Author name cluster

Yiwei Li

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.

17 papers

1 author row

YNIMG Journal 2025 Journal Article

Causal evidence for the role of prefrontal theta oscillations in emotion regulation using neurofeedback training

Yiwei Li
Sijin Li
Yuyao Tang
Shuai Hao
Dandan Zhang

BACKGROUND: Prefrontal theta oscillations play a critical role in cognitive control processes that facilitate emotion regulation. However, causal evidence linking their modulation to improved emotion regulation outcomes remains limited. RESULTS: Using a double-blind, randomized controlled design, this study demonstrates that EEG neurofeedback (NF) targeting prefrontal theta oscillations significantly enhances emotion regulation abilities. Participants in the experimental group exhibited specific upregulation of prefrontal theta power during NF training, leading to reduced negative emotional ratings and diminished late positive potential (LPP) amplitudes. Mediation analysis further revealed that NF-induced enhancements in prefrontal theta power partially mediated the relationship between training efficiency and emotion regulation benefits. CONCLUSION: These findings underscore the theoretical significance of prefrontal theta oscillations as a core mechanism of cognitive control in emotional contexts. This study advances our understanding of the neural underpinnings of emotion regulation and highlights theta-NF as a promising, non-invasive intervention for enhancing emotional resilience.

Details DOI

YNIMG Journal 2025 Journal Article

Decoding cortical folding patterns in marmosets using machine learning and large language model

Yue Wu
Xuesong Gao
Zhengliang Liu
Pengcheng Wang
Zihao Wu
Yiwei Li
Tuo Zhang
Tianming Liu

Macroscale neuroimaging results have revealed significant differences in the structural and functional connectivity patterns of gyri and sulci in the primate cerebral cortex. Despite these findings, understanding these differences at the molecular level has remained challenging. This study leverages a comprehensive dataset of whole-brain in situ hybridization (ISH) data from marmosets, with updates continuing through 2024, to systematically analyze cortical folding patterns. Utilizing advanced machine learning algorithm and large language model (LLM), we identified genes with significant transcriptomic differences between concave (sulci) and convex (gyri) cortical patterns. Further, gene enrichment analysis, neural migration analysis, and axon guidance pathway analysis were employed to elucidate the molecular mechanisms underlying these structural and functional differences. Our findings provide new insights into the molecular basis of cortical folding, demonstrating the potential of LLM in enhancing our understanding of brain structural and functional connectivity.

Details DOI

NeurIPS Conference 2025 Conference Paper

Every Rollout Counts: Optimal Resource Allocation for Efficient Test-Time Scaling

Xinglin Wang
Yiwei Li
Shaoxiong Feng
Peiwen Yuan
Yueqi Zhang
Jiayi Shi
Chuyi Tan
Boyuan Pan

Test-Time Scaling (TTS) improves the performance of Large Language Models (LLMs) by using additional inference-time computation to explore multiple reasoning paths through search. Yet how to allocate a fixed rollout budget most effectively during search remains underexplored, often resulting in inefficient use of compute at test time. To bridge this gap, we formulate test-time search as a resource allocation problem and derive the optimal allocation strategy that maximizes the probability of obtaining a correct solution under a fixed rollout budget. Within this formulation, we reveal a core limitation of existing search methods: solution-level allocation tends to favor reasoning directions with more candidates, leading to theoretically suboptimal and inefficient use of compute. To address this, we propose Direction-Oriented Resource Allocation (DORA), a provably optimal method that mitigates this bias by decoupling direction quality from candidate count and allocating resources at the direction level. To demonstrate DORA’s effectiveness, we conduct extensive experiments on challenging mathematical reasoning benchmarks including MATH500, AIME2024, and AIME2025. The empirical results show that DORA consistently outperforms strong baselines with comparable computational cost, achieving state-of-the-art accuracy. We hope our findings contribute to a broader understanding of optimal TTS for LLMs.

