Yicheng Liu

AAAI Conference 2026 Conference Paper

Do Retrieval Augmented Language Models Know When They Don’t Know?

• Youchao Zhou
• Heyan Huang
• Yicheng Liu
• Rui Dai
• Xinglin Wang
• Xingchen Zhang
• Shumin Shi
• Yang Deng

Existing large language models (LLMs) occasionally generate plausible yet factually incorrect responses, known as hallucinations. Two main approaches have been proposed to mitigate hallucinations: retrieval-augmented language models (RALMs) and refusal post-training. However, current research predominantly focuses on their individual effectiveness while overlooking the evaluation of the refusal capability of RALMs. Ideally, if RALMs know when they do not know, they should refuse to answer. In this study, we ask the fundamental question: Do RALMs know when they don’t know? Specifically, we investigate three questions. First, are RALMs well calibrated with respect to different internal and external knowledge states? We examine the influence of various factors. Contrary to expectations, when all retrieved documents are irrelevant, RALMs still tend to refuse questions they could have answered correctly. Next, given the model's pronounced over-refusal behavior, we raise a second question: How does a RALM's refusal ability align with its calibration quality? Our results show that the over-refusal problem can be mitigated through in-context fine-tuning. However, we observe that improved refusal behavior does not necessarily imply better calibration or higher overall accuracy. Finally, we ask: Can we combine refusal-aware RALMs with uncertainty-based answer abstention to mitigate over-refusal? We develop a simple yet effective refusal mechanism for refusal-post-trained RALMs that improves their overall answer quality by balancing refusal and correct answers. Our study provides a more comprehensive understanding of the factors influencing RALM behavior. Meanwhile, we emphasize that uncertainty estimation for RALMs remains an open problem deserving deeper investigation.

ICLR Conference 2025 Conference Paper

CG-Bench: Clue-grounded Question Answering Benchmark for Long Video Understanding

• Guo Chen 0006
• Yicheng Liu
• Yifei Huang 0002
• Baoqi Pei
• Jilan Xu
• Yuping He
• Tong Lu 0002
• Yali Wang 0001

The existing video understanding benchmarks for multimodal large language models (MLLMs) mainly focus on short videos. The few benchmarks for long video understanding often rely on multiple-choice questions (MCQs). Due to the limitations of MCQ evaluations and the advanced reasoning abilities of MLLMs, models can often answer correctly by combining short video insights with elimination, without truly understanding the content. To bridge this gap, we introduce CG-Bench, a benchmark for clue-grounded question answering in long videos. CG-Bench emphasizes the model's ability to retrieve relevant clues, enhancing evaluation credibility. It includes 1,219 manually curated videos organized into 14 primary, 171 secondary, and 638 tertiary categories, making it the largest benchmark for long video analysis. The dataset features 12,129 QA pairs in three question types: perception, reasoning, and hallucination. To address the limitations of MCQ-based evaluation, we develop two novel clue-based methods: clue-grounded white box and black box evaluations, assessing whether models generate answers based on accurate video understanding. We evaluated multiple closed-source and open-source MLLMs on CG-Bench. The results show that current models struggle significantly with long videos compared to short ones, and there is a notable gap between open-source and commercial models. We hope CG-Bench will drive the development of more reliable and capable MLLMs for long video comprehension.

NeurIPS Conference 2025 Conference Paper

Conditioning Matters: Training Diffusion Policies is Faster Than You Think

• Zibin Dong
• Yicheng Liu
• Yinchuan Li
• Hang Zhao
• Jianye Hao

Diffusion policies have emerged as a mainstream paradigm for building vision-language-action (VLA) models. Although they demonstrate strong robot control capabilities, their training efficiency remains suboptimal. In this work, we identify a fundamental challenge in conditional diffusion policy training: when generative conditions are hard to distinguish, the training objective degenerates into modeling the marginal action distribution, a phenomenon we term loss collapse. To overcome this, we propose Cocos, a simple yet general solution that modifies the source distribution in the conditional flow matching to be condition-dependent. By anchoring the source distribution around semantics extracted from condition inputs, Cocos encourages stronger condition integration and prevents the loss collapse. We provide theoretical justification and extensive empirical results across simulation and real-world benchmarks. Our method achieves faster convergence and higher success rates than existing approaches, matching the performance of large-scale pre-trained VLAs using significantly fewer gradient steps and parameters. Cocos is lightweight, easy to implement, and compatible with diverse policy architectures, offering a general-purpose improvement to diffusion policy training.

