Institution profile

Carnegie Mellon University

Academic institutionnorthamerica · us
Official website
Research library4,693linked papers
Opportunities0open roles
Selected work

Representative Papers

Explaining the Success of Nearest Neighbor Methods in Prediction

May 31, 2018Found. Trends Mach. Learn.

Despite widespread empirical success, the theoretical foundations and practical deployment guidelines for k-nearest neighbors (k-NN) in predictive tasks remain inadequately understood. Method: We establish the first non-asymptotic error bound framework tailored for real-world deployment, replacing conventional smoothness or margin assumptions with verifiable cluster structure as the key success criterion. We integrate approximate nearest neighbor techniques—including LSH and graph-based indexing—and unify k-NN theory with emerging paradigms such as random forests, graphon modeling, and crowdsourcing. We further introduce a novel distance-learning perspective, characterizing how ensemble methods implicitly learn neighborhood structure. Contribution/Results: Evaluated on time-series forecasting, recommender systems, and medical image segmentation, our framework demonstrates that high accuracy is guaranteed solely under cluster-structured data—enhancing both theoretical interpretability and engineering tractability. It provides actionable, error-tolerance-driven guidance for data volume and hyperparameter selection.

144 citations10 influentialRead paper

Tree Search for Language Model Agents

Jul 01, 2024arXiv.org

To address the limited multi-step decision-making capability of language model agents in real-world web environments—characterized by weak planning and insufficient utilization of environmental feedback—this paper proposes Environment-aware Best-First Search (Env-BFS), a reasoning-time, environment-integrated tree search algorithm. Env-BFS explicitly models the action space, enables interactive environment exploration and backtracking, and achieves computationally scalable planning via dynamic LM API scheduling. As the first tree search framework validated on realistic web benchmarks (VisualWebArena and WebArena), it features a plug-and-play design compatible with any state-of-the-art agent. Experiments demonstrate significant improvements: success rates increase by 39.7% to 26.4% on VisualWebArena and by 28.0% to 19.2% on WebArena—establishing new SOTA performance. The code and models are publicly released.

85 citations14 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

“It’s the only thing I can trust”: Envisioning Large Language Model Use by Autistic Workers for Communication Assistance

Mar 05, 2024International Conference on Human Factors in Computing Systems

This study investigates the applicability, opportunities, and risks of large language models (LLMs) as workplace social communication aids for autistic adults. Employing GPT-4 in a human–AI comparative experimental design (LLM vs. disguised human advisor), complemented by qualitative interviews with 11 autistic adult participants and expert review by vocational coaches, the research identifies a significant cognitive divergence: participants exhibited high trust and strong preference for LLM-generated advice, whereas employment coaches raised concerns regarding normative biases and insufficient contextual adaptation. The findings yield autism-centered design principles that prioritize user autonomy while integrating professional safety guardrails—balancing adaptive support with ethical accountability. This work establishes the first empirical foundation for developing AI-assisted tools tailored to neurodiverse populations, offering both a theoretical framework and actionable guidelines for inclusive, evidence-informed AI design in vocational contexts.

38 citations7 influentialRead paper

Everybody Prune Now: Structured Pruning of LLMs with only Forward Passes

Feb 08, 2024arXiv.org

Existing structured pruning methods for large language models (LLMs) heavily rely on backpropagation, incurring substantial memory and computational overhead. To address this, we propose Bonsai—the first fully backpropagation-free, gradient-agnostic forward-pass pruning method for LLMs. Bonsai estimates module importance via forward perturbation analysis and performs module-level structured pruning without gradient computation. On a single NVIDIA A6000 GPU, Bonsai efficiently prunes the 8B-parameter LLaMA-3 model at 50% sparsity: memory consumption is reduced to one-half to one-third of conventional backward-based methods; pruning speed doubles; inference latency improves by 100%; and accuracy remains state-of-the-art. By eliminating dependence on gradient computation, Bonsai significantly broadens the feasibility of deploying compressed LLMs on resource-constrained hardware.

25 citations4 influentialRead paper
Recent publications

Latest Papers