PDF Details

YNIMG Journal 2025 Journal Article

fNIRS neurofeedback facilitates emotion regulation: Exploring individual differences over the ventrolateral prefrontal cortex

Yiwei Li
Sijin Li
Hua Li
Yuyao Tang
Dandan Zhang

The ventrolateral prefrontal cortex (VLPFC) plays a pivotal role in emotion regulation, yet the effectiveness of neurofeedback (NF) training targeting the VLPFC remains uncertain, suggesting significant individual differences in outcomes. In this study, we aimed to clarify these differences by enrolling 90 participants, randomly assigned to either an experimental group or a sham group (n = 48/42). Participants in the experimental group underwent VLPFCNF training over eight sessions across two consecutive days, while those in the sham group received random signals from functional near-infrared spectroscopy (fNIRS). To investigate individual variability, participants in the experimental group were further categorized as high or low-efficacy groups based on their training efficiency, determined by the regression slope of VLPFC activity over the sessions. Our results revealed a significant reduction in negative emotions and increased VLPFC activity during emotion regulation in the high-efficacy group, compared to both the low-efficacy group and sham group. Importantly, the benefit in emotion regulation, as reflected by decreased negativity ratings, was predicted by NF training efficiency. Furthermore, the enhancement of VLPFC activity during emotion regulation fully mediated the relationship between NF training efficiency and emotion regulation benefits. Participants with higher VLPFCNF training efficiency exhibited greater engagement of the VLPFC during emotion regulation, leading to superior emotional outcomes. Additionally, VLPFCNF training efficiency was linked to the habitual use of reappraisal strategies in daily life. This study provides novel causal evidence that VLPFCNF training can effectively enhance emotion regulation, highlighting the importance of individual differences in training outcomes. Our findings suggest that NF training targeting the VLPFC offers a promising and personalized intervention strategy for improving emotion regulation, with potential applications for treating emotional disorders. This research underscores the potential of personalized NF approaches, offering new avenues for tailored therapeutic interventions in the future.

Details DOI

JBHI Journal 2025 Journal Article

MambaSAM: A Visual Mamba-Adapted SAM Framework for Medical Image Segmentation

Pengchen Liang
Leijun Shi
Bin Pu
Renkai Wu
Jianguo Chen
Lixin Zhou
Lite Xu
Zhuangzhuang Chen

The Segment Anything Model (SAM) has shown exceptional versatility in segmentation tasks across various natural image scenarios. However, its application to medical image segmentation poses significant challenges due to the intricate anatomical details and domain-specific characteristics inherent in medical images. To address these challenges, we propose a novel VMamba adapter framework that integrates a lightweight, trainable Visual Mamba (VMamba) branch with the pre-trained SAM ViT encoder. The VMamba adapter accurately captures multi-scale contextual correlations, integrates global and local information, and reduces ambiguities arising from local features only. Specifically, we propose a novel cross-branch attention (CBA) mechanism to facilitate effective interaction between the SAM and VMamba branches. This mechanism enables the model to learn and adapt more efficiently to the nuances of medical images, extracting rich, complementary features that enhance its representational capacity. Beyond architectural enhancements, we streamline the segmentation workflow by eliminating the need for prompt-driven input mechanisms. This results in an autonomous prediction model that reduces manual input requirements and improves operational efficiency. In addition, our method introduces only minimal additional trainable parameters, offering an efficient solution for medical image segmentation. Extensive evaluations of four medical image datasets demonstrate that our VMamba adapter framework achieves state-of-the-art performance. Specifically, on the ACDC dataset with limited training data, our method achieves an average Dice coefficient improvement of 0. 18 and reduces the Hausdorff distance by 20. 38 mm compared to the AutoSAM.

