arXiv subjects

Boqi Chen

Publications and source records attributed to Boqi Chen.

At least 19 recordsLinked to original sources

arXiv2026

GPAgentBench-2K: Benchmarking Large Language Model Agents in Complex Clinical Action Space

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.

cs.CL↗

arXiv2026

Benchmarking and Enhancing Text-to-Image Models for Generating Visual Representations in Early Arithmetic Education

AI systems are increasingly used to support educational content creation, yet it remains unclear whether they can generate outputs that faithfully represent the pedagogical concepts they are intended to teach. Thus, we introduce equation-to-visual generation, a task that, in contrast to conventional image generation, requires producing pedagogically meaningful visuals from arithmetic equations while precisely preserving their numerical and relational structure. Informed by interviews with teachers and an analysis of educational materials, we construct E2V-Bench, a benchmark spanning four pedagogically grounded visual types, along with automatic metrics for evaluating visual correctness. Our evaluation reveals that recent text-to-image (T2I) models frequently fail on this task, with errors dominated by incorrect object counts and broken relational structure. Building on this, we explore benchmark-guided enhancement strategies. These strategies improve representative models, while the remaining gap calls for stronger numerical and relational grounding in future T2I models.

cs.CV↗

arXiv2026

Projectional Decoding: Towards Semantic-Aware LLM Generation

Large language models (LLMs) are increasingly used to generate software artifacts across many software engineering (SE) tasks, yet ensuring the semantic validity of these artifacts remains a fundamental challenge. Existing constrained decoding techniques can enforce syntactic correctness and, in some cases, specific semantic rules, but lack a general representation that bridges LLM-generated text with the reasoning required for semantic validation in SE. In this paper, we propose projectional decoding, a novel conceptual framework that integrates domain semantics directly into the generation process by maintaining, alongside text, a partial graph model as the primary artifact representation throughout generation. This abstract representation enables incremental semantic validation by explicitly capturing uncertainty and natively supporting error detection, while guiding generation toward semantically valid outputs with provable guarantees. We present preliminary results on a program generation task which demonstrate the potential of this approach to improve the semantic validity of LLM-generated artifacts. We also discuss how projectional decoding can enable verifiable automation with LLMs across various SE activities.

cs.SE↗

arXiv2026

LLM-Assisted Repository-Level Generation with Structured Spec-Driven Engineering

State-of-the-art Large Language Models (LLMs) excel in code generation at the function level. However, the output quality significantly declines when scaling to repository-level systems. Current workflows relying only on natural language prompts suffer from inherent ambiguity and a lack of verifiability. To address this, we propose structured spec-driven engineering (SSDE), a paradigm that leverages structured artifacts to guide LLM generation. We argue that structured specifications as LLM inputs make high-quality, repository-level code generation a tangible goal, while at the same time offering superior verifiability, leading to significant potential for improvement. We first investigate the feasibility of this vision through a pilot study generating Model-View-Controller (MVC) business logic for three software systems using five LLMs, and then highlight the potential, challenges, and future roadmap for SSDE.

cs.SE↗

arXiv2026

Revisiting Greedy Decoding for Visual Question Answering: A Calibration Perspective

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.

cs.CL↗

arXiv2026

Structure- and Event-Driven Frameworks for State Machine Modeling with Large Language Models

UML state machine design is a critical process in software engineering. Traditionally, state machines are manually crafted by experienced engineers based on natural language requirements-a time-consuming and error-prone procedure. Many automated approaches exist but they require structured NL requirements. In this paper, we investigate the capabilities of current Large Language Models to fully automate UML state machine generation via specialized State Machine Frameworks (SMFs) from non-structured NL requirements. We evaluate two types of state-of-the-art LLMs using single-step and multi-step prompting approaches: a non-reasoning LLM GPT-4o and a reasoning-focused LLM Claude 3.5 Sonnet, and introduce a novel Hybrid Approach that uses the output from a Single-Prompt Baseline as an initial draft state machine, which is then refined through an SMF. In our study, two distinct SMFs are developed based on human approaches: (i) a Structure-Driven SMF, in which state machine components (states, transitions, guards, actions, etc.) are generated in sequential steps, and (ii) an Event-Driven SMF, where identified events iteratively guide state machine construction. Our experiments indicate that while LLMs demonstrate a promising ability to generate state machine models from the Single-Prompt Baseline (e.g., F1-scores of 0.90 for states and 0.75 for transitions using Claude 3.5 Sonnet), their performance is not yet fully sufficient for a fully automated solution (e.g., F1-scores of 0.23 for guards and 0.00 for actions for GPT-4o). Our proposed Hybrid Approach improves the performance of the non-reasoning LLM (GPT-4o) to a similar level as the reasoning LLM (Claude 3.5 Sonnet) but does not further improve the reasoning LLM. Our evaluation highlights both the potential and the limitations of current LLMs for automated state machine design, providing a baseline for future research in this domain.

