Author name cluster

Rui Zheng

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.

15 papers

2 author rows

EAAI Journal 2026 Journal Article

A dual-stream foreground-aware enhancement network with spiralscan-Mamba for vision-based occupancy prediction in autonomous driving

Rui Zheng
Nannan Liu
Yanyin Guo
Chuiyi Deng
Zhuoyi Zhao
Zhiheng Liu
Junwei Li

In autonomous driving perception, conventional three-dimensional object detection uses a closed set of categories and consequently fails to respond to previously unseen objects. By contrast, occupancy prediction estimates per-voxel occupancy and semantics, delivering finer geometric detail and approximate distance while generalizing beyond known classes to support downstream planning. Despite these advantages, prevailing occupancy architectures suffer from foreground–background imbalance and peripheral feature sparsity, which hinder accurate reasoning about the more critical foreground categories for autonomous driving and distant objects. To address these limitations, we propose a dual-stream framework that combines voxel-based modeling to strengthen foreground representation with a bird’s-eye-view stream for efficient global feature extraction, employs an adaptive foreground-global information fusion module to flexibly integrate complementary information across the two streams. Additionally, we introduce a Mamba-based spiral scanning mechanism that propagates structural priors from dense central areas to peripheral regions, effectively enriching boundary features. Experimental results demonstrate that our method achieves competitive state-of-the-art performance, with notably stronger foreground-category prediction and performance robust to low-resolution inputs. In particular, the proposed framework significantly enhances the representation and prediction of critical traffic participants such as vehicles and pedestrians, thereby improving the reliability of perception for real-world autonomous driving applications. The code will be publicly available at: https: //github. com/Bryan-ZhengRui/DFEOcc.

Details DOI

AAAI Conference 2026 Conference Paper

MetaAct-RL: Training Language Models for Reasoning Through Meta-Action-Based Reinforcement Learning

Zhiheng Xi
Yuhui Wang
Yiwen Ding
Guanyu Li
Senjie Jin
Shichun Liu
Jixuan Huang
Dingwen Yang

Outcome-based reinforcement learning has made notable advances in training language models (LMs) for reasoning. However, without explicit incentives and controls, this paradigm has limitations and instability in eliciting high-quality reasoning trajectories with diverse actions—particularly for models whose pretraining lacked extensive reasoning-related data. To this end, we introduce MetaAct-RL, a new RL framework that frames LMs’ thinking as sequential decision making over meta-actions. In this framework, the model chooses and executes a high-level action at each step—such as forward reasoning, critique, or refinement—to gradually reach the correct answer. To encourage deeper exploration, richer action diversity, and to improve sampling efficiency in the RL optimization process, MetaAct-RL incorporates appropriate length-based reward and regularization, and a key-state restart mechanism. Extensive experiments across six benchmarks show that MetaAct-RL improves reasoning performance by 7.99 on Llama3.2-1B and 7.17 on Llama3.1-8B relative to vanilla RL method. Moreover, on the challenging AIME-2024, our method outperforms the vanilla RL by 7.5 with Qwen2.5-1.5B.

PDF Details DOI

AAAI Conference 2026 Conference Paper

Time-Frequency Token Advantage Clipping for Training Efficient Large Reasoning Model

Rong Bao
Bo Wang
Xiao Wang
Hongyu Li
Rui Zheng
Leszek Rutkowski
Qi Zhang
Liang Ding

Long Chain-of-Thought (CoT) reasoning enhances large reasoning models' performance but suffers from severe inefficiencies, as models often overthink simple problems or underthink complex ones. Current sequence-level optimizations, like length penalties, are too coarse-grained to distinguish core logic from verbose language, precluding the necessary token-level control for efficient reasoning CoT. To overcome these limitations, we introduce Time-Frequency token Advantage Clipping (TFAC), a novel training framework designed to build efficient large reasoning models via token-level interventions. Specifically, TFAC functions along two dimensions: 1) The Frequency Dimension: It discourages inefficient loops and encourages deeper exploration by dynamically reducing the advantage scores of high-entropy tokens that are repeatedly generated within a single reasoning path. 2) The Time Dimension: It reduces excessive overthinking of the system by establishing a historical baseline for the occurrence count of each critical token in previously successful trajectories, and clipping the advantages of tokens that exceed this baseline during training. Crucially, to preserve the model's exploratory capabilities on novel problems, this suppression mechanism is automatically disabled when no historical record of success is available. Experiments conducted on the Deepseek-Distill-32B and Qwen3-8B models show that TFAC outperforms leading baseline methods, improving performance by 2.3 and 3.1 percentage points, respectively, while simultaneously reducing inference costs by 35% and 28% in scenarios where correct answers are generated. These results validate the significant efficacy of TFAC in training large reasoning models that are both powerful and highly efficient.