Details DOI

NeurIPS Conference 2025 Conference Paper

Mind the Quote: Enabling Quotation-Aware Dialogue in LLMs via Plug-and-Play Modules

Yueqi Zhang
Peiwen Yuan
Yiwei Li
Shaoxiong Feng
Xinglin Wang
Jiayi Shi
Chuyi Tan
Boyuan Pan

Human–AI conversation frequently relies on quoting earlier text—“check it with the formula I just highlighted”—yet today’s large language models (LLMs) lack an explicit mechanism for locating and exploiting such spans. We formalize the challenge as span-conditioned generation, decomposing each turn into the dialogue history, a set of token-offset quotation spans, and an intent utterance. Building on this abstraction, we introduce a quotation-centric data pipeline that automatically synthesizes task-specific dialogues, verifies answer correctness through multi-stage consistency checks, and yields both a heterogeneous training corpus and the first benchmark covering five representative scenarios. To meet the benchmark’s zero-overhead and parameter-efficiency requirements, we propose QuAda, a lightweight training-based method that attaches two bottleneck projections to every attention head, dynamically amplifying or suppressing attention to quoted spans at inference time while leaving the prompt unchanged and updating < 2. 8% of backbone weights. Experiments across models show that QuAda is suitable for all scenarios and generalizes to unseen topics, offering an effective, plug-and-play solution for quotation-aware dialogue.

PDF Details

AAAI Conference 2025 Conference Paper

SceneX: Procedural Controllable Large-Scale Scene Generation

Mengqi Zhou
Yuxi Wang
Jun Hou
Shougao Zhang
Yiwei Li
Chuanchen Luo
Junran Peng
Zhaoxiang Zhang

Developing comprehensive explicit world models is crucial for understanding and simulating real-world scenarios. Recently, Procedural Controllable Generation (PCG) has gained significant attention in large-scale scene generation by enabling the creation of scalable, high-quality assets. However, PCG faces challenges such as limited modular diversity, high expertise requirements, and challenges in managing the diverse elements and structures in complex scenes. In this paper, we introduce a large-scale scene generation framework, SceneX, which can automatically produce high-quality procedural models according to designers' textual descriptions. Specifically, the proposed method comprises two components, PCGHub and PCGPlanner. The former encompasses an extensive collection of accessible procedural assets and thousands of hand-craft API documents to perform as a standard protocol for PCG controller. The latter aims to generate executable actions for Blender to produce controllable and precise 3D assets guided by the user's instructions. Extensive experiments demonstrated the capability of our method in controllable large-scale scene generation, including nature scenes and unbounded cities, as well as scene editing such as asset placement and season translation.

PDF Details DOI

NeurIPS Conference 2025 Conference Paper

Silencer: From Discovery to Mitigation of Self-Bias in LLM-as-Benchmark-Generator

Peiwen Yuan
Yiwei Li
Shaoxiong Feng
Xinglin Wang
Yueqi Zhang
Jiayi Shi
Chuyi Tan
Boyuan Pan

LLM-as-Benchmark-Generator methods have been widely studied as a supplement to human annotators for scalable evaluation, while the potential biases within this paradigm remain underexplored. In this work, we systematically define and validate the phenomenon of inflated performance in models evaluated on their self-generated benchmarks, referred to as self-bias, and attribute it to sub-biases arising from question domain, language style, and wrong labels. On this basis, we propose Silencer, a general framework that leverages the heterogeneity between multiple generators at both the sample and benchmark levels to neutralize bias and generate high-quality, self-bias-silenced benchmark. Experimental results across various settings demonstrate that Silencer can suppress self-bias to near zero, significantly improve evaluation effectiveness of the generated benchmark (with an average improvement from 0. 655 to 0. 833 in Pearson correlation with high-quality human-annotated benchmark), while also exhibiting strong generalizability.

