Page Not Found
Page not found. Your pixels are in another canvas.
Sitemap
A list of all the posts and pages found on the site. For you robots out there is an XML version available for digesting as well.
Pages
About me
Yizhu Wen — Ph.D. student in Computer Science at Indiana University Bloomington, working on AI security.
Posts
publications
Independent Modulation of Discrete-Time Markov Chain Signals Generated from Tunneling Magnetoresistance Effect
Published in IEEE Electron Device Letters (IEEE EDL'24), 2024
Independent modulation of discrete-time Markov chain signals generated from the tunneling magnetoresistance effect.
Download here
DiffImpute: Tabular Data Imputation with Denoising Diffusion Probabilistic Model
Published in IEEE International Conference on Multimedia and Expo (ICME'24), 2024
A denoising diffusion probabilistic model for imputing missing values in tabular data, with four tailored denoising backbones (MLP, ResNet, Transformer, U-Net).
Download here
Secure-IRS: Defending Against Adversarial Physical-Layer Sensing in ISAC System
Published in International Conference on Computing, Networking and Communications (ICNC'25), 2025
Uses an intelligent reflecting surface to defend integrated sensing and communication systems against adversarial physical-layer sensing.
Download here
Web Intellectual Property at Risk: Preventing Unauthorized Real-Time Retrieval by Large Language Models
Published in Conference on Empirical Methods in Natural Language Processing, Main Conference (EMNLP'25), 2025
Protecting web content from being retrieved and reused in real time by large language models without authorization.
Download here
Mental Damage: Caption Poisoning Attacks on Retrieval-Augmented Text-to-Music Generation
Published in IEEE Symposium on Security and Privacy Workshops, ArtSec (IEEE S&P'26 Workshops), 2026
Poisoning the captions in a retrieval corpus steers a retrieval-augmented text-to-music pipeline toward an attacker’s target while staying aligned with the user’s query.
Download here
Position: Generative Engine Optimization Creates Underexamined Risks, Governance Must Target Concentration, Disclosure, and Academic Blind Spots
Published in International Conference on Machine Learning (ICML'26), 2026
A position paper arguing that the governance of Generative Engine Optimization must target market concentration, disclosure, and blind spots in academic evaluation.
Download here
Ultrasound Watermark: Real-time Acoustic Watermarking against Voice Scams on Smartphones
Published in The 9th ACM/IEEE Workshop on Security and Privacy in Edge Computing (EdgeSP'26), 2026
A real-time acoustic watermarking scheme that runs on commodity smartphones to defend against voice scams.
EchoGuard: Continuous Ultrasound-based Human Presence Verification Against AI Agents
Published in The 9th ACM/IEEE Workshop on Security and Privacy in Edge Computing (EdgeSP'26), 2026
Continuously verifies that a real human — not an automated AI agent — is present at a device, using ultrasound sensing.
WAV2PWN: Leveraging Adversarial Transfer Patterns for Surrogate Selection in Black-Box SSL-Based ASR Attacks
Published in Under review, 2027
Choosing surrogate models for black-box attacks on self-supervised speech recognition by exploiting adversarial transfer patterns.
AuscultRAG: Retrieval-Augmented Audio-LLM for Cardiopulmonary Condition Classification
Published in Under review, 2027
A retrieval-augmented audio LLM for classifying cardiopulmonary conditions from auscultation recordings.
Mitigating Cross-Tier Poisoning in Hierarchical Permission-Aware RAG
Published in Under review, 2027
Defending hierarchical, permission-aware retrieval-augmented generation against poisoning that crosses permission tiers.
Claim–Evidence Prompt Injection for Targeted Answer Contamination in Retrieval-Augmented Generation
Published in Under review, 2027
A prompt-injection attack that pairs a claim with fabricated supporting evidence to contaminate specific answers in RAG pipelines.
PoisonedEar: Knowledge Poisoning Attacks on RAG Multimodal Reasoning in Audio-centric Language Models
Published in Under review, 2027
Knowledge poisoning attacks against retrieval-augmented multimodal reasoning in audio-centric language models.
Real-Time Speech Watermarking for Live Voice Communication: Synthetic Speech Provenance and Post-Hoc Recording Attribution
Published in Under review, 2027
Watermarking live voice communication in real time, for both synthetic-speech provenance and post-hoc attribution of recordings.
Download here
SoK: Is Audio Watermarking Robust to Removal Attacks?
Published in Under review, 2027
A taxonomy of 22 audio watermarking schemes and a large-scale measurement study across 22 classes of removal attacks (109 configurations). No surveyed scheme withstands all of them.
Download here