Ziyuan Yang @ NTU

# AI security

# AI in healthcare

# biometrics

# federated learning

# trustworthy AI

# privacy-preserving learning

# backdoor attack

# adversarial defense

# medical imaging

# representation learning

My Research

My research interests encompass AI Security, AI in Healthcare, Biometrics, Federated Learning, and AI Efficiency. I am particularly interested in building trustworthy and privacy-preserving learning systems that remain effective in real-world healthcare and security scenarios.

* Equal contribution, ^† Corresponding author (for publications)

2026

Zero-Sacrifice Persistent-Robustness Adversarial Defense for Pre-Trained Encoders

Zhuxin Lei, Ziyuan Yang^†, Yi Zhang^†

International Conference on Learning Representations (ICLR), 2026

[TL;DR] [Paper] [Code]

This paper proposes a persistent-robustness adversarial defense for pre-trained encoders that improves robustness without sacrificing clean-task utility.

CCF-A # AI Security # Adversarial Defense # Robust Learning

Zero-Sacrifice Persistent-Robustness Adversarial Defense for Pre-Trained Encoders

Zhuxin Lei, Ziyuan Yang^†, Yi Zhang^†

International Conference on Learning Representations (ICLR), 2026

[TL;DR] [Paper] [Code]

This paper proposes a persistent-robustness adversarial defense for pre-trained encoders that improves robustness without sacrificing clean-task utility.

CCF-A # AI Security # Adversarial Defense # Robust Learning

Poisoned Distillation: Injecting Backdoors Into Distilled Datasets Without Raw Data Access

Ziyuan Yang, Ming Yan, Yi Zhang^†, Joey Tianyi Zhou

AAAI Conference on Artificial Intelligence (AAAI), 2026

[TL;DR] [Paper] [Code]

This work studies how backdoors can be injected into distilled datasets without access to the original raw training data, revealing new security risks for dataset distillation pipelines.

CCF-A # AI Security # Backdoor Attack # Trustworthy AI

Poisoned Distillation: Injecting Backdoors Into Distilled Datasets Without Raw Data Access

Ziyuan Yang, Ming Yan, Yi Zhang^†, Joey Tianyi Zhou

AAAI Conference on Artificial Intelligence (AAAI), 2026

[TL;DR] [Paper] [Code]

This work studies how backdoors can be injected into distilled datasets without access to the original raw training data, revealing new security risks for dataset distillation pipelines.

CCF-A # AI Security # Backdoor Attack # Trustworthy AI

FedPalm: a General Federated Learning Framework for Closed-And Open-Set Palmprint Verification

Ziyuan Yang, Yingyu Chen, Chengrui Gao, Andrew Beng Jin Teoh, Bob Zhang, Yi Zhang^†

IEEE Transactions on Information Forensics and Security (IEEE T-IFS), 2026

[TL;DR] [Paper] [Code]

This paper proposes FedPalm, a unified federated learning framework for palmprint verification that combines personalized local textural experts with a shared global expert to achieve robust performance in both closed-set and open-set scenarios while preserving biometric privacy.

CCF-A JCR-Q1 # Federated Learning # Biometrics

FedPalm: a General Federated Learning Framework for Closed-And Open-Set Palmprint Verification

Ziyuan Yang, Yingyu Chen, Chengrui Gao, Andrew Beng Jin Teoh, Bob Zhang, Yi Zhang^†

IEEE Transactions on Information Forensics and Security (IEEE T-IFS), 2026

[TL;DR] [Paper] [Code]

CCF-A JCR-Q1 # Federated Learning # Biometrics

One CT Unified Model Training Framework to Rule All Scanning Protocols

Fengzhi Xu*, Ziyuan Yang*, Zexin Lu, Yingyu Chen, Fenglei Fan, Hongming Shan, Yi Zhang^†

ACM Multimedia (ACM MM), 2026

Rio de Janeiro, Brazil, Nov. 10-14, 2026

2026

Zero-Sacrifice Persistent-Robustness Adversarial Defense for Pre-Trained Encoders

Zero-Sacrifice Persistent-Robustness Adversarial Defense for Pre-Trained Encoders

Poisoned Distillation: Injecting Backdoors Into Distilled Datasets Without Raw Data Access

Poisoned Distillation: Injecting Backdoors Into Distilled Datasets Without Raw Data Access

FedPalm: a General Federated Learning Framework for Closed-And Open-Set Palmprint Verification

FedPalm: a General Federated Learning Framework for Closed-And Open-Set Palmprint Verification

One CT Unified Model Training Framework to Rule All Scanning Protocols

One CT Unified Model Training Framework to Rule All Scanning Protocols

Deep Learning in Palmprint Recognition: a Comprehensive Survey

Deep Learning in Palmprint Recognition: a Comprehensive Survey

Low-Level Dataset Distillation for Medical Image Enhancement

Low-Level Dataset Distillation for Medical Image Enhancement

LURE: Latent Space Unblocking for Multi-Concept Reawakening in Diffusion Models

LURE: Latent Space Unblocking for Multi-Concept Reawakening in Diffusion Models

Tuning Less, Learning More: toward a More Generalizable Image Manipulation Detector

Tuning Less, Learning More: toward a More Generalizable Image Manipulation Detector

FACT: Fuzzy Alignment with Comorbidity Topology for Reliable Multi-Label Medical Image Diagnosis

FACT: Fuzzy Alignment with Comorbidity Topology for Reliable Multi-Label Medical Image Diagnosis

