Kaidi Xu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Kaidi Xu has a total of 152 co-authors. Here are the top co-authors by number of shared publications:
Jinhao Duan
Department of Computer Science, University of North Carolina at Chapel Hill, Chapel Hill, USA
10 shared publications- Xiaoshuang Shi
University of Electronic Science and Technology of China
6 shared publications - Jinhao Duan
Drexel University
5 shared publications Chenan Wang
Drexel University
4 shared publications- Hao Cheng
The Hong Kong University of Science and Technology (Guangzhou)
4 shared publications Bhavya Kailkhura
Lawrence Livermore National Laboratory, Livermore, USA
3 shared publications
Latest Publications
- Open Access
- 30 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-11
MyoFLAME: Inverse-Physics Muscle Activation for Statistical 3D Face Models
- 30 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-11
Open AccessMyoFLAME: Inverse-Physics Muscle Activation for Statistical 3D Face Models
- Open Access
- 1 November 2026
- Published by MDPI AG in Electronics
- Vol. 15 (21) , 5064
CATT: Complexity-Aware Thinker–Talker Orchestration for Low-Latency Spoken Responses in Digital Humans
Digital humans must speak promptly, but thinking-enabled large language models (LLMs) answer only after reasoning. CATT (Complexity-Aware Thinker–Talker) routes easy turns to a direct reply; hard turns get an answer-free acknowledgment, progress speech from background reasoning, then the final answer. It needs no fine-tuning and is plug-and-play: its orchestration code runs unchanged when the model endpoint, router, or speech and playback layer is replaced. Evaluated on the transcript-to-speech subsystem (local Qwen3-32B, 738 isolated turns, and virtual playback), CATT starts speaking about 6.3 s earlier than a prompt-matched thinking baseline. The first substantive content comes about 10.5 s earlier on closed-form math and science questions and 3.0 s earlier on open-ended requests (about 1 s later when routed to reasoning). The final closed-form answer arrives about 3.2 s later at the same accuracy. An LLM-judged listening rating, a proxy that includes waiting, places CATT about 1 point above the thinking baseline but 0.42 below a non-thinking baseline on open-ended requests. Both latency gains held for three model families accessed through hosted APIs and in an end-to-end avatar case study. A limited ReVerT reproduction on the same reasoning started speaking 8.4 s after CATT. Real users, microphones, networks and multi-user load remain untested.
- 1 November 2026
- Published by MDPI AG in Electronics
- Vol. 15 (21) , 5064
Open AccessCATT: Complexity-Aware Thinker–Talker Orchestration for Low-Latency Spoken Responses in Digital Humans
Digital humans must speak promptly, but thinking-enabled large language models (LLMs) answer only after reasoning. CATT (Complexity-Aware Thinker–Talker) routes easy turns to a direct reply; hard turns get an answer-free acknowledgment, progress speech from background reasoning, then the final answer. It needs no fine-tuning and is plug-and-play: its orchestration code runs unchanged when the model endpoint, router, or speech and playback layer is replaced. Evaluated on the transcript-to-speech subsystem (local Qwen3-32B, 738 isolated turns, and virtual playback), CATT starts speaking about 6.3 s earlier than a prompt-matched thinking baseline. The first substantive content comes about 10.5 s earlier on closed-form math and science questions and 3.0 s earlier on open-ended requests (about 1 s later when routed to reasoning). The final closed-form answer arrives about 3.2 s later at the same accuracy. An LLM-judged listening rating, a proxy that includes waiting, places CATT about 1 point above the thinking baseline but 0.42 below a non-thinking baseline on open-ended requests. Both latency gains held for three model families accessed through hosted APIs and in an end-to-end avatar case study. A limited ReVerT reproduction on the same reasoning started speaking 8.4 s after CATT. Real users, microphones, networks and multi-user load remain untested.
