Haizhong Zheng
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Haizhong Zheng has a total of 25 co-authors. Here are the top co-authors by number of shared publications:
- Atul Prakash
University of Michigan, Ann Arbor, USA
4 shared publications - Alanson P. Sample
University of Michigan, Ann Arbor, USA
1 shared publication - Beidi Chen
Carnegie Mellon University
1 shared publication - Beijie Liu
University of Southern California, Los Angeles, CA, USA
1 shared publication Bhavya Kailkhura
Lawrence Livermore National Laboratory
1 shared publicationBokai Lai
School of Informatics Xiamen University,Department of Artificial Intelligence,Xiamen,China
1 shared publication
Latest Publications
- Open Access
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-23
EchoScriptor: Automatic Lifelogging Narratives via Activity-Based Audio–Language Model
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-23
Open AccessEchoScriptor: Automatic Lifelogging Narratives via Activity-Based Audio–Language Model
- Open Access
- 22 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 2212-2231
Wave : Leveraging Architecture Observation for Privacy-Preserving Model Oversight
Large Language Models (LLMs) inference increasingly require mechanisms that provide runtime visibility into what is actually executing, without exposing model weights or code. We present WAVE, a hardware-grounded monitoring framework that leverages GPU performance counters (PMCs) to observe LLM inference. WAVE is built on the insight that legitimate executions of a given model must satisfy hardware-constrained invariants, such as memory accesses, instruction mix, and tensor-core utilization, induced by the model's linear-algebraic structure. WAVE collects lightweight PMC traces and applies a two-stage pipeline: (1) inferring architectural properties (e.g., parameter count, layer depth, hidden dimension, batch size) from the observed traces; and (2) using an SMT-based consistency checker to assess whether the execution aligns with the provisioned compute and the claimed model's constraints. We evaluate WAVE on common open-source LLM architectures, such as LLaMA, GPT, and Qwen, across multiple GPU architectures, including NVIDIA Ada Lovelace, Hopper, and Blackwell. Results show that WAVE recovers key model parameters with an average error of 6.8% and identifies disguised executions under realistic perturbations. By grounding oversight in hardware invariants, WAVE provides a practical avenue for continuous, privacy-preserving runtime monitoring of LLM services.
- 22 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 2212-2231
Open AccessWave : Leveraging Architecture Observation for Privacy-Preserving Model Oversight
Large Language Models (LLMs) inference increasingly require mechanisms that provide runtime visibility into what is actually executing, without exposing model weights or code. We present WAVE, a hardware-grounded monitoring framework that leverages GPU performance counters (PMCs) to observe LLM inference. WAVE is built on the insight that legitimate executions of a given model must satisfy hardware-constrained invariants, such as memory accesses, instruction mix, and tensor-core utilization, induced by the model's linear-algebraic structure. WAVE collects lightweight PMC traces and applies a two-stage pipeline: (1) inferring architectural properties (e.g., parameter count, layer depth, hidden dimension, batch size) from the observed traces; and (2) using an SMT-based consistency checker to assess whether the execution aligns with the provisioned compute and the claimed model's constraints. We evaluate WAVE on common open-source LLM architectures, such as LLaMA, GPT, and Qwen, across multiple GPU architectures, including NVIDIA Ada Lovelace, Hopper, and Blackwell. Results show that WAVE recovers key model parameters with an average error of 6.8% and identifies disguised executions under realistic perturbations. By grounding oversight in hardware invariants, WAVE provides a practical avenue for continuous, privacy-preserving runtime monitoring of LLM services.