PDF Details DOI

AAAI Conference 2026 Conference Paper

What Makes a Good Speech Tokenizer for LLM-Centric Speech Generation? A Systematic Study

Xiaoran Fan
Zhichao Sun
Yangfan Gao
Jingfei Xiong
Hang Yan
Yifei Cao
Jiajun Sun
Shuo Li

Speech-language models (SLMs) offer a promising path toward unifying speech and text understanding and generation. However, challenges remain in achieving effective cross-modal alignment and high-quality speech generation. In this work, we systematically investigate the role of speech tokenizer designs in LLM-centric SLMs, augmented by speech heads and speaker modeling. We compare coupled, semi-decoupled, and fully decoupled speech tokenizers under a fair SLM framework and find that decoupled tokenization significantly improves alignment and synthesis quality. To address the information density mismatch between speech and text, we introduce multi-token prediction (MTP) into SLMs, enabling each hidden state to decode multiple speech tokens. This leads to up to 12× faster decoding and a substantial drop in word error rate (from 6.07 to 3.01). Furthermore, we propose a speaker-aware generation paradigm and introduce RoleTriviaQA, a large-scale role-playing knowledge QA benchmark with diverse speaker identities. Experiments demonstrate that our methods enhance both knowledge understanding and speaker consistency.

PDF Details DOI

AAAI Conference 2025 Conference Paper

Alleviating Shifted Distribution in Human Preference Alignment through Meta-Learning

Shihan Dou
Yan Liu
Enyu Zhou
Songyang Gao
Tianlong Li
Limao Xiong
Xin Zhao
Haoxiang Jia

The capability of the reward model (RM) is crucial for the success of Reinforcement Learning from Human Feedback (RLHF) in aligning with human preferences. However, as training progresses, the output space distribution of the policy model shifts. The RM, initially trained on responses sampled from the output distribution of the early policy model, gradually loses its ability to distinguish between responses from the newly shifted distribution. This issue is further compounded when the RM, trained on a specific data distribution, struggles to generalize to examples outside of that distribution. These two issues can be united as a challenge posed by the shifted distribution of the environment. To surmount this challenge, we introduce MetaRM, a novel method leveraging meta-learning to adapt the RM to the shifted environment distribution. MetaRM optimizes the RM in an alternating way, by preserving both the preferences of the original preference pairs, as well as maximizing discrimination power over new examples of the shifted distribution. Extensive experiments demonstrate that MetaRM can iteratively enhance the performance of human preference alignment by improving the RM's capacity to identify subtle differences in samples of shifted distributions.

PDF Details DOI

ICLR Conference 2025 Conference Paper

Have the VLMs Lost Confidence? A Study of Sycophancy in VLMs

Shuo Li
Tao Ji
Xiaoran Fan
Linsheng Lu
Leyi Yang
Yuming Yang 0001
Zhiheng Xi
Rui Zheng

In the study of LLMs, sycophancy represents a prevalent hallucination that poses significant challenges to these models. Specifically, LLMs often fail to adhere to original correct responses, instead blindly agreeing with users' opinions, even when those opinions are incorrect or malicious. However, research on sycophancy in visual language models (VLMs) has been scarce. In this work, we extend the exploration of sycophancy from LLMs to VLMs, introducing the MM-SY benchmark to evaluate this phenomenon. We present evaluation results from multiple representative models, addressing the gap in sycophancy research for VLMs. To mitigate sycophancy, we propose a synthetic dataset for training and employ methods based on prompts, supervised fine-tuning, and DPO. Our experiments demonstrate that these methods effectively alleviate sycophancy in VLMs. Additionally, we probe VLMs to assess the semantic impact of sycophancy and analyze the attention distribution of visual tokens. Our findings indicate that the ability to prevent sycophancy is predominantly observed in higher layers of the model. The lack of attention to image knowledge in these higher layers may contribute to sycophancy, and enhancing image attention at high layers proves beneficial in mitigating this issue.