- Open Access
- 7 October 2026
- Published by Springer Nature in npj Artificial Intelligence
FramePoison: attacks on medical RAG that target framing, not just facts
- 7 October 2026
- Published by Springer Nature in npj Artificial Intelligence
Open AccessFramePoison: attacks on medical RAG that target framing, not just facts
- Open Access
- 25 July 2026
- Published by Springer Nature in npj Artificial Intelligence
- Vol. 2 (1)
SURE: data-efficient safety guardrailing via internal representations and uncertainty-weighted pseudo-labels
- 25 July 2026
- Published by Springer Nature in npj Artificial Intelligence
- Vol. 2 (1)
Open AccessSURE: data-efficient safety guardrailing via internal representations and uncertainty-weighted pseudo-labels
- 25 June 2026
- Published by Springer Nature
- p. 414-423
Can Multimodal LLMs ‘See’ Science Instruction? Benchmarking Pedagogical Reasoning in K–12 Classroom Videos
- 25 June 2026
- Published by Springer Nature
- p. 414-423
Can Multimodal LLMs ‘See’ Science Instruction? Benchmarking Pedagogical Reasoning in K–12 Classroom Videos
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 5010-5019
Transfer Attack for Bad and Good: Explain and Boost Adversarial Transferability across Multimodal Large Language Models
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 5010-5019
Open AccessTransfer Attack for Bad and Good: Explain and Boost Adversarial Transferability across Multimodal Large Language Models
- Open Access
Conformal Lesion Segmentation for 3D Medical Images
Medical image segmentation serves as a critical component of precision medicine, enabling accurate localization and delineation of pathological regions, such as lesions. However, existing models empirically apply fixed thresholds (e.g., 0.5) to differentiate lesions from the background, offering no statistical guarantees on key metrics such as the false negative rate (FNR). This lack of principled risk control undermines their reliable deployment in high-stakes clinical applications, especially in challenging scenarios like 3D lesion segmentation (3D-LS). To address this issue, we propose a risk-constrained framework, termed Conformal Lesion Segmentation (CLS), that calibrates data-driven thresholds via conformalization to ensure the test-time FNR remains below a target tolerance $\varepsilon$ under desired risk levels. CLS begins by holding out a calibration set to analyze the threshold setting for each sample under the FNR tolerance, drawing on the idea of conformal prediction. We define an FNR-specific loss function and identify the critical threshold at which each calibration data point just satisfies the target tolerance. Given a user-specified risk level $α$, we then determine the approximate $1-α$ quantile of all the critical thresholds in the calibration set as the test-time confidence threshold. By conformalizing such critical thresholds, CLS generalizes the statistical regularities observed in the calibration set to new test data, providing rigorous FNR constraint while yielding more precise and reliable segmentations. We validate the statistical soundness and predictive performance of CLS on six 3D-LS datasets across five backbone models, and conclude with actionable insights for deploying risk-aware segmentation in clinical practice.Open AccessConformal Lesion Segmentation for 3D Medical Images
Medical image segmentation serves as a critical component of precision medicine, enabling accurate localization and delineation of pathological regions, such as lesions. However, existing models empirically apply fixed thresholds (e.g., 0.5) to differentiate lesions from the background, offering no statistical guarantees on key metrics such as the false negative rate (FNR). This lack of principled risk control undermines their reliable deployment in high-stakes clinical applications, especially in challenging scenarios like 3D lesion segmentation (3D-LS). To address this issue, we propose a risk-constrained framework, termed Conformal Lesion Segmentation (CLS), that calibrates data-driven thresholds via conformalization to ensure the test-time FNR remains below a target tolerance $\varepsilon$ under desired risk levels. CLS begins by holding out a calibration set to analyze the threshold setting for each sample under the FNR tolerance, drawing on the idea of conformal prediction. We define an FNR-specific loss function and identify the critical threshold at which each calibration data point just satisfies the target tolerance. Given a user-specified risk level $α$, we then determine the approximate $1-α$ quantile of all the critical thresholds in the calibration set as the test-time confidence threshold. By conformalizing such critical thresholds, CLS generalizes the statistical regularities observed in the calibration set to new test data, providing rigorous FNR constraint while yielding more precise and reliable segmentations. We validate the statistical soundness and predictive performance of CLS on six 3D-LS datasets across five backbone models, and conclude with actionable insights for deploying risk-aware segmentation in clinical practice. - Open Access
- 8 September 2025
- Published by Association for Computing Machinery (ACM)
- p. 37-42
LIFT: Automating Symbolic Execution Optimization with Large Language Models for AI Networks
- 8 September 2025
- Published by Association for Computing Machinery (ACM)
- p. 37-42
Open AccessLIFT: Automating Symbolic Execution Optimization with Large Language Models for AI Networks
- Open Access
LIFT: Automating Symbolic Execution Optimization with Large Language Models for AI Networks
Dynamic Symbolic Execution (DSE) is a key technique in program analysis, widely used in software testing, vulnerability discovery, and formal verification. In distributed AI systems, DSE plays a crucial role in identifying hard-to-detect bugs, especially those arising from complex network communication patterns. However, traditional approaches to symbolic execution are often hindered by scalability issues and inefficiencies, particularly in large-scale systems. This paper introduces LIFT (Large-language-model Integrated Functional-equivalent-IR Transformation), a novel framework that leverages Large Language Models (LLMs) to automate the optimization of Intermediate Representations (IRs) in symbolic execution. LIFT addresses the challenges of symbolic execution by providing a scalable, context-sensitive solution for IR transformation. The framework consists of two phases: IR Analysis and Optimization, where LLMs optimize time-intensive IR blocks, and Symbolic Execution and Validation, which includes benchmarking and semantic verification to ensure correctness and generalizability. Experiments on real-world binaries demonstrated significant performance improvements, including a 53.5\% reduction in execution time for bigtest and a 10.24\% reduction for random, along with reductions in IR statements, PUT instructions, and temporary variables. These results demonstrate that LLMs simplify IRs while maintaining functional correctness, enhancing symbolic execution in distributed AI systems.