- 3 November 2024
- Published by Springer Nature
- p. 166-182
Leveraging Hierarchical Feature Sharing for Efficient Dataset Condensation
- 3 November 2024
- Published by Springer Nature
- p. 166-182
Leveraging Hierarchical Feature Sharing for Efficient Dataset Condensation
- 28 August 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-8
Meta Combination Framework for 3D Point Cloud Analysis with Distillation
Recent advancements in 3D point cloud analysis have introduced various network architectures to the field. Nevertheless, the absence of a standard framework for describing these networks makes it difficult to systematically compare and analyze them. Furthermore, the network size of these models tends to be large due to the emergence of Large Language Models. We propose a method called MetaPointDistill that employs unified building blocks and model compression techniques to mitigate the aforementioned issues. To explain, when comparing systems, we need a unified framework structure for fair and convenient comparison. At the same time, the scale of this framework cannot be too large, so model compression technology needs to be used. Specifically, we partition the building blocks into Neighborhood Updater, Neighborhood Aggregator, Point Updater, and Spatial Embedding Generator, and explore their optimal combinations. Additionally, we employ knowledge distillation to obtain a model with improved performance and reduced size. We conduct experiments on multiple tasks to evaluate the efficacy of our framework. The results demonstrate that MetaPointDistill outperforms prior state-of-the-art methods by 1.2% mIoU and 0.4% OA while utilizing only 13% and 35% FLOPs on the S3DIS and ScanObjectNN datasets, respectively. Overall, our proposed framework exhibits significant potential for addressing challenges encountered in practical production processes.
- 28 August 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-8
Meta Combination Framework for 3D Point Cloud Analysis with Distillation
Recent advancements in 3D point cloud analysis have introduced various network architectures to the field. Nevertheless, the absence of a standard framework for describing these networks makes it difficult to systematically compare and analyze them. Furthermore, the network size of these models tends to be large due to the emergence of Large Language Models. We propose a method called MetaPointDistill that employs unified building blocks and model compression techniques to mitigate the aforementioned issues. To explain, when comparing systems, we need a unified framework structure for fair and convenient comparison. At the same time, the scale of this framework cannot be too large, so model compression technology needs to be used. Specifically, we partition the building blocks into Neighborhood Updater, Neighborhood Aggregator, Point Updater, and Spatial Embedding Generator, and explore their optimal combinations. Additionally, we employ knowledge distillation to obtain a model with improved performance and reduced size. We conduct experiments on multiple tasks to evaluate the efficacy of our framework. The results demonstrate that MetaPointDistill outperforms prior state-of-the-art methods by 1.2% mIoU and 0.4% OA while utilizing only 13% and 35% FLOPs on the S3DIS and ScanObjectNN datasets, respectively. Overall, our proposed framework exhibits significant potential for addressing challenges encountered in practical production processes.
- Open Access
Coverage-centric Coreset Selection for High Pruning Rates
One-shot coreset selection aims to select a representative subset of the training data, given a pruning rate, that can later be used to train future models while retaining high accuracy. State-of-the-art coreset selection methods pick the highest importance examples based on an importance metric and are found to perform well at low pruning rates. However, at high pruning rates, they suffer from a catastrophic accuracy drop, performing worse than even random sampling. This paper explores the reasons behind this accuracy drop both theoretically and empirically. We first propose a novel metric to measure the coverage of a dataset on a specific distribution by extending the classical geometric set cover problem to a distribution cover problem. This metric helps explain why coresets selected by SOTA methods at high pruning rates perform poorly compared to random sampling because of worse data coverage. We then propose a novel one-shot coreset selection method, Coverage-centric Coreset Selection (CCS), that jointly considers overall data coverage upon a distribution as well as the importance of each example. We evaluate CCS on five datasets and show that, at high pruning rates (e.g., 90%), it achieves significantly better accuracy than previous SOTA methods (e.g., at least 19.56% higher on CIFAR10) as well as random selection (e.g., 7.04% higher on CIFAR10) and comparable accuracy at low pruning rates. We make our code publicly available at https://github.com/haizhongzheng/Coverage-centric-coreset-selection.