PDF Details

NeurIPS Conference 2025 Conference Paper

Stitch and Tell: A Structured Data Augmentation Method for Spatial Understanding

Yin Hang
Xiaomin He
Peiwen Yuan
Yiwei Li
Jiayi Shi
Wenxiao Fan
Shaoxiong Feng

Existing vision-language models often suffer from spatial hallucinations, i. e. , generating incorrect descriptions about the relative positions of objects in an image. We argue that this problem mainly stems from the asymmetric properties between images and text. To enrich the spatial understanding ability of vision-language models, we propose a simple, annotation-free, plug-and-play method named Stitch and Tell (abbreviated as SiTe), which injects structured spatial supervision into multimodal data. It constructs stitched image–text pairs by stitching images along a spatial axis and generating spatially-aware captions or question answer pairs based on the layout of stitched image, without relying on costly advanced models or human involvement. We evaluate SiTe across three architectures including LLaVA-v1. 5-7B, LLaVA-Qwen2-1. 5B and HALVA-7B, two training datasets, and thirteen benchmarks. Experiments show that SiTe improves spatial understanding tasks such as $\text{MME}_{\text{Position}}$ (+5. 50\%) and Spatial-MM (+4. 19\%), while maintaining or improving performance on general vision-language benchmarks. Our findings suggest that explicitly injecting spatially-aware structure into training data offers an effective way to mitigate spatial hallucinations and improve spatial understanding, while preserving general vision-language capabilities.

PDF Details

YNIMG Journal 2024 Journal Article

Enhancing ventrolateral prefrontal cortex activation mitigates social pain and modifies subsequent social attitudes: Insights from TMS and fMRI

Sijin Li
Xueying Cao
Yiwei Li
Yuyao Tang
Si Cheng
Dandan Zhang

Social pain, a multifaceted emotional response triggered by interpersonal rejection or criticism, profoundly impacts mental well-being and social interactions. While prior research has implicated the right ventrolateral prefrontal cortex (rVLPFC) in mitigating social pain, the precise neural mechanisms and downstream effects on subsequent social attitudes remain elusive. This study employed transcranial magnetic stimulation (TMS) integrated with fMRI recordings during a social pain task to elucidate these aspects. Eighty participants underwent either active TMS targeting the rVLPFC (n = 41) or control stimulation at the vertex (n = 39). Our results revealed that TMS-induced rVLPFC facilitation significantly reduced self-reported social pain, confirming the causal role of the rVLPFC in social pain relief. Functional connectivity analyses demonstrated enhanced interactions between the rVLPFC and the dorsolateral prefrontal cortex, emphasizing the collaborative engagement of prefrontal regions in emotion regulation. Significantly, we observed that negative social feedback led to negative social attitudes, whereas rVLPFC activation countered this detrimental effect, showcasing the potential of the rVLPFC as a protective buffer against adverse social interactions. Moreover, our study uncovered the impact role of the hippocampus in subsequent social attitudes, a relationship particularly pronounced during excitatory TMS over the rVLPFC. These findings offer promising avenues for improving mental health within the intricate dynamics of social interactions. By advancing our comprehension of the neural mechanisms underlying social pain relief, this research introduces novel intervention strategies for individuals grappling with social distress. Empowering individuals to modulate rVLPFC activation may facilitate reshaping social attitudes and successful reintegration into communal life.

Details DOI

NeurIPS Conference 2024 Conference Paper

Eye-gaze Guided Multi-modal Alignment for Medical Representation Learning

Chong Ma
Hanqi Jiang
Wenting Chen
Yiwei Li
Zihao Wu
Xiaowei Yu
Zhengliang Liu
Lei Guo

In the medical multi-modal frameworks, the alignment of cross-modality features presents a significant challenge. However, existing works have learned features that are implicitly aligned from the data, without considering the explicit relationships in the medical context. This data-reliance may lead to low generalization of the learned alignment relationships. In this work, we propose the Eye-gaze Guided Multi-modal Alignment (EGMA) framework to harness eye-gaze data for better alignment of medical visual and textual features. We explore the natural auxiliary role of radiologists' eye-gaze data in aligning medical images and text, and introduce a novel approach by using eye-gaze data, collected synchronously by radiologists during diagnostic evaluations. We conduct downstream tasks of image classification and image-text retrieval on four medical datasets, where EGMA achieved state-of-the-art performance and stronger generalization across different datasets. Additionally, we explore the impact of varying amounts of eye-gaze data on model performance, highlighting the feasibility and utility of integrating this auxiliary data into multi-modal alignment framework.