IJCAI Conference 2025 Conference Paper

CycSeq: Leveraging Cyclic Data Generation for Accurate Perturbation Prediction in Single-Cell RNA-Seq

• Yicheng Liu
• Sai Wu
• Tianyun Zhang
• Chang Yao
• Ning Shen

Understanding and predicting the effects of cellular perturbations using single-cell sequencing technology remains a critical and challenging problem in biotechnology. In this work, we introduce CycSeq, a deep learning framework that leverages cyclic data generation and recent advances in neural architectures to predict single-cell responses under specified perturbations across multiple cell lines, while also generating the corresponding single-cell expression profiles. Specifically, CycSeq addresses the challenge of learning heterogeneous perturbation responses from unpaired single-cell gene expression data by generating pseudo-pairs through cyclic data generation. Experimental results demonstrate that CycSeq outperforms existing methods in perturbation prediction tasks, as evaluated using computational metrics such as R-squared and MAE. Furthermore, CycSeq employs a unified architecture that integrates information from multiple cell lines, enabling robust predictions even for long-tail cell lines with limited training data. The source code is publicly available at https: //github. com/yczju/cycseq.

ICRA Conference 2025 Conference Paper

Diffusion-Based Generative Models for 3D Occupancy Prediction in Autonomous Driving

• Yunshen Wang
• Yicheng Liu
• Tianyuan Yuan
• Yucheng Mao
• Yingshi Liang
• Xiuyu Yang
• Honggang Zhang
• Hang Zhao 0021

Accurately predicting 3D occupancy grids from visual inputs is critical for autonomous driving, but current discriminative methods struggle with noisy data, incomplete observations, and the complex structures inherent in 3D scenes. In this work, we reframe 3D occupancy prediction as a generative modeling task using diffusion models, which learn the underlying data distribution and incorporate 3D scene priors. This approach enhances prediction consistency, noise robustness, and better handles the intricacies of 3D spatial structures. Our extensive experiments show that diffusion-based generative models outperform state-of-the-art discriminative approaches, delivering more realistic and accurate occupancy predictions, especially in occluded or low-visibility regions. Moreover, the improved predictions significantly benefit downstream planning tasks, highlighting the practical advantages of our method for real-world autonomous driving applications.

NeurIPS Conference 2025 Conference Paper

Eagle 2.5: Boosting Long-Context Post-Training for Frontier Vision-Language Models

• Guo Chen
• Zhiqi Li
• Shihao Wang
• Jindong Jiang
• Yicheng Liu
• Lidong Lu
• De-An Huang
• Wonmin Byeon

We introduce Eagle2. 5, a frontier vision-language model (VLM) for long-context multimodal learning. Our work addresses the challenges in long video comprehension and high-resolution image understanding, introducing a generalist framework for both tasks. The proposed training framework incorporates Automatic Degrade Sampling and Image Area Preservation, two techniques that preserve contextual integrity and visual details. The framework also includes numerous efficiency optimizations in the pipeline for long-context data training. Finally, we propose Eagle-Video-110K, a novel dataset that integrates both story-level and clip-level annotations, facilitating long-video understanding. Eagle2. 5 demonstrates substantial improvements on long-context multimodal benchmarks, providing a robust solution to the limitations of existing VLMs. Notably, our best model Eagle2. 5-8B achieves 72. 4\% on Video-MME with 512 input frames, matching the results of top-tier commercial model such as GPT-4o and large-scale open-source models like Qwen2. 5-VL-72B and InternVL2. 5-78B.