CPL-IQA: Blind Image Quality Assessment via Convolutional Prototype Learning

CPL-IQA: Blind Image Quality Assessment via Convolutional Prototype Learning

Palmprint De-Identification via Diffusion Model for High-Quality and Diverse Synthesis

Palmprint De-Identification via Diffusion Model for High-Quality and Diverse Synthesis

Harmless Copyright Protection for CT Dataset

Harmless Copyright Protection for CT Dataset

PrismAgent: Illuminating Harm in Memes via a Zero-Shot Interpretable Multi-Agent Framework

PrismAgent: Illuminating Harm in Memes via a Zero-Shot Interpretable Multi-Agent Framework

Causal Gradient Intervention for Debiased and Evidence-Grounded Medical Visual Question Answering

Causal Gradient Intervention for Debiased and Evidence-Grounded Medical Visual Question Answering

Double Banking on Knowledge: a Unified All-In-One Framework for Unpaired Multi-Modality Semi-Supervised Medical Image Segmentation

Double Banking on Knowledge: a Unified All-In-One Framework for Unpaired Multi-Modality Semi-Supervised Medical Image Segmentation

Channel Group-Wise Exciting Network for Video-Based Dorsal Hand Re-Identification in Uncontrollable Environment

Channel Group-Wise Exciting Network for Video-Based Dorsal Hand Re-Identification in Uncontrollable Environment

Multi-Label Medical Diagnosis Using Spatial-Disease Feature Condensation and Kolmogorov–Arnold Layers

Multi-Label Medical Diagnosis Using Spatial-Disease Feature Condensation and Kolmogorov–Arnold Layers

Directly Encrypting DNA Sequences for DNA Storage via Automata Cryptography

Directly Encrypting DNA Sequences for DNA Storage via Automata Cryptography

2025

Patient-Level Anatomy Meets Scanning-Level Physics: Personalized Federated Low-Dose CT Denoising Empowered by Large Language Model

Patient-Level Anatomy Meets Scanning-Level Physics: Personalized Federated Low-Dose CT Denoising Empowered by Large Language Model

Enhancing Federated Learning through Exploring Filter-Aware Relationships and Personalizing Local Structures

Enhancing Federated Learning through Exploring Filter-Aware Relationships and Personalizing Local Structures

Hypernetwork-Based Physics-Driven Personalized Federated Learning for CT Imaging

Hypernetwork-Based Physics-Driven Personalized Federated Learning for CT Imaging

Beyond Static Features: a Novel Dynamic Palmprint Verification Framework Empowered by Generative Models

Beyond Static Features: a Novel Dynamic Palmprint Verification Framework Empowered by Generative Models

Bridging the Divide Between Left and Right Palmprints for Cross-Chirality Verification

Bridging the Divide Between Left and Right Palmprints for Cross-Chirality Verification

A Novel Privacy-Enhancing Framework for Low-Dose CT Denoising

A Novel Privacy-Enhancing Framework for Low-Dose CT Denoising

SF2Net: Sequence Feature Fusion Network for Palmprint Verification

SF2Net: Sequence Feature Fusion Network for Palmprint Verification

Identity and Style Feature Decoupling Network for Cross-Domain Palmprint Recognition

Identity and Style Feature Decoupling Network for Cross-Domain Palmprint Recognition

FedRIR: Rethinking Information Representation in Federated Learning

FedRIR: Rethinking Information Representation in Federated Learning

Single Source Domain Generalization for Palm Biometrics

Single Source Domain Generalization for Palm Biometrics

BGC-Net: Bilateral Graph Convolutional Network for Weakly-Supervised Semantic Segmentation of Large-Scale Point Clouds

BGC-Net: Bilateral Graph Convolutional Network for Weakly-Supervised Semantic Segmentation of Large-Scale Point Clouds

Trustworthy Disentangled Framework for Multi-Label Medical Image Classification with Multimodal Refinement

Trustworthy Disentangled Framework for Multi-Label Medical Image Classification with Multimodal Refinement

Modality Modulation and Dual Consistency for Multi-Modality Semi-Supervised Medical Image Segmentation

Modality Modulation and Dual Consistency for Multi-Modality Semi-Supervised Medical Image Segmentation

Plaintext-Free Deep Learning for Privacy-Preserving Medical Image Analysis through Frequency Information Embedding

Plaintext-Free Deep Learning for Privacy-Preserving Medical Image Analysis through Frequency Information Embedding

Multi-Order Extension Codes for Palmprint Recognition

Multi-Order Extension Codes for Palmprint Recognition

PPD-IQA: a Bottom-Up and Top-Down Combined Approach for Blind Image Quality Assessment via Prototype-Prompted Disentangling

PPD-IQA: a Bottom-Up and Top-Down Combined Approach for Blind Image Quality Assessment via Prototype-Prompted Disentangling

ROPREM: A Prototype-Free Retrieval Method for Automatic Check-Out

ROPREM: A Prototype-Free Retrieval Method for Automatic Check-Out

STAND-Net: Lightning-Fast MRI Reconstruction via Stage-Aware Deep Unrolling with Neural Tensor Decomposition

STAND-Net: Lightning-Fast MRI Reconstruction via Stage-Aware Deep Unrolling with Neural Tensor Decomposition

2024

A Dual-Level Cancelable Framework for Palmprint Verification and Hack-Proof Data Storage

A Dual-Level Cancelable Framework for Palmprint Verification and Hack-Proof Data Storage

Physics-Driven Spectrum-Consistent Federated Learning for Palmprint Verification

Physics-Driven Spectrum-Consistent Federated Learning for Palmprint Verification

Inject Backdoor in Measured Data to Jeopardize Full-Stack Medical Image Analysis System