PDF Details DOI

NeurIPS Conference 2024 Conference Paper

Instruction Embedding: Latent Representations of Instructions Towards Task Identification

Yiwei Li
Jiayi Shi
Shaoxiong Feng
Peiwen Yuan
Xinglin Wang
Boyuan Pan
Heda Wang
Yao Hu

Instruction data is crucial for improving the capability of Large Language Models (LLMs) to align with human-level performance. Recent research LIMA demonstrates that alignment is essentially a process where the model adapts instructions' interaction style or format to solve various tasks, leveraging pre-trained knowledge and skills. Therefore, for instructional data, the most important aspect is the task it represents, rather than the specific semantics and knowledge information. The latent representations of instructions play roles for some instruction-related tasks like data selection and demonstrations retrieval. However, they are always derived from text embeddings, encompass overall semantic information that influences the representation of task categories. In this work, we introduce a new concept, instruction embedding, and construct Instruction Embedding Benchmark (IEB) for its training and evaluation. Then, we propose a baseline Prompt-based Instruction Embedding (PIE) method to make the representations more attention on tasks. The evaluation of PIE, alongside other embedding methods on IEB with two designed tasks, demonstrates its superior performance in accurately identifying task categories. Moreover, the application of instruction embeddings in four downstream tasks showcases its effectiveness and suitability for instruction-related tasks.

PDF Details DOI

AAAI Conference 2024 Conference Paper

Turning Dust into Gold: Distilling Complex Reasoning Capabilities from LLMs by Leveraging Negative Data

Yiwei Li
Peiwen Yuan
Shaoxiong Feng
Boyuan Pan
Bin Sun
Xinglin Wang
Heda Wang
Kan Li

Large Language Models (LLMs) have performed well on various reasoning tasks, but their inaccessibility and numerous parameters hinder wide application in practice. One promising way is distilling the reasoning ability from LLMs to small models by the generated chain-of-thought reasoning paths. In some cases, however, LLMs may produce incorrect reasoning chains, especially when facing complex mathematical problems. Previous studies only transfer knowledge from positive samples and drop the synthesized data with wrong answers. In this work, we illustrate the merit of negative data and propose a model specialization framework to distill LLMs with negative samples besides positive ones. The framework consists of three progressive steps, covering from training to inference stages, to absorb knowledge from negative data. We conduct extensive experiments across arithmetic reasoning tasks to demonstrate the role of negative data in distillation from LLM.

PDF Details DOI

NeurIPS Conference 2023 Conference Paper

Better Correlation and Robustness: A Distribution-Balanced Self-Supervised Learning Framework for Automatic Dialogue Evaluation

Peiwen Yuan
Xinglin Wang
Jiayi Shi
Bin Sun
Yiwei Li

Turn-level dialogue evaluation models (TDEMs), using self-supervised learning (SSL) framework, have achieved state-of-the-art performance in open-domain dialogue evaluation. However, these models inevitably face two potential problems. First, they have low correlations with humans on medium coherence samples as the SSL framework often brings training data with unbalanced coherence distribution. Second, the SSL framework leads TDEM to nonuniform score distribution. There is a danger that the nonuniform score distribution will weaken the robustness of TDEM through our theoretical analysis. To tackle these problems, we propose Better Correlation and Robustness (BCR), a distribution-balanced self-supervised learning framework for TDEM. Given a dialogue dataset, BCR offers an effective training set reconstructing method to provide coherence-balanced training signals and further facilitate balanced evaluating abilities of TDEM. To get a uniform score distribution, a novel loss function is proposed, which can adjust adaptively according to the uniformity of score distribution estimated by kernel density estimation. Comprehensive experiments on 17 benchmark datasets show that vanilla BERT-base using BCR outperforms SOTA methods significantly by 11. 3% on average. BCR also demonstrates strong generalization ability as it can lead multiple SOTA methods to attain better correlation and robustness.