Details

ICLR Conference 2025 Conference Paper

RMB: Comprehensively benchmarking reward models in LLM alignment

Enyu Zhou
Guodong Zheng
Binghai Wang
Zhiheng Xi
Shihan Dou
Rong Bao
Wei Shen
Limao Xiong

Reward models (RMs) guide the alignment of large language models (LLMs), steering them toward behaviors preferred by humans. Evaluating RMs is the key to better aligning LLMs. However, the current evaluation of RMs may not directly correspond to their alignment performance due to the limited distribution of evaluation data and evaluation methods that are not closely related to alignment objectives. To address these limitations, we propose RMB, a comprehensive RM benchmark that covers over 49 real-world scenarios and includes both pairwise and Best-of-N (BoN) evaluations to better reflect the effectiveness of RMs in guiding alignment optimization. We demonstrate a positive correlation between our benchmark and the downstream alignment task performance. Based on our benchmark, we conduct extensive analysis on the state-of-the-art RMs, revealing their generalization defects that were not discovered by previous benchmarks, and highlighting the potential of generative RMs. Furthermore, we delve into open questions in reward models, specifically examining the effectiveness of majority voting for the evaluation of reward models and analyzing the impact factors of generative RMs, including the influence of evaluation criteria and instructing methods. We will release our evaluation code and datasets upon publication.

Details

EAAI Journal 2024 Journal Article

An adaptive network fusing light detection and ranging height-sliced bird’s-eye view and vision for place recognition

Rui Zheng
Zuo Jiang
Yibin Ye
Yang Ren
Hui Zeng
Junwei Li
Zhiyuan Zhang

Place recognition, a fundamental component of robotic perception, aims to identify previously visited locations within an environment. In this study, we present a novel global descriptor that uses height-sliced Bird’s Eye View (BEV) from Light Detection and Ranging (LiDAR) and vision images, to facilitate high-recall place recognition in autonomous driving field. Our descriptor generation network, incorporates an adaptive weights generation branch to learn weights of visual and LiDAR features, enhancing its adaptability to different environments. The generated descriptor exhibits excellent yaw-invariance. The entire network is trained using a self-designed quadruplet loss, which discriminates inter-class boundaries and alleviates overfitting to one particular modality. We evaluate our approach on three benchmarks derived from two public datasets and achieve optimal performance on these evaluation sets. Our approach demonstrates excellent generalization ability and efficient runtime, which are indicative of its practical viability in real-world scenarios. For those interested in applying this Artificial Intelligence contribution to engineering, the implementation of our approach can be found at: https: //github. com/Bryan-ZhengRui/LocFuse.

Details DOI

JBHI Journal 2024 Journal Article

Anatomical Prior and Inter-Slice Consistency for Semi-Supervised Vertebral Structure Detection in 3D Ultrasound Volume

Hongye Zeng
Kang Zhou
Songhan Ge
Yuchong Gao
Jianhao Zhao
Shenghua Gao
Rui Zheng

Three-dimensional (3D) ultrasound imaging technique has been applied for scoliosis assessment, but the current assessment method only uses coronal projection images and cannot illustrate the 3D deformity and vertebra rotation. The vertebra detection is essential to reveal 3D spine information, but the detection task is challenging due to complex data and limited annotations. We propose VertMatch to detect vertebral structures in 3D ultrasound volume containing a detector and classifier. The detector network finds the potential positions of structures on transverse slice globally, and then the local patches are cropped based on detected positions. The classifier is used to distinguish whether the patches contain real vertebral structures and screen the predicted positions from the detector. VertMatch utilizes unlabeled data in a semi-supervised manner, and we develop two novel techniques for semi-supervised learning: 1) anatomical prior is used to acquire high-quality pseudo labels; 2) inter-slice consistency is used to utilize more unlabeled data by inputting multiple adjacent slices. Experimental results demonstrate that VertMatch can detect vertebra accurately in ultrasound volume and outperforms state-of-the-art methods. Moreover, VertMatch is also validated in automatic spinous process angle measurement on forty subjects with scoliosis, and the results illustrate that it can be a promising approach for the 3D assessment of scoliosis.

