Large Language Models (LLMs) show great potential as clinical agents, yet existing benchmarks reduce clinical workflows to static predictions or unconstrained Markov Decision Processes (MDPs) with coarse action sets. To address this, we introduce GPAgentBench-2K, the first Constrained MDP (CMDP) LLM-agent benchmark for primary-care clinical decision-making, constructed from expert-validated records of real-world GP encounters. Our environment models a full spectrum of six foundational clinical actions, imposes a topological workflow prior over the action space, and operationalizes safety-informed abstention as a first-class outcome. Evaluating 16 state-of-the-art LLMs reveals a significant performance degradation as the action space scales. Crucially, we uncover a clinical quality-safety gap: even frontier models with the highest diagnosis accuracy violate safety constraints in over half of high-risk cases. Finally, we establish a reference point using Constrained Group Relative Policy Optimization (C-GRPO), and show that while explicitly modeling constraints improves performance over unconstrained RL methods, it remains far from clinically acceptable safety.
@inproceedings{chen2026gpagentbench,author={Chen, Boqi and Liu, Xudong and Ao, Yunke and Do, Heejin and Qiu, Jianing},title={{GPAgentBench-2K}: Benchmarking Large Language Model Agents in Complex Clinical Action Space},booktitle={Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP)},year={2026},}
EMNLP
Revisiting Greedy Decoding for Visual Question Answering: A Calibration Perspective
Boqi Chen , Xudong Liu , Yunke Ao, and 1 more author
In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP) , 2026
Stochastic sampling strategies are widely adopted in large language models (LLMs) to balance output coherence and diversity. These heuristics are often inherited in Multimodal LLMs (MLLMs) without task-specific justification. However, we contend that stochastic decoding can be suboptimal for Visual Question Answering (VQA). VQA is a closed-ended task with head-heavy answer distributions where uncertainty is usually epistemic, arising from missing or ambiguous visual evidence rather than plausible continuations. In this work, we provide a theoretical formalization of the relationship between model calibration and predictive accuracy, and derive the sufficient conditions for greedy decoding optimality. Extensive experiments provide empirical evidence for the superiority of greedy decoding over stochastic sampling across multiple benchmarks. Furthermore, we propose Greedy Decoding for Reasoning Models, which outperforms both stochastic sampling and standard greedy decoding in multimodal reasoning scenarios. Overall, our results caution against naively inheriting LLMs decoding heuristics in MLLMs and demonstrate that greedy decoding can be an efficient yet strong default for VQA.
@inproceedings{chen2026greedy,author={Chen, Boqi and Liu, Xudong and Ao, Yunke and Qiu, Jianing},title={Revisiting Greedy Decoding for Visual Question Answering: A Calibration Perspective},booktitle={Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP)},year={2026},}
ECCV
Controllable Egocentric Video Generation via Occlusion-Aware Sparse 3D Hand Joints
Chenyangguang Zhang* , Botao Ye* , Boqi Chen*, and 4 more authors
* Equal contribution.
In European Conference on Computer Vision (ECCV) , 2026
Motion-controllable video generation is crucial for egocentric applications in virtual reality and embodied AI. However, existing methods often struggle to achieve 3D-consistent fine-grained hand articulation. By adopting on 2D trajectories or implicit poses, they collapse 3D geometry into spatially ambiguous signals or over rely on human-centric priors. Under severe egocentric occlusions, this causes motion inconsistencies and hallucinated artifacts, as well as preventing cross-embodiment generalization to robotic hands. To address these limitations, we propose a novel framework that generates egocentric videos from a single reference frame, leveraging sparse 3D hand joints as embodiment-agnostic control signals with clear semantic and geometric structures. We introduce an efficient control module that resolves occlusion ambiguities while fully preserving 3D information. Specifically, it extracts occlusion-aware features from the source reference frame by penalizing unreliable visual signals from hidden joints, and employs a 3D-based weighting mechanism to robustly handle dynamically occluded target joints during motion propagation. Concurrently, the module directly injects 3D geometric embeddings into the latent space to strictly enforce structural consistency. To facilitate robust training and evaluation, we develop an automated annotation pipeline that yields over one million high-quality egocentric video clips paired with precise hand trajectories. Additionally, we register humanoid kinematic and camera data to construct a cross-embodiment benchmark. Extensive experiments demonstrate that our approach significantly outperforms state-of-the-art baselines, generating high-fidelity egocentric videos with realistic interactions and exhibiting exceptional cross-embodiment generalization to robotic hands.