cs.SE↗

arXiv2026

Controllable Egocentric Video Generation via Occlusion-Aware Sparse 3D Hand Joints

Controllable video generation for complex hand-object interactions is a critical step toward building visual world models. However, existing methods often struggle to achieve fine-grained, 3D-consistent hand articulation in generated videos. By relying on dense 2D trajectories or implicit pose representations, they collapse crucial geometric structures into spatially ambiguous signals, leading to severe motion inconsistencies and hallucinated artifacts under egocentric occlusions. To address this, we propose leveraging sparse 3D hand joints as explicit control signals with three key advantages: explicit geometry to resolve occlusions, an intuitive interface for interactive editing, and cross-embodiment generalization to robotic hands. Built upon this, our efficient control module extracts occlusion-aware features from the source reference frame by penalizing unreliable visual features from hidden joints, and employs a 3D-based weighting mechanism to handle dynamically occluded target joints during motion propagation. Meanwhile, it directly injects 3D geometric embeddings into the latent space to enforce structural consistency. To facilitate robust training and evaluation, we develop an automated annotation pipeline, yielding 1M high-quality egocentric video clips paired with precise hand trajectories. Experiments demonstrate that our approach outperforms state-of-the-art baselines, generating high-fidelity egocentric videos with realistic hand-object interactions.

cs.CV↗

arXiv2026

MEDSYN: Benchmarking Multi-EviDence SYNthesis in Complex Clinical Cases for Multimodal Large Language Models

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 ($\it{e.g.}$, 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.

cs.CL↗

arXiv2026

Mask What Matters: Mitigating Object Hallucinations in Multimodal Large Language Models with Object-Aligned Visual Contrastive Decoding

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.

cs.CV↗

arXiv2026

Evaluating Implicit Regulatory Compliance in LLM Tool Invocation via Logic-Guided Synthesis

The integration of large language models (LLMs) into autonomous agents has enabled complex tool use, yet in high-stakes domains, these systems must strictly adhere to regulatory standards beyond simple functional correctness. However, existing benchmarks often overlook implicit regulatory compliance, thus failing to evaluate whether LLMs can autonomously enforce mandatory safety constraints. To fill this gap, we introduce LogiSafetyGen, a framework that converts unstructured regulations into Linear Temporal Logic oracles and employs logic-guided fuzzing to synthesize valid, safety-critical traces. Building on this framework, we construct LogiSafetyBench, a benchmark comprising 240 human-verified tasks that require LLMs to generate Python programs that satisfy both functional objectives and latent compliance rules. Evaluations of 13 state-of-the-art (SOTA) LLMs reveal that larger models, despite achieving better functional correctness, frequently prioritize task completion over safety, which results in non-compliant behavior.

cs.CL↗

arXiv2025

Hierarchical Evaluation of Software Design Capabilities of Large Language Models of Code

Large language models (LLMs) are being increasingly adopted in the software engineering domain, yet the robustness of their grasp on core software design concepts remains unclear. We conduct an empirical study to systematically evaluate their understanding of cohesion (intra-module) and coupling (inter-module). We programmatically generate poorly designed code fragments and test the DeepSeek-R1 model family ($14$B, $32$B, $70$B) under varying levels of guidance, from simple \textit{Verification} to \textit{Guided} and \textit{Open-ended Generation}, while varying contextual noise by injecting distractor elements. While models exhibit a solid baseline understanding of both concepts in ideal conditions, their practical knowledge is fragile and highly asymmetrical. Reasoning about coupling proves brittle; performance collapses in noisy, open-ended scenarios, with F1 scores dropping by over $50\%$. In contrast, the models' analysis of cohesion is remarkably robust to internal noise in guided tasks, showing little performance degradation. However, this resilience also fails when all guidance is removed. Reasoning-trace analysis confirms these failure modes, revealing \textit{cognitive shortcutting} for coupling versus a more exhaustive (yet still failing) analysis for cohesion. To summarize, while LLMs can provide reliable assistance for recognizing design flaws, their ability to reason autonomously in noisy, realistic contexts is limited, highlighting the critical need for more scalable and robust program understanding capabilities.