Details DOI

NeurIPS Conference 2024 Conference Paper

DACO: Towards Application-Driven and Comprehensive Data Analysis via Code Generation

Xueqing Wu
Rui Zheng
Jingzhen Sha
Te-Lin Wu
Hanyu Zhou
Mohan Tang
Kai-Wei Chang
Nanyun Peng

Data analysis is a crucial analytical process essential for deriving insights from real-world databases. As shown in Figure 1, the need for data analysis typically arises from specific application scenarios, and requires diverse reasoning skills including mathematical reasoning, logical reasoning, and strategic reasoning. Existing work often focus on simple factual retrieval or arithmetic resolutions and thus are insufficient for addressing complex real-world queries. This work aims to propose new resources and benchmarks on this crucial yet challenging and under-explored task. Due to the prohibitively high cost of collecting expert annotations, we use large language models (LLMs) enhanced by code generation to automatically generate high-quality data analysis, which will later be refined by human annotators. We construct the DACO dataset, containing (1) 440 databases (of tabular data) collected from real-world scenarios, (2) ~2k automatically generated query-answer pairs that can serve as weak supervision for model training, and (3) a concentrated but high-quality test set with human refined annotations that serves as our main evaluation benchmark. Experiments show that while LLMs like GPT-4 exhibit promising data analysis capabilities, they are still evaluated as less helpful than human-written analysis on 58. 1% cases. Leveraging our weak supervision data, we experiment with various fine-tuning methods, including supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF). Our trained model outperforms existing baselines for table question answering, and RLHF further boosts the helpfulness of generated analysis on 58. 5% cases. Data and code are released at https: //github. com/shirley-wu/daco.

PDF Details DOI

ICLR Conference 2024 Conference Paper

Improving Generalization of Alignment with Human Preferences through Group Invariant Learning

Rui Zheng
Wei Shen
Yuan Hua
Wenbin Lai
Shihan Dou
Yuhao Zhou 0005
Zhiheng Xi
Xiao Wang 0001

The success of AI assistants based on language models (LLMs) hinges crucially on Reinforcement Learning from Human Feedback (RLHF), which enables the generation of responses more aligned with human preferences. As universal AI assistants, there's a growing expectation for them to perform consistently across various domains. However, previous work shows that Reinforcement Learning (RL) often exploits shortcuts to attain high rewards and overlooks challenging samples. This focus on quick reward gains undermines both the stability in training and the model's ability to generalize to new, unseen data. In this work, we propose a novel approach that can learn a consistent policy via RL across various data groups or domains. Given the challenges associated with acquiring group annotations, our method automatically classifies data into different groups, deliberately maximizing performance variance. Then, we optimize the policy to perform well on challenging groups. Lastly, leveraging the established groups, our approach adaptively adjusts the exploration space, allocating more learning capacity to more challenging data and preventing the model from over-optimizing on simpler data. Experimental results indicate that our approach significantly enhances training stability and model generalization.

Details

ICML Conference 2024 Conference Paper

Linear Alignment: A Closed-form Solution for Aligning Human Preferences without Tuning and Feedback

Songyang Gao
Qiming Ge
Wei Shen
Shihan Dou
Junjie Ye 0005
Xiao Wang 0001
Rui Zheng
Yicheng Zou

The success of AI assistants based on Language Models (LLMs) hinges on Reinforcement Learning from Human Feedback (RLHF) to comprehend and align with user intentions. However, traditional alignment algorithms, such as PPO, are hampered by complex annotation and training requirements. This reliance limits the applicability of RLHF and hinders the development of professional assistants tailored to diverse human preferences. In this work, we introduce Linear Alignment, a novel algorithm that aligns language models with human preferences in one single inference step, eliminating the reliance on data annotation and model training. Linear alignment incorporates a new parameterization for policy optimization under divergence constraints, which enables the extraction of optimal policy in a closed-form manner and facilitates the direct estimation of the aligned response. Extensive experiments on both general and personalized preference datasets demonstrate that linear alignment significantly enhances the performance and efficiency of LLM alignment across diverse scenarios.

