneven from Canada to the USA at Peace Bridge and Lewiston-Queenston Bridge, which may lead to a high anomaly score. By calculating the frequency of the top 5% abnormal patterns by hour of the day, the results show that for cars from the USA to Canada, the frequency of abnormal waiting time patterns is the highest during noon while for trucks in the same direction, it is the highest during the afternoon peak hours. For Canada to US direction, the frequency of abnormal border wait time patterns for both cars and trucks reaches to the peak during the afternoon. The analysis of abnormal spatial-temporal wait times patterns is promising to improve the border crossing management

Lei Lin, Qian Wang, Adel W. Sadek

Traffic accident data are usually noisy, contain missing values, and heterogeneous. How to select the most important variables to improve real-time traffic accident risk prediction has become a concern of many recent studies. This paper proposes a novel variable selection method based on the Frequent Pattern tree (FP tree) algorithm. First, all the frequent patterns in the traffic accident dataset are discovered. Then for each frequent pattern, a new criterion, called the Relative Object Purity Ratio (ROPR) which we proposed, is calculated. This ROPR is added to the importance score of the variables that differentiate one frequent pattern from the others. To test the proposed method, a dataset was compiled from the traffic accidents records detected by only one detector on interstate highway I-64 in Virginia in 2005. This dataset was then linked to other variables such as real-time traffic information and weather conditions. Both the proposed method based on the FP tree algorithm, as well as the widely utilized, random forest method, were then used to identify the important variables or the Virginia dataset. The results indicate that there are some differences between the variables deemed important by the FP tree and those selected by the random forest method. Following this, two baseline models (i.e. a nearest neighbor (k-NN) method and a Bayesian network) were developed to predict accident risk based on the variables identified by both the FP tree method and the random forest method. The results show that the models based on the variable selection using the FP tree performed better than those based on the random forest method for several versions of the k-NN and Bayesian network models.The best results were derived from a Bayesian network model using variables from FP tree. That model could predict 61.11% of accidents accurately while having a false alarm rate of 38.16%.

Yang Liu, Chengjie Sun, Lei Lin et al.

In this paper, we proposed a sentence encoding-based model for recognizing text entailment. In our approach, the encoding of sentence is a two-stage process. Firstly, average pooling was used over word-level bidirectional LSTM (biLSTM) to generate a first-stage sentence representation. Secondly, attention mechanism was employed to replace average pooling on the same sentence for better representations. Instead of using target sentence to attend words in source sentence, we utilized the sentence's first-stage representation to attend words appeared in itself, which is called "Inner-Attention" in our paper . Experiments conducted on Stanford Natural Language Inference (SNLI) Corpus has proved the effectiveness of "Inner-Attention" mechanism. With less number of parameters, our model outperformed the existing best sentence encoding-based approach by a large margin.

Yaming Sun, Lei Lin, Duyu Tang et al.

We present a method to leverage radical for learning Chinese character embedding. Radical is a semantic and phonetic component of Chinese character. It plays an important role as characters with the same radical usually have similar semantic meaning and grammatical usage. However, existing Chinese processing algorithms typically regard word or character as the basic unit but ignore the crucial radical information. In this paper, we fill this gap by leveraging radical for learning continuous representation of Chinese character. We develop a dedicated neural architecture to effectively learn character embedding and apply it on Chinese character similarity judgement and Chinese word segmentation. Experiment results show that our radical-enhanced method outperforms existing embedding learning algorithms on both tasks.