Research & Open Source

Projects

Featured Research

Research Code ECCV 2026

TARS

MinMax Token-Adaptive Preference Strategy

A token-adaptive preference optimization method that uses a min-max objective to reduce hallucinations in multimodal large language models.

MLLM ReliabilityPreference Optimization

Project Paper Code

Overview of visual adversarial perturbation for hallucination reduction

Research Code NeurIPS 2025

VAP

Visual Adversarial Perturbation

A visual perturbation method that reduces object hallucinations without modifying the base vision-language model, evaluated across eight models.

Object HallucinationVision-Language Models

Project Paper Poster Code

Debiased high-confidence logit alignment framework

Research Code ICCV 2025

DHAT

Debiased High-Confidence Adversarial Training

Debiased high-confidence logit alignment for adversarial training, with code for model training and robustness evaluation.

Adversarial TrainingLogit Alignment

Paper Code

High-frequency feature disentanglement and recalibration framework

Research Code Neural Networks 2026

HFDR

High-Frequency Disentanglement and Recalibration

A frequency recalibration module that addresses the low-frequency bias induced by adversarial training.

Frequency AnalysisAdversarial Robustness

Paper Code

Research Software

Reusable frameworks and implementations that support reproducible experiments.

Framework

AdverRobust

A configurable PyTorch framework for adversarial training and robustness evaluation across multiple methods, attacks, and image datasets.

PyTorchRobustness Evaluation