IJCAI Conference 2025 Conference Paper

Egocentric Object-Interaction Anticipation with Retentive and Predictive Learning

• Guo Chen
• Yifei Huang
• Yin-Dong Zheng
• Yicheng Liu
• Jiahao Wang
• Tong Lu

Egocentric object-interaction anticipation is critical for applications like augmented reality and robotics, but existing methods struggle with misaligned egocentric encoding, insufficient supervision, and underutilized historical context. These limitations stem from a lack of focus on retention, i. e. , retaining long-term object-centric interactions, and prediction, i. e. , future-centric encoding and future uncertainty modeling. We introduce EgoAnticipator, a novel Retentive and Predictive Learning framework that addresses these challenges. Our approach combines retentive pre-training for domain-specific encoding, predictive pre-training for future uncertainty modeling, and mirror distillation to transfer future-informed knowledge. Additionally, we propose long-term memory prompting to integrate historical interaction cues. We evaluate the effectiveness of our framework using the Ego4D short-term object interaction anticipation benchmark, covering both STAv1 and STAv2. Extensive experiments demonstrate that our framework outperforms existing methods, while ablation studies highlight the effectiveness of each design inside our retentive and predictive learning framework.

ECAI Conference 2024 Conference Paper

CorrAdaptor: Adaptive Local Context Learning for Correspondence Pruning

• Wei Zhu
• Yicheng Liu
• Yuping He
• Tangfei Liao
• Kang Zheng
• Xiaoqiu Xu
• Tao Wang
• Tong Lu

In the fields of computer vision and robotics, accurate pixel-level correspondences are essential for enabling advanced tasks such as structure-from-motion and simultaneous localization and mapping. Recent correspondence pruning methods usually focus on learning local consistency through k-nearest neighbors, which makes it difficult to capture robust context for each correspondence. We propose CorrAdaptor, a novel architecture that introduces a dual-branch structure capable of adaptively adjusting local contexts through both explicit and implicit local graph learning. Specifically, the explicit branch uses KNN-based graphs tailored for initial neighborhood identification, while the implicit branch leverages a learnable matrix to softly assign neighbors and adaptively expand the local context scope, significantly enhancing the model’s robustness and adaptability to complex image variations. Moreover, we design a motion injection module to integrate motion consistency into the network to suppress the impact of outliers and refine local context learning, resulting in substantial performance improvements. The experimental results on extensive correspondence-based tasks indicate that our CorrAdaptor achieves state-of-the-art performance both qualitatively and quantitatively.

AAAI Conference 2024 Conference Paper

Deep Variational Incomplete Multi-View Clustering: Exploring Shared Clustering Structures

• Gehui Xu
• Jie Wen
• Chengliang Liu
• Bing Hu
• Yicheng Liu
• Lunke Fei
• Wei Wang

Incomplete multi-view clustering (IMVC) aims to reveal shared clustering structures within multi-view data, where only partial views of the samples are available. Existing IMVC methods primarily suffer from two issues: 1) Imputation-based methods inevitably introduce inaccurate imputations, which in turn degrade clustering performance; 2) Imputation-free methods are susceptible to unbalanced information among views and fail to fully exploit shared information. To address these issues, we propose a novel method based on variational autoencoders. Specifically, we adopt multiple view-specific encoders to extract information from each view and utilize the Product-of-Experts approach to efficiently aggregate information to obtain the common representation. To enhance the shared information in the common representation, we introduce a coherence objective to mitigate the influence of information imbalance. By incorporating the Mixture-of-Gaussians prior information into the latent representation, our proposed method is able to learn the common representation with clustering-friendly structures. Extensive experiments on four datasets show that our method achieves competitive clustering performance compared with state-of-the-art methods.

ICML Conference 2024 Conference Paper

Language-Driven Cross-Modal Classifier for Zero-Shot Multi-Label Image Recognition

• Yicheng Liu
• Jie Wen 0001
• Chengliang Liu 0003
• Xiaozhao Fang
• Zuoyong Li
• Yong Xu 0001
• Zheng Zhang 0006

Large-scale pre-trained vision-language models (e. g. , CLIP) have shown powerful zero-shot transfer capabilities in image recognition tasks. Recent approaches typically employ supervised fine-tuning methods to adapt CLIP for zero-shot multi-label image recognition tasks. However, obtaining sufficient multi-label annotated image data for training is challenging and not scalable. In this paper, we propose a new language-driven framework for zero-shot multi-label recognition that eliminates the need for annotated images during training. Leveraging the aligned CLIP multi-modal embedding space, our method utilizes language data generated by LLMs to train a cross-modal classifier, which is subsequently transferred to the visual modality. During inference, directly applying the classifier to visual inputs may limit performance due to the modality gap. To address this issue, we introduce a cross-modal mapping method that maps image embeddings to the language modality while retaining crucial visual information. Comprehensive experiments demonstrate that our method outperforms other zero-shot multi-label recognition methods and achieves competitive results compared to few-shot methods.