PDF Details

AAAI Conference 2023 Conference Paper

Heterogeneous-Branch Collaborative Learning for Dialogue Generation

Yiwei Li
Shaoxiong Feng
Bin Sun
Kan Li

With the development of deep learning, advanced dialogue generation methods usually require a greater amount of computational resources. One promising approach to obtaining a high-performance and lightweight model is knowledge distillation, which relies heavily on the pre-trained powerful teacher. Collaborative learning, also known as online knowledge distillation, is an effective way to conduct one-stage group distillation in the absence of a well-trained large teacher model. However, previous work has a severe branch homogeneity problem due to the same training objective and the independent identical training sets. To alleviate this problem, we consider the dialogue attributes in the training of network branches. Each branch learns the attribute-related features based on the selected subset. Furthermore, we propose a dual group-based knowledge distillation method, consisting of positive distillation and negative distillation, to further diversify the features of different branches in a steadily and interpretable way. The proposed approach significantly improves branch heterogeneity and outperforms state-of-the-art collaborative learning methods on two widely used open-domain dialogue datasets.

PDF Details DOI

AAAI Conference 2023 Conference Paper

Towards Diverse, Relevant and Coherent Open-Domain Dialogue Generation via Hybrid Latent Variables

Bin Sun
Yitong Li
Fei Mi
Weichao Wang
Yiwei Li
Kan Li

Conditional variational models, using either continuous or discrete latent variables, are powerful for open-domain dialogue response generation. However, previous works show that continuous latent variables tend to reduce the coherence of generated responses. In this paper, we also found that discrete latent variables have difficulty capturing more diverse expressions. To tackle these problems, we combine the merits of both continuous and discrete latent variables and propose a Hybrid Latent Variable (HLV) method. Specifically, HLV constrains the global semantics of responses through discrete latent variables and enriches responses with continuous latent variables. Thus, we diversify the generated responses while maintaining relevance and coherence. In addition, we propose Conditional Hybrid Variational Transformer (CHVT) to construct and to utilize HLV with transformers for dialogue generation. Through fine-grained symbolic-level semantic information and additive Gaussian mixing, we construct the distribution of continuous variables, prompting the generation of diverse expressions. Meanwhile, to maintain the relevance and coherence, the discrete latent variable is optimized by self-separation training. Experimental results on two dialogue generation datasets (DailyDialog and Opensubtitles) show that CHVT is superior to traditional transformer-based variational mechanism w.r.t. diversity, relevance and coherence metrics. Moreover, we also demonstrate the benefit of applying HLV to fine-tuning two pre-trained dialogue models (PLATO and BART-base).

PDF Details DOI

AIIM Journal 2021 Journal Article

NIA-Network: Towards improving lung CT infection detection for COVID-19 diagnosis

Wei Li
Jinlin Chen
Ping Chen
Lequan Yu
Xiaohui Cui
Yiwei Li
Fang Cheng
Wen Ouyang

During pandemics (e. g. , COVID-19) physicians have to focus on diagnosing and treating patients, which often results in that only a limited amount of labeled CT images is available. Although recent semi-supervised learning algorithms may alleviate the problem of annotation scarcity, limited real-world CT images still cause those algorithms producing inaccurate detection results, especially in real-world COVID-19 cases. Existing models often cannot detect the small infected regions in COVID-19 CT images, such a challenge implicitly causes that many patients with minor symptoms are misdiagnosed and develop more severe symptoms, causing a higher mortality. In this paper, we propose a new method to address this challenge. Not only can we detect severe cases, but also detect minor symptoms using real-world COVID-19 CT images in which the source domain only includes limited labeled CT images but the target domain has a lot of unlabeled CT images. Specifically, we adopt Network-in-Network and Instance Normalization to build a new module (we term it NI module) and extract discriminative representations from CT images from both source and target domains. A domain classifier is utilized to implement infected region adaptation from source domain to target domain in an Adversarial Learning manner, and learns domain-invariant region proposal network (RPN) in the Faster R-CNN model. We call our model NIA-Network (Network-in-Network, Instance Normalization and Adversarial Learning), and conduct extensive experiments on two COVID-19 datasets to validate our approach. The experimental results show that our model can effectively detect infected regions with different sizes and achieve the highest diagnostic accuracy compared with existing SOTA methods.

Details DOI