cs.SE↗

arXiv2025

YoNoSplat: You Only Need One Model for Feedforward 3D Gaussian Splatting

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/.

cs.CV↗

arXiv2025

SHERPA: A Model-Driven Framework for Large Language Model Execution

Recently, large language models (LLMs) have achieved widespread application across various fields. Despite their impressive capabilities, LLMs suffer from a lack of structured reasoning ability, particularly for complex tasks requiring domain-specific best practices, which are often unavailable in the training data. Although multi-step prompting methods incorporating human best practices, such as chain-of-thought and tree-of-thought, have gained popularity, they lack a general mechanism to control LLM behavior. In this paper, we propose SHERPA, a model-driven framework to improve the LLM performance on complex tasks by explicitly incorporating domain-specific best practices into hierarchical state machines. By structuring the LLM execution processes using state machines, SHERPA enables more fine-grained control over their behavior via rules or decisions driven by machine learning-based approaches, including LLMs. We show that SHERPA is applicable to a wide variety of tasks-specifically, code generation, class name generation, and question answering-replicating previously proposed approaches while further improving the performance. We demonstrate the effectiveness of SHERPA for the aforementioned tasks using various LLMs. Our systematic evaluation compares different state machine configurations against baseline approaches without state machines. Results show that integrating well-designed state machines significantly improves the quality of LLM outputs, and is particularly beneficial for complex tasks with well-established human best practices but lacking data used for training LLMs.

cs.AI↗

arXiv2025

WeChat-YATT: A Scalable, Simple, Efficient, and Production Ready Training Library

Reinforcement Learning from Human Feedback (RLHF) has emerged as a prominent paradigm for training large language models and multimodal systems. Despite the notable advances enabled by existing RLHF training frameworks, significant challenges remain to scale to complex multimodal workflows and adapt to dynamic workloads. In particular, current systems often encounter limitations related to controller scalability when managing large models, as well as inefficiencies in orchestrating intricate RLHF pipelines, especially in scenarios that require dynamic sampling and resource allocation. In this paper, we introduce WeChat-YATT Yet Another Transformer Trainer in WeChat, a simple, scalable, and balanced RLHF training framework specifically designed to address these challenges. WeChat-YATT features a parallel controller programming model that enables flexible and efficient orchestration of complex RLHF workflows, effectively mitigating bottlenecks associated with centralized controller architectures and facilitating scalability in large-scale data scenarios. In addition, we propose a dynamic placement schema that adaptively partitions computational resources and schedules workloads, thereby significantly reducing hardware idle time and improving GPU utilization under variable training conditions. We evaluate WeChat-YATT across diverse experimental scenarios, demonstrating its substantial throughput improvements over state-of-the-art RLHF training frameworks. Furthermore, WeChat-YATT has been successfully deployed to train models that support WeChat product features for a large-scale user base, underscoring its effectiveness and robustness in real-world applications. We have made WeChat-YATT publicly available at https://www.github.com/tencent/WeChat-YATT.

cs.LG↗

arXiv2025

Accurate and Consistent Graph Model Generation from Text with Large Language Models

Graph model generation from natural language description is an important task with many applications in software engineering. With the rise of large language models (LLMs), there is a growing interest in using LLMs for graph model generation. Nevertheless, LLM-based graph model generation typically produces partially correct models that suffer from three main issues: (1) syntax violations: the generated model may not adhere to the syntax defined by its metamodel, (2) constraint inconsistencies: the structure of the model might not conform to some domain-specific constraints, and (3) inaccuracy: due to the inherent uncertainty in LLMs, the models can include inaccurate, hallucinated elements. While the first issue is often addressed through techniques such as constraint decoding or filtering, the latter two remain largely unaddressed. Motivated by recent self-consistency approaches in LLMs, we propose a novel abstraction-concretization framework that enhances the consistency and quality of generated graph models by considering multiple outputs from an LLM. Our approach first constructs a probabilistic partial model that aggregates all candidate outputs and then refines this partial model into the most appropriate concrete model that satisfies all constraints. We evaluate our framework on several popular open-source and closed-source LLMs using diverse datasets for model generation tasks. The results demonstrate that our approach significantly improves both the consistency and quality of the generated graph models.

