Building controllable and robust AI systems.

I am an undergraduate student at the University of Maryland, currently working on adversarial attacks and related AI research under the guidance of Andi Zhang.

Explore My Research Resume GitHub ↗

Research

Research I am pursuing

My work connects generative models, adversarial learning, and robust neural computation.

Diffusion Models · Adversarial Machine Learning

Subject-Preserving Adversarial Image Generation

I develop and evaluate adversarial-generation pipelines for personalized diffusion models, including segmentation-guided construction, classifier-guided latent optimization, subject-specific LoRA integration, and reproducible GPU experiments.

I design controlled carrier conditions, implement attack variants, run preservation and transfer evaluations, and analyze how attack evidence is distributed outside the primary subject.

Diffusion modelsLoRASegmentationAdversarial ML

Ongoing research

Spiking Neural Network (SNN) Robustness

I implement and evaluate spiking neural networks (SNNs), automate experiments across model checkpoints and configurations, and debug attack, detection, and reprogramming pipelines.

My analysis covers SNN robustness, detector behavior, temporal spiking activity, latency, and operation cost while keeping the experimental framework extensible to new datasets and attack families.

PyTorchSNNsReprogrammingRobustness evaluation

Publications

Selected research

2027

Let the Carrier Carry the Attack: Preserving the Subject in Adversarial Image Generation

A carrier-guided framework for preserving personalized subjects while supporting targeted adversarial image generation and cross-model transfer.

arXiv ↗

Qualitative comparison of No-Carrier, Non-Target Carrier, and Target Carrier adversarial image generation — Figure 1. Qualitative comparison of the No-Carrier baseline and carrier-based conditions.

2027

Rethinking Inter-LoRA Orthogonality in Adapter Merging: Insights from Orthogonal Monte Carlo Dropout

Title visually masked until the submission deadline.

AISTATS — on submission

Selected projects

From models to practical systems

Machine learning

Photovoltaic Output Forecasting

A CNN-LSTM pipeline for ultra-short-term time-series prediction and analysis of photovoltaic power output.

Python · CNN-LSTM · Time series

Cybersecurity

COFE Coin Security Mechanism Optimization

A security-focused study of the COFE Coin mechanism, examining system risks and proposing improvements to strengthen its protection model.

Digital currency · Security analysis · Mechanism design

Security systems

Centralized Security Dashboard

An executive-facing dashboard plan for consolidating security events, vulnerabilities, and organizational risk signals.

Dashboard design · Security · Project planning

Responsible AI

NIST AI Risk Management Accessibility

Concise inductive and deductive approaches for making AI risk-management concepts easier to apply in model assessment.

NIST AI RMF · Model assessment

Contact

Contact Me

I am actively seeking PhD positions and welcome research collaborations in AI and AI security.

ljiang19@umd.eduWeChat · dj884290GitHub ↗