@inproceedings{zhang2026egocentric,author={Zhang, Chenyangguang and Ye, Botao and Chen, Boqi and Delitzas, Alexandros and Wang, Fangjinhua and Pollefeys, Marc and Wang, Xi},title={Controllable Egocentric Video Generation via Occlusion-Aware Sparse 3D Hand Joints},booktitle={European Conference on Computer Vision (ECCV)},year={2026},}
ACL
MEDSYN: Benchmarking Multi-EviDence SYNthesis in Complex Clinical Cases for Multimodal Large Language Models
Boqi Chen , Xudong Liu , Jiachuan Peng, and 5 more authors
In Findings of the Association for Computational Linguistics: ACL 2026 , 2026
Multimodal large language models (MLLMs) have shown great potential in medical applications, yet existing benchmarks inadequately capture real-world clinical complexity. We introduce MEDSYN, a multilingual, multimodal benchmark of highly complex clinical cases with up to 7 distinct visual clinical evidence (CE) types per case. Mirroring clinical workflow, we evaluate 18 MLLMs on differential diagnosis (DDx) generation and final diagnosis (FDx) selection. While top models often match or even outperform human experts on DDx generation, all MLLMs exhibit a much larger DDx–FDx performance gap compared to expert clinicians, indicating a failure mode in synthesis of heterogeneous CE types. Ablations attribute this failure to (i) overreliance on less discriminative textual CE (e.g., medical history) and (ii) a cross-modal CE utilization gap. We introduce Evidence Sensitivity to quantify the latter and show that a smaller gap correlates with higher diagnostic accuracy. Finally, we demonstrate how it can be used to guide interventions to improve model performance. We will open-source our benchmark and code.
@inproceedings{chen2026medsyn,author={Chen, Boqi and Liu, Xudong and Peng, Jiachuan and Frey-Marti, Marianne and Zheng, Bang and Lam, Kyle and Li, Lin and Qiu, Jianing},title={{MEDSYN}: Benchmarking Multi-EviDence SYNthesis in Complex Clinical Cases for Multimodal Large Language Models},booktitle={Findings of the Association for Computational Linguistics: ACL 2026},year={2026},}
EACL
Mask What Matters: Mitigating Object Hallucinations in Multimodal Large Language Models with Object-Aligned Visual Contrastive Decoding
Boqi Chen , Xudong Liu , and Jianing Qiu
In Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics: SRW , 2026
We study object hallucination in Multimodal Large Language Models (MLLMs) and improve visual contrastive decoding (VCD) by constructing an object-aligned auxiliary view. We leverage object-centric attention in self-supervised Vision Transformers. In particular, we remove the most salient visual evidence to construct an auxiliary view that disrupts unsupported tokens and produces a stronger contrast signal. Our method is prompt-agnostic, model-agnostic, and can be seamlessly plugged into the existing VCD pipeline with little computation overhead, i.e., a single cacheable forward pass. Empirically, our method demonstrates consistent gains on two popular object hallucination benchmarks across two MLLMs.
@inproceedings{chen2026mask,author={Chen, Boqi and Liu, Xudong and Qiu, Jianing},title={Mask What Matters: Mitigating Object Hallucinations in Multimodal Large Language Models with Object-Aligned Visual Contrastive Decoding},booktitle={Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics: SRW},year={2026},}
ICLR
YoNoSplat: You Only Need One Model for Feedforward 3D Gaussian Splatting
Botao Ye , Boqi Chen , Haofei Xu, and 2 more authors
In International Conference on Learning Representations (ICLR) , 2026
Fast and flexible 3D scene reconstruction from unstructured image collections remains a significant challenge. We present YoNoSplat, a feedforward model that reconstructs high-quality 3D Gaussian Splatting representations from an arbitrary number of images. Our model is highly versatile, operating effectively with both posed and unposed, calibrated and uncalibrated inputs. YoNoSplat predicts local Gaussians and camera poses for each view, which are aggregated into a global representation using either predicted or provided poses. To overcome the inherent difficulty of jointly learning 3D Gaussians and camera parameters, we introduce a novel mixing training strategy. This approach mitigates the entanglement between the two tasks by initially using ground-truth poses to aggregate local Gaussians and gradually transitioning to a mix of predicted and ground-truth poses, which prevents both training instability and exposure bias. We further resolve the scale ambiguity problem by a novel pairwise camera-distance normalization scheme and by embedding camera intrinsics into the network. Moreover, YoNoSplat also predicts intrinsic parameters, making it feasible for uncalibrated inputs. YoNoSplat demonstrates exceptional efficiency, reconstructing a scene from 100 views (at 280x518 resolution) in just 2.69 seconds on an NVIDIA GH200 GPU. It achieves state-of-the-art performance on standard benchmarks in both pose-free and pose-dependent settings. Our project page is at https://botaoye.github.io/yonosplat/.
@inproceedings{ye2026yonosplat,author={Ye, Botao and Chen, Boqi and Xu, Haofei and Barath, Daniel and Pollefeys, Marc},title={{YoNoSplat}: You Only Need One Model for Feedforward 3D Gaussian Splatting},booktitle={International Conference on Learning Representations (ICLR)},year={2026},}
2025
MICCAI
Revisiting Automatic Data Curation for Vision Foundation Models in Digital Pathology
Boqi Chen , Cédric Vincent-Cuaz , Lydia A. Schoenpflug, and 12 more authors
In Medical Image Computing and Computer Assisted Intervention (MICCAI) , 2025
This work investigates unsupervised tile-level data curation for digital pathology foundation models, using hierarchical clustering over large-scale tile embeddings and balanced sampling strategies to improve downstream clinical performance.