Open AccessCoverage-centric Coreset Selection for High Pruning Rates
One-shot coreset selection aims to select a representative subset of the training data, given a pruning rate, that can later be used to train future models while retaining high accuracy. State-of-the-art coreset selection methods pick the highest importance examples based on an importance metric and are found to perform well at low pruning rates. However, at high pruning rates, they suffer from a catastrophic accuracy drop, performing worse than even random sampling. This paper explores the reasons behind this accuracy drop both theoretically and empirically. We first propose a novel metric to measure the coverage of a dataset on a specific distribution by extending the classical geometric set cover problem to a distribution cover problem. This metric helps explain why coresets selected by SOTA methods at high pruning rates perform poorly compared to random sampling because of worse data coverage. We then propose a novel one-shot coreset selection method, Coverage-centric Coreset Selection (CCS), that jointly considers overall data coverage upon a distribution as well as the importance of each example. We evaluate CCS on five datasets and show that, at high pruning rates (e.g., 90%), it achieves significantly better accuracy than previous SOTA methods (e.g., at least 19.56% higher on CIFAR10) as well as random selection (e.g., 7.04% higher on CIFAR10) and comparable accuracy at low pruning rates. We make our code publicly available at https://github.com/haizhongzheng/Coverage-centric-coreset-selection.
- Open Access
Efficient Adversarial Training with Transferable Adversarial Examples
Adversarial training is an effective defense method to protect classification models against adversarial attacks. However, one limitation of this approach is that it can require orders of magnitude additional training time due to high cost of generating strong adversarial examples during training. In this paper, we first show that there is high transferability between models from neighboring epochs in the same training process, i.e., adversarial examples from one epoch continue to be adversarial in subsequent epochs. Leveraging this property, we propose a novel method, Adversarial Training with Transferable Adversarial Examples (ATTA), that can enhance the robustness of trained models and greatly improve the training efficiency by accumulating adversarial perturbations through epochs. Compared to state-of-the-art adversarial training methods, ATTA enhances adversarial accuracy by up to 7.2% on CIFAR10 and requires 12~14x less training time on MNIST and CIFAR10 datasets with comparable model robustness.
Open AccessEfficient Adversarial Training with Transferable Adversarial Examples
Adversarial training is an effective defense method to protect classification models against adversarial attacks. However, one limitation of this approach is that it can require orders of magnitude additional training time due to high cost of generating strong adversarial examples during training. In this paper, we first show that there is high transferability between models from neighboring epochs in the same training process, i.e., adversarial examples from one epoch continue to be adversarial in subsequent epochs. Leveraging this property, we propose a novel method, Adversarial Training with Transferable Adversarial Examples (ATTA), that can enhance the robustness of trained models and greatly improve the training efficiency by accumulating adversarial perturbations through epochs. Compared to state-of-the-art adversarial training methods, ATTA enhances adversarial accuracy by up to 7.2% on CIFAR10 and requires 12~14x less training time on MNIST and CIFAR10 datasets with comparable model robustness.
- Open Access
Analyzing the Interpretability Robustness of Self-Explaining Models
Recently, interpretable models called self-explaining models (SEMs) have been proposed with the goal of providing interpretability robustness. We evaluate the interpretability robustness of SEMs and show that explanations provided by SEMs as currently proposed are not robust to adversarial inputs. Specifically, we successfully created adversarial inputs that do not change the model outputs but cause significant changes in the explanations. We find that even though current SEMs use stable co-efficients for mapping explanations to output labels, they do not consider the robustness of the first stage of the model that creates interpretable basis concepts from the input, leading to non-robust explanations. Our work makes a case for future work to start examining how to generate interpretable basis concepts in a robust way.
Open AccessAnalyzing the Interpretability Robustness of Self-Explaining Models
Recently, interpretable models called self-explaining models (SEMs) have been proposed with the goal of providing interpretability robustness. We evaluate the interpretability robustness of SEMs and show that explanations provided by SEMs as currently proposed are not robust to adversarial inputs. Specifically, we successfully created adversarial inputs that do not change the model outputs but cause significant changes in the explanations. We find that even though current SEMs use stable co-efficients for mapping explanations to output labels, they do not consider the robustness of the first stage of the model that creates interpretable basis concepts from the input, leading to non-robust explanations. Our work makes a case for future work to start examining how to generate interpretable basis concepts in a robust way.