Open AccessLIFT: Automating Symbolic Execution Optimization with Large Language Models for AI Networks
Dynamic Symbolic Execution (DSE) is a key technique in program analysis, widely used in software testing, vulnerability discovery, and formal verification. In distributed AI systems, DSE plays a crucial role in identifying hard-to-detect bugs, especially those arising from complex network communication patterns. However, traditional approaches to symbolic execution are often hindered by scalability issues and inefficiencies, particularly in large-scale systems. This paper introduces LIFT (Large-language-model Integrated Functional-equivalent-IR Transformation), a novel framework that leverages Large Language Models (LLMs) to automate the optimization of Intermediate Representations (IRs) in symbolic execution. LIFT addresses the challenges of symbolic execution by providing a scalable, context-sensitive solution for IR transformation. The framework consists of two phases: IR Analysis and Optimization, where LLMs optimize time-intensive IR blocks, and Symbolic Execution and Validation, which includes benchmarking and semantic verification to ensure correctness and generalizability. Experiments on real-world binaries demonstrated significant performance improvements, including a 53.5\% reduction in execution time for bigtest and a 10.24\% reduction for random, along with reductions in IR statements, PUT instructions, and temporary variables. These results demonstrate that LLMs simplify IRs while maintaining functional correctness, enhancing symbolic execution in distributed AI systems.
- 7 July 2025
- Published by Drexel University Libraries
Know when to trust
Though Foundation Models (FMs), such as Large Language Models (LLMs) and Large Vision-Language Models (LVLMs), are demonstrated to be powerful in various real-world tasks, it is essential to know when users should trust their decisions in safety-critical scenarios. In this dissertation, I study the reliable decision-making of FMs from three perspectives: 1. Know when FM decisions to be correct (Chapter 2), i.e., how to identify the correctness of model responses? To achieve this, I develop advanced Uncertainty Quantification (UQ) of LLMs. Specifically, I study how the uncertainty should be expressed in a single-turn question-answering format, and how the uncertainty should be propagated in multi-step agentic decision-making processes. 2. Know how to correct FM Decisions (Chapter 3), i.e., how to correct model decisions to be truthful? To handle this, I first build uncertainty-driven hallucination detection methods on FMs. I then develop intervention mechanisms that utilize hallucination detection as rewards to encourage truthful decoding. These interventions are conducted from either the latent subspace or the output space of FMs, over both visual captioning and strategic reasoning benchmarks. 3. Know when to trust FM decisions (Chapter 4), i.e., the trustworthiness of FMs. I study the general trustworthiness, such as toxicity, fairness, and adversarial robustness, of LLMs and compare how real-world deployment, e.g., model pruning and quantization, affects trustworthy behaviors. I also study membership inference privacy issues in image generation models such as Diffusion Models (DMs) and design new membership inference attacks for these models.
- 7 July 2025
- Published by Drexel University Libraries
Know when to trust
Though Foundation Models (FMs), such as Large Language Models (LLMs) and Large Vision-Language Models (LVLMs), are demonstrated to be powerful in various real-world tasks, it is essential to know when users should trust their decisions in safety-critical scenarios. In this dissertation, I study the reliable decision-making of FMs from three perspectives: 1. Know when FM decisions to be correct (Chapter 2), i.e., how to identify the correctness of model responses? To achieve this, I develop advanced Uncertainty Quantification (UQ) of LLMs. Specifically, I study how the uncertainty should be expressed in a single-turn question-answering format, and how the uncertainty should be propagated in multi-step agentic decision-making processes. 2. Know how to correct FM Decisions (Chapter 3), i.e., how to correct model decisions to be truthful? To handle this, I first build uncertainty-driven hallucination detection methods on FMs. I then develop intervention mechanisms that utilize hallucination detection as rewards to encourage truthful decoding. These interventions are conducted from either the latent subspace or the output space of FMs, over both visual captioning and strategic reasoning benchmarks. 3. Know when to trust FM decisions (Chapter 4), i.e., the trustworthiness of FMs. I study the general trustworthiness, such as toxicity, fairness, and adversarial robustness, of LLMs and compare how real-world deployment, e.g., model pruning and quantization, affects trustworthy behaviors. I also study membership inference privacy issues in image generation models such as Diffusion Models (DMs) and design new membership inference attacks for these models.