Huan Liu

Few-Shot Disaster Classification via Causal Disentanglement and Cross-Modal Flattening

Current AI is too data-hungry for fast disaster response. My research uses causal learning to identify a crisis from just a few examples.

Program: FURI

Accelerating Convergence to Effective Capacity in Lightweight Vision Models via Self-Competitive Distillation

Optimizing AI training through self-competition will bring highly accurate, real-time computer vision to everyday mobile and edge devices.

Program: FURI

Why do Generative AI Models Snitch? Investigating Deceptive Tool-calling Behaviors in Safety-aligned Language Models

Investigating why safety-trained AI models deceive users will help create assistants that users can trust with sensitive data.

Program: GCSP research stipend

EduCBM: Concept Bottleneck Models for Interpretable Education

Teaching AI to explain its reasoning helps stakeholders understand not just what is right, but why it is right.

Program: FURI

Merging Large Language Models: Threats and Opportunities

AI models can be "merged" to expand their knowledge. We assess its effect on safety and detectability in creating trustworthy AI systems.

Program: FURI

LLM-Multimodal Misinformation Detection

Analyzing discrepancies between AI-generated text and photos can improve the detection of fake news created by AI.

Program: FURI

Disambiguating Human Language in Large Language Models and Analyzing its Effect to Improve NLP Task Accuracy

Helping AI comprehend ambiguous human text will make it more reliable and effective in conversations and critical tasks alike.

Program: FURI

Understanding Machine vs Human Generated Text in News Articles

Designing an algorithm to identify if a text article is written by a human or is machine generated will help defend against disinformation.

Program: FURI

A Systematic Study of Pooled Testing Procedures for SARS-CoV-2

Analyzing and optimizing group testing strategies can increase the throughput of COVID-19 testing while reducing time and cost.

Program: FURI

Learning and Sampling from the Causal Graph of Multi Categorical Distributions Via the Generative Adversarial Network

Generating fake data is an important task for data researchers who want to avoid data collection and privacy issues.

Program: FURI

Twitter Bot Detection Software for Disaster Relief Applications

Program: FURI

Disaster Visualization and Tracking Using Social Media Data

Program: FURI

The Lifecycle of Botnets on Social Media

Program: FURI

Personalized Browser History Anonymization

Program: FURI