cs.SE↗

arXiv2025

MCeT: Behavioral Model Correctness Evaluation using Large Language Models

Behavioral model diagrams, e.g., sequence diagrams, are an essential form of documentation that are typically designed by system engineers from requirements documentation, either fully manually or assisted by design tools. With the growing use of Large Language Models (LLM) as AI modeling assistants, more automation will be involved in generating diagrams. This necessitates the advancement of automatic model correctness evaluation tools. Such a tool can be used to evaluate both manually and AI automatically generated models; to provide feedback to system engineers, and enable AI assistants to self-evaluate and self-enhance their generated models. In this paper, we propose MCeT, the first fully automated tool to evaluate the correctness of a behavioral model, sequence diagrams in particular, against its corresponding requirements text and produce a list of issues that the model has. We utilize LLMs for the correctness evaluation tasks as they have shown outstanding natural language understanding ability. However, we show that directly asking an LLM to compare a diagram to requirements finds less than 35% of issues that experienced engineers can find. We propose to supplement the direct check with a fine-grained, multi-perspective approach; we split the diagram into atomic, non-divisible interactions, and split the requirements text into atomic, self-contained items. We compare the diagram with atomic requirements and each diagram-atom with the requirements. We also propose a self-consistency checking approach that combines perspectives to mitigate LLM hallucinated issues. Our combined approach improves upon the precision of the direct approach from 0.58 to 0.81 in a dataset of real requirements. Moreover, the approach finds 90% more issues that the experienced engineers found than the direct approach, and reports an average of 6 new issues per diagram.

cs.SE↗

arXiv2025

NOCTA: Non-Greedy Objective Cost-Tradeoff Acquisition for Longitudinal Data

In many critical domains, features are not freely available at inference time: each measurement may come with a cost of time, money, and risk. Longitudinal prediction further complicates this setting because both features and labels evolve over time, and missing measurements at earlier timepoints may become permanently unavailable. We propose NOCTA, a Non-Greedy Objective Cost-Tradeoff Acquisition framework that sequentially acquires the most informative features at inference time while accounting for both temporal dynamics and acquisition cost. NOCTA is driven by a novel objective, NOCT, which evaluates a candidate set of future feature-time acquisitions by its expected predictive loss together with its acquisition cost. Since NOCT depends on unobserved future trajectories at inference time, we develop two complementary estimators: (i) NOCT-Contrastive, which learns an embedding of partial observations utilizing the induced distribution over future acquisitions, and (ii) NOCT-Amortized, which directly predicts NOCT for candidate plans with a neural network. Experiments on synthetic and real-world medical datasets demonstrate that both NOCTA estimators outperform existing baselines, achieving higher accuracy at lower acquisition costs.

cs.LG↗

arXiv2025

LLM-based Satisfiability Checking of String Requirements by Consistent Data and Checker Generation

Requirements over strings, commonly represented using natural language (NL), are particularly relevant for software systems due to their heavy reliance on string data manipulation. While individual requirements can usually be analyzed manually, verifying properties (e.g., satisfiability) over sets of NL requirements is particularly challenging. Formal approaches (e.g., SMT solvers) may efficiently verify such properties, but are known to have theoretical limitations. Additionally, the translation of NL requirements into formal constraints typically requires significant manual effort. Recently, large language models (LLMs) have emerged as an alternative approach for formal reasoning tasks, but their effectiveness in verifying requirements over strings is less studied. In this paper, we introduce a hybrid approach that verifies the satisfiability of NL requirements over strings by using LLMs (1) to derive a satisfiability outcome (and a consistent string, if possible), and (2) to generate declarative (i.e., SMT) and imperative (i.e., Python) checkers, used to validate the correctness of (1). In our experiments, we assess the performance of four LLMs. Results show that LLMs effectively translate natural language into checkers, even achieving perfect testing accuracy for Python-based checkers. These checkers substantially help LLMs in generating a consistent string and accurately identifying unsatisfiable requirements, leading to more than doubled generation success rate and F1-score in certain cases compared to baselines without generated checkers.

cs.SE↗