Details

ICML Conference 2024 Conference Paper

Training Large Language Models for Reasoning through Reverse Curriculum Reinforcement Learning

Zhiheng Xi
Wenxiang Chen
Boyang Hong
Senjie Jin
Rui Zheng
Wei He 0024
Yiwen Ding
Shichun Liu

In this paper, we propose R $^3$: Learning R easoning through R everse Curriculum R einforcement Learning (RL), a novel method that employs only outcome supervision to achieve the benefits of process supervision for large language models. The core challenge in applying RL to complex reasoning is to identify a sequence of actions that result in positive rewards and provide appropriate supervision for optimization. Outcome supervision provides sparse rewards for final results without identifying error locations, whereas process supervision offers step-wise rewards but requires extensive manual annotation. R $^3$ overcomes these limitations by learning from correct demonstrations. Specifically, R $^3$ progressively slides the start state of reasoning from a demonstration’s end to its beginning, facilitating easier model exploration at all stages. Thus, R $^3$ establishes a step-wise curriculum, allowing outcome supervision to offer step-level signals and precisely pinpoint errors. Using Llama2-7B, our method surpasses RL baseline on eight reasoning tasks by $4. 1$ points on average. Notably, in program-based reasoning, 7B-scale models perform comparably to larger models or closed-source models with our R $^3$.

Details

YNICL Journal 2021 Journal Article

Hippocampal subfield and anterior-posterior segment volumes in patients with sporadic amyotrophic lateral sclerosis

Shuangwu Liu
Qingguo Ren
Gaolang Gong
Yuan Sun
Bing Zhao
Xiaotian Ma
Na Zhang
Suyu Zhong

Neuroimaging studies of hippocampal volumes in patients with amyotrophic lateral sclerosis (ALS) have reported inconsistent results. Our aims were to demonstrate that such discrepancies are largely due to atrophy of different regions of the hippocampus that emerge in different disease stages of ALS and to explore the existence of co-pathology in ALS patients. We used the well-validated King's clinical staging system for ALS to classify patients into different disease stages. We investigated in vivo hippocampal atrophy patterns across subfields and anterior-posterior segments in different King's stages using structural MRI in 76 ALS patients and 94 health controls (HCs). The thalamus, corticostriatal tract and perforant path were used as structural controls to compare the sequence of alterations between these structures and the hippocampal subfields. Compared with HCs, ALS patients at King's stage 1 had lower volumes in the bilateral posterior subiculum and presubiculum; ALS patients at King's stage 2 exhibited lower volumes in the bilateral posterior subiculum, left anterior presubiculum and left global hippocampus; ALS patients at King's stage 3 showed significantly lower volumes in the bilateral posterior subiculum, dentate gyrus and global hippocampus. Thalamic atrophy emerged at King's stage 3. White matter tracts remained normal in a subset of ALS patients. Our study demonstrated that the pattern of hippocampal atrophy in ALS patients varies greatly across King's stages. Future studies in ALS patients that focus on the hippocampus may help to further clarify possible co-pathologies in ALS.

Details DOI

IJCAI Conference 2020 Conference Paper

GestureDet: Real-time Student Gesture Analysis with Multi-dimensional Attention-based Detector

Rui Zheng
Fei Jiang
Ruimin Shen

Students’ gestures, hand-raising, stand-up, and sleeping, indicates the engagement of students in classrooms and partially reflects teaching quality. Therefore, fast and automatically recognizing these gestures are of great importance. Due to limited computational resources in primary and secondary schools, we propose a real-time student behavior detector based on light-weight MobileNetV2-SSD to reduce the dependency of GPUs. Firstly, we build a large-scale corpus from real schools to capture various behavior gestures. Based on such a corpus, we transfer the gesture recognition task into object detections. Secondly, we design a multi-dimensional attention-based detector, named GestureDet, for real-time and accurate gesture analysis. The multi-dimensional attention mechanisms simultaneously consider all the dimensions of the training set, aiming to pay more attention to discriminative features and samples that are important for the final performance. Specifically, the spatial attention is constructed with stacked dilated convolution layers to generate a soft and learnable mask for re-weighting foreground and background features; the channel attention introduces the context modeling and squeeze-and-excitation module to focus on discriminative features; the batch attention discriminates important samples with a new designed reweight strategy. Experimental results demonstrate the effectiveness and versatility of GestureDet, which achieves 75. 2% mAP on real student behavior dataset, and 74. 5% on public PASCAL VOC dataset at 20fps on embedding device Nvidia Jetson TX2. Code will be made publicly available.

PDF Details DOI