Institution profile

University of Virginia

Academic institutionnorthamerica · us
Official website
Research library870linked papers
Opportunities0open roles
Selected work

Representative Papers

Reward Is Enough: LLMs Are In-Context Reinforcement Learners

May 21, 2025arXiv.org

This work investigates whether large language models (LLMs) can spontaneously perform reinforcement learning–like optimization during inference—termed *in-context reinforcement learning* (ICRL). To this end, we propose an ICRL framework that requires no parameter updates and instead relies solely on multi-round context augmentation and scalar reward feedback: at each round, the prompt is dynamically reconstructed based on prior model responses and (human- or self-generated) rewards, enabling online improvement of output quality. Our key contribution is the first empirical demonstration that LLMs possess gradient-free, inference-time reward maximization capability, supporting closed-loop self-feedback optimization and extending beyond conventional test-time learning paradigms. Evaluated on Game of 24, creative writing, and ScienceWorld, ICRL significantly outperforms Self-Refine and Reflexion. Notably, it maintains robust performance gains even when using self-assessed rewards.

16 citations1 influentialRead paper

Conceptual Modeling: Topics, Themes, and Technology Trends

Mar 30, 2023ACM Computing Surveys

Despite rapid technological advances, the enduring value and evolutionary trajectory of conceptual modeling in digital-era information systems development remain inadequately understood. Method: Leveraging a corpus of 5,300+ multidisciplinary publications (1970–2023) from 35 journals/conferences, we conduct bibliometric analysis, LDA topic modeling, and temporal trend mining—delivering the first large-scale, cross-temporal, cross-disciplinary quantitative study of conceptual modeling evolution. Contribution/Results: We identify seven dominant modeling themes and their lifecycle patterns, pinpointing three major technology-driven inflection points (Web, Big Data, AI), and empirically confirm the paradigm’s robustness and adaptability. Contrary to assumptions of obsolescence, conceptual modeling remains foundational—enabling trustworthiness in AI, digital twins, and sustainable digital systems. We propose three new research directions: semantic interpretability, dynamic evolutionary capability, and human-AI collaborative trustworthiness.

16 citations1 influentialRead paper

Reasoning Beyond Chain-of-Thought: A Latent Computational Mode in Large Language Models

Jan 12, 2026

This study investigates the internal mechanisms underlying reasoning in large language models, challenging the prevailing assumption that chain-of-thought (CoT) prompting is necessary to elicit reasoning capabilities. By applying sparse autoencoders (SAEs) to dissect model representations, the authors identify a small set of latent features causally linked to reasoning behavior. Leveraging latent steering techniques, they directly activate these features without relying on CoT prompts. Experiments across multiple models and reasoning benchmarks demonstrate that manipulating just a single such feature can substantially improve reasoning accuracy—matching or even surpassing the performance of standard CoT prompting—while yielding more efficient generations. This work provides the first evidence that large language models harbor reasoning-oriented computational pathways that can be externally activated, establishing CoT as one effective—but not exclusive—means of engaging this intrinsic reasoning mechanism.

3 citationsRead paper

LabelAny3D: Label Any Object 3D in the Wild

Jan 04, 2026arXiv.org

This work addresses the performance limitations of existing monocular 3D object detection methods in wild scenes, primarily caused by the scarcity of high-quality 3D annotations. To overcome this, the authors propose a 3D scene reconstruction approach based on an analysis-by-synthesis framework, which leverages foundation models to drive an automated annotation pipeline. This pipeline efficiently generates open-vocabulary, large-scale 3D bounding box annotations from single 2D images and constructs the COCO3D benchmark dataset based on MS-COCO. By eliminating reliance on manual labeling or constrained environments, the method breaks through traditional bottlenecks in 3D annotation. The resulting pseudo-labels significantly improve performance across multiple monocular 3D detection benchmarks and surpass existing automatic annotation techniques in quality, demonstrating strong scalability and effectiveness for open-world 3D recognition.

3 citationsRead paper

MoCo: A One-Stop Shop for Model Collaboration Research

Jan 29, 2026

This work addresses the lack of a unified benchmark in existing research on model collaboration, which hinders systematic evaluation of diverse collaborative strategies. To bridge this gap, we propose MoCo—a modular Python library that, for the first time, systematically integrates 26 collaboration methods, supporting multi-granularity information exchange at the levels of routing, text, logits, and parameters. MoCo incorporates 25 diverse evaluation datasets and provides an extensible framework for heterogeneous model collaboration alongside efficient analysis tools, thereby establishing model collaboration as a distinct research paradigm. Experimental results demonstrate that collaborative strategies outperform single-model baselines in 61.0% of (model, dataset) configurations, with the best-performing method achieving a performance gain of up to 25.8%, highlighting the substantial advantages of collaboration in complex tasks.

2 citationsRead paper
Recent publications

Latest Papers