Institution profile

University of Washington

Academic institutionnorthamerica · us
Official website
Research library2,246linked papers
Opportunities0open roles
Selected work

Representative Papers

Findings of the BabyLM Challenge: Sample-Efficient Pretraining on Developmentally Plausible Corpora

Apr 10, 2025Proceedings of the BabyLM Challenge at the 27th Conference on Computational Natural Language Learning

Large language models (LLMs) suffer from low data efficiency, typically requiring trillion-word corpora for effective pretraining. Method: Inspired by child language acquisition, this work proposes a cognitively grounded, highly efficient pretraining paradigm using only a developmentally appropriate corpus of under 100 million tokens. We systematically demonstrate—contrary to prevailing assumptions—that such small-scale data can surpass trillion-parameter models’ performance when combined with short-sequence training, knowledge distillation, and multi-task evaluation (covering syntactic competence, downstream task transfer, and out-of-distribution generalization); notably, curriculum learning proves ineffective in this low-data regime. Contribution/Results: Leveraging the LTG-BERT architecture, our best-performing model achieves state-of-the-art results across diverse benchmarks, significantly outperforming standard large baselines. The project yields over 30 empirically validated guidelines—identifying both viable strategies and dead ends—for efficient pretraining, thereby establishing a novel paradigm for cognitive modeling and environmentally sustainable (“green”) AI.

105 citations18 influentialRead paper

VisuoTactile 6D Pose Estimation of an In-Hand Object using Vision and Tactile Sensor Data

Jan 04, 2026IEEE Robotics and Automation Letters

This work addresses the challenge of accurately estimating the 6D pose of grasped objects under severe occlusion, where vision-only approaches often fail. To overcome this limitation, we propose a multimodal method that fuses visual and fingertip tactile sensing. Tactile signals are uniformly represented as contact point clouds, and a pixel-wise dense visual-tactile feature fusion network is designed to enable high-precision pose estimation. To facilitate training, we extend the NVIDIA DISE-based synthetic data generation pipeline to jointly produce realistic RGB images and corresponding tactile point clouds. Experimental results on a real robotic platform demonstrate that our approach significantly outperforms vision-only baselines, and that the model trained on synthetic data generalizes effectively to real-world scenarios.

55 citations6 influentialRead paper

Medical Hallucinations in Foundation Models and Their Impact on Healthcare

Feb 26, 2025arXiv.org

Medical foundation models may generate “hallucinations”—factual, logical, or evidence-inconsistent errors—that jeopardize clinical decision-making and patient safety. To address this, we first propose a multidimensional taxonomy of medical hallucinations and establish a real-world, clinician-annotated benchmark dataset derived from authentic clinical cases; we further validate its clinical impact via an international physician survey. Methodologically, we integrate expert annotation, empirical behavioral surveys, and large language model (LLM) evaluation to systematically assess the efficacy of chain-of-thought (CoT) reasoning and retrieval-augmented generation (RAG) in mitigating hallucinations. Results show both techniques significantly reduce hallucination rates, yet residual hallucinations remain clinically hazardous. Building on these findings, we introduce a patient-safety-centered AI governance and ethics framework, offering theoretical foundations and actionable pathways for responsible deployment of medical AI. (149 words)

41 citationsRead paper

A Tale of Santa Claus, Hypergraphs and Matroids

Jul 19, 2018ACM-SIAM Symposium on Discrete Algorithms

This paper studies the Santa Claus problem with matroid constraints: allocate indivisible items (gifts) to children to maximize the minimum utility among children, where child $i$’s value for item $j$ is either $0$ or $p_j$. It introduces matroid constraints—modeling general resource allocation restrictions—for the first time in this context. Methodologically, it simplifies and generalizes Haxell’s augmenting tree technique, yielding a unified and concise approximation framework that integrates matroid theory, hypergraph matching, linear programming relaxation, and rounding. The algorithm achieves a $(4+varepsilon)$-approximation ratio, substantially improving upon the previous best $12.33$-approximation. Moreover, it serves as a black-box improvement for the LP relaxation bound of the classical Santa Claus problem. This work establishes a new paradigm for fair allocation under combinatorial constraints, bridging fairness, discrete optimization, and structural constraint modeling.

25 citations2 influentialRead paper

From hanging out to figuring it out: Socializing online as a pathway to computational thinking

May 21, 2020New Media & Society

This study addresses a key challenge in educational platforms: effectively transforming adolescents’ online social interactions into opportunities for computational thinking development. Focusing on over 14,000 comments from the Scratch platform, the research introduces the novel concept of “participatory debugging,” wherein users cultivate computational thinking through collaborative troubleshooting. Employing a mixed-methods approach that integrates inductive analysis, content analysis, and longitudinal qualitative analysis, the study identifies three critical social antecedents that support this practice: sustained community engagement, identifiable problems, and topical permeability. The findings not only demonstrate the prevalence of participatory debugging but also establish a theoretical framework linking interest-driven social interaction to computational thinking learning, offering empirical grounding and design implications for creating socially engaging, learning-oriented platforms.

10 citations2 influentialRead paper
Recent publications

Latest Papers