ICML Conference 2023 Conference Paper

VectorMapNet: End-to-end Vectorized HD Map Learning

• Yicheng Liu
• Tianyuan Yuan
• Yue Wang
• Yilun Wang
• Hang Zhao 0021

Autonomous driving systems require High-Definition (HD) semantic maps to navigate around urban roads. Existing solutions approach the semantic mapping problem by offline manual annotation, which suffers from serious scalability issues. Recent learning-based methods produce dense rasterized segmentation predictions to construct maps. However, these predictions do not include instance information of individual map elements and require heuristic post-processing to obtain vectorized maps. To tackle these challenges, we introduce an end-to-end vectorized HD map learning pipeline, termed VectorMapNet. VectorMapNet takes onboard sensor observations and predicts a sparse set of polylines in the bird’s-eye view. This pipeline can explicitly model the spatial relation between map elements and generate vectorized maps that are friendly to downstream autonomous driving tasks. Extensive experiments show that VectorMapNet achieve strong map learning performance on both nuScenes and Argoverse2 dataset, surpassing previous state-of-the-art methods by 14. 2 mAP and 14. 6mAP. Qualitatively, VectorMapNet is capable of generating comprehensive maps and capturing fine-grained details of road geometry. To the best of our knowledge, VectorMapNet is the first work designed towards end-to-end vectorized map learning from onboard observations.

NeurIPS Conference 2020 Conference Paper

Bidirectional Convolutional Poisson Gamma Dynamical Systems

• Wenchao Chen
• Chaojie Wang
• Bo Chen
• Yicheng Liu
• Hao Zhang
• Mingyuan Zhou

Incorporating the natural document-sentence-word structure into hierarchical Bayesian modeling, we propose convolutional Poisson gamma dynamical systems (PGDS) that introduce not only word-level probabilistic convolutions, but also sentence-level stochastic temporal transitions. With word-level convolutions capturing phrase-level topics and sentence-level transitions capturing how the topic usages evolve over consecutive sentences, we aggregate the topic proportions of all sentences of a document as its feature representation. To consider not only forward but also backward sentence-level information transmissions, we further develop a bidirectional convolutional PGDS to incorporate the full contextual information to represent each sentence. For efficient inference, we construct a convolutional-recurrent inference network, which provides both sentence-level and document-level representations, and introduce a hybrid Bayesian inference scheme combining stochastic-gradient MCMC and amortized variational inference. Experimental results on a variety of document corpora demonstrate that the proposed models can extract expressive multi-level latent representations, including interpretable phrase-level topics and sentence-level temporal transitions as well as discriminative document-level features, achieving state-of-the-art document categorization performance while being memory and computation efficient.

IJCAI Conference 2020 Conference Paper

Switching Poisson Gamma Dynamical Systems

• Wenchao Chen
• Bo Chen
• Yicheng Liu
• Qianru Zhao
• Mingyuan Zhou

We propose Switching Poisson gamma dynamical systems (SPGDS) to model sequentially observed multivariate count data. Different from previous models, SPGDS assigns its latent variables into mixture of gamma distributed parameters to model complex sequences and describe the nonlinear dynamics, meanwhile, capture various temporal dependencies. For efficient inference, we develop a scalable hybrid stochastic gradient-MCMC and switching recurrent autoencoding variational inference, which is scalable to large scale sequences and fast in out-of-sample prediction. Experiments on both unsupervised and supervised tasks demonstrate that the proposed model not only has excellent fitting and prediction performance on complex dynamic sequences, but also separates different dynamical patterns within them.