@inproceedings{chen2025revisiting,author={Chen, Boqi and Vincent-Cuaz, C{\'e}dric and Schoenpflug, Lydia A. and Madeira, Manuel and Fournier, Lisa and Subramanian, Vaishnavi and Andani, Sonali and Ruiperez-Campillo, Samuel and Vogt, Julia E. and Luisier, Rapha{\"e}lle and Thanou, Dorina and Koelzer, Viktor H. and Frossard, Pascal and Campanella, Gabriele and R{\"a}tsch, Gunnar},title={Revisiting Automatic Data Curation for Vision Foundation Models in Digital Pathology},booktitle={Medical Image Computing and Computer Assisted Intervention (MICCAI)},year={2025},doi={10.1007/978-3-032-04978-0_53},}
Nat. Mach. Intell.
Histopathology-Based Protein Multiplex Generation Using Deep Learning
Sonali Andani , Boqi Chen , Joanna Ficek-Pascual, and 7 more authors
HistoPlexer generates spatially resolved protein multiplexes from standard H&E histopathology images and supports tumor microenvironment characterization.
@article{andani2025histopathology,author={Andani, Sonali and Chen, Boqi and Ficek-Pascual, Joanna and Heinke, Simon and Casanova, Ruben and Hild, Bernard Friedrich and Sobottka, Bettina and Bodenmiller, Bernd and Koelzer, Viktor H. and R{\"a}tsch, Gunnar},title={Histopathology-Based Protein Multiplex Generation Using Deep Learning},journal={Nature Machine Intelligence},year={2025},doi={10.1038/s42256-025-01074-y},}
WACV
Generalizable Single-Source Cross-Modality Medical Image Segmentation via Invariant Causal Mechanisms
Boqi Chen , Yuanzhi Zhu , Yunke Ao, and 5 more authors
In The IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , 2025
This work uses causality-inspired invariant representations and diffusion-based augmentation to improve single-source domain generalization for cross-modality medical image segmentation.
@inproceedings{chen2025generalizable,author={Chen, Boqi and Zhu, Yuanzhi and Ao, Yunke and Caprara, Sebastiano and Sutter, Reto and R{\"a}tsch, Gunnar and Konukoglu, Ender and Susmelj, Anna},title={Generalizable Single-Source Cross-Modality Medical Image Segmentation via Invariant Causal Mechanisms},booktitle={The IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},year={2025},}
IROS
MIXPINN: Mixed-Material Simulations by Physics-Informed Neural Network
Xintian Yuan , Yunke Ao , Boqi Chen, and 1 more author
In The IEEE/RSJ International Conference on Intelligent Robots and Systems , 2025
MIXPINN is a physics-informed graph neural network framework for efficient mixed-material biomechanical simulation with soft-rigid interactions.
@inproceedings{yuan2025mixpinn,author={Yuan, Xintian and Ao, Yunke and Chen, Boqi and Fuernstahl, Philipp},title={MIXPINN: Mixed-Material Simulations by Physics-Informed Neural Network},booktitle={The IEEE/RSJ International Conference on Intelligent Robots and Systems},year={2025},}
2023
MedIA
Generative Appearance Replay for Continual Unsupervised Domain Adaptation
Boqi Chen , Kevin Thandiackal , Pushpak Pati, and 1 more author
GarDA adapts a segmentation model sequentially across unlabeled domains using generative replay, without requiring access to previously seen data.
@article{chen2023generative,author={Chen, Boqi and Thandiackal, Kevin and Pati, Pushpak and Goksel, Orcun},title={Generative Appearance Replay for Continual Unsupervised Domain Adaptation},journal={Medical Image Analysis},year={2023},doi={10.1016/j.media.2023.102843},}
2022
ECCV
Differentiable Zooming for Multiple Instance Learning on Whole-Slide Images
Kevin Thandiackal* , Boqi Chen* , Pushpak Pati, and 4 more authors
* Equal contribution.
In European Conference on Computer Vision (ECCV) , 2022
ZoomMIL learns multi-level zooming for whole-slide image classification, improving representation quality while reducing compute.
@inproceedings{thandiackal2022differentiable,author={Thandiackal, Kevin and Chen, Boqi and Pati, Pushpak and Jaume, Guillaume and Williamson, Drew F. K. and Gabrani, Maria and Goksel, Orcun},title={Differentiable Zooming for Multiple Instance Learning on Whole-Slide Images},booktitle={European Conference on Computer Vision (ECCV)},year={2022},doi={10.1007/978-3-031-19803-8_41},}