AAMAS Conference 2017 Conference Paper

Stability of Generalized Two-sided Markets with Transaction Thresholds

• Zhiyuan Li
• Yicheng Liu
• Pingzhong Tang
• Tingting Xu
• Wei Zhan

We study a type of generalized two-sided markets where a set of sellers, each with multiple units of the same divisible good, trade with a set of buyers. Possible trade of each unit between a buyer and a seller generates a given welfare (to be split among them), indicated by the weight of the edge between them. What makes the markets interesting is a special type of constraints, called transaction threshold constraints, which essentially mean that the amount of goods traded between a pair of agents can either be zero or above a certain edge-specific threshold. This constraints has originally been motivated from the water-right market domain by Liu et. al. where minimum thresholds must be imposed to mitigate administrative and other costs. The same constraints have been witnessed in several other market domains. Without the threshold constraints, it is known that the seminal result by Shapley and Shubick holds: the social welfare maximizing assignments between buyers and sellers are in the core. In other words, by algorithmically optimizing the market, one can obtain desirable incentive properties for free. This is no longer the case for markets with threshold constraints: the model considered in this paper. We first demonstrate a counterexample where no optimal assignment (with respect to any way to split the trade welfare) is in the core. Motivated by this observation, we study the stability of the optimal assignments from the following two perspectives: 1) by relaxing the definition of core; 2) by restricting the graph structure. For the first line, we show that the optimal assignments are pairwise stable, and no coalition can benefit twice as large when they deviate. For the second line, we exactly characterize the graph structure ∗ This work was supported in part by the National Natural Science Foundation of China Grant 61303077, 61561146398, a Tsinghua University Initiative Scientific Research Grant and a China National Youth 1000-talent program. Appears in: Proc. of the 16th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2017), S. Das, E. Durfee, K. Larson, M. Winikoff (eds.), May 8–12, 2017, São Paulo, Brazil. Copyright c 2017, International Foundation for Autonomous Agents and Multiagent Systems (www. ifaamas. org). All rights reserved. for the nonemptyness of core: the core is nonempty if and only if the market graph is a tree. Last but not least, we compliment our previous results by quantitatively measuring the welfare loss caused by the threshold constraints: the optimal welfare without transaction thresholds is no greater than constant times of that with transaction thresholds. We evaluate and confirm our theoretical results using real data from a water-right market.

AAAI Conference 2016 Conference Paper

Optimizing Trading Assignments in Water Right Markets

• Yicheng Liu
• Pingzhong Tang
• Tingting Xu
• Hang Zheng

Over the past two decades, water markets have been successfully fielded in countries such as Australia, the United states, Chile, China, etc. Water users, mainly irrigators, have benefited immensely from water markets. However, the current water market design also faces certain serious barriers. It has been pointed out that transaction costs, which exists in most markets, induce great welfare loss. For example, for water markets in western China discussed in this paper, the influence of transaction costs is significant. Another important barrier is the locality of trades due to geographical constraints. Based on the water market at Xiying Irrigation, one of the most successful water market in western China, we model the water market as a graph with minimum transaction thresholds on edges. Our goal is to maximize the transaction volume or welfare. We prove that the existence of transaction costs results in no polynomial time approximation scheme (PTAS) to maximize social welfare (MAX SNP-hard). The complexities on special graphs are also presented. From a practical point of view, however, optimal social welfare can be obtained via a well-designed mixed integer linear program and can be approximated near optimally at a large scale via a heuristic algorithm. Both algorithms are tested on data sets generated from real historical trading data. Our study also suggests the importance of reducing transaction costs, for example, institutional costs in water market design. Our work opens a potentially important avenue of market design within the agenda of computational sustainability.

AAMAS Conference 2016 Conference Paper

Single Item Auctions with Discrete Action Spaces (Extended Abstract)

• Yicheng Liu
• Pingzhong Tang

AAAI Conference 2014 Conference Paper

Internally Stable Matchings and Exchanges

• Yicheng Liu
• Pingzhong Tang
• Wenyi Fang

Stability is a central concept in exchange-based mechanism design. It imposes a fundamental requirement that no subset of agents could beneficially deviate from the outcome prescribed by the mechanism. However, deployment of stability in an exchange mechanism presents at least two challenges. First, it reduces social welfare and sometimes prevents the mechanism from producing a solution. Second, it might incur computational cost to clear the mechanism. In this paper, we propose an alternative notion of stability, coined internal stability, under which we analyze the social welfare bounds and computational complexity. Our contributions are as follows: for both pairwise matchings and limitedlength exchanges, for both unweighted and weighted graphs, (1) we prove desirable tight social welfare bounds; (2) we analyze the computational complexity for clearing the matchings and exchanges. Extensive experiments on the kidney exchange domain demonstrate that the optimal welfare under internal stability is very close to the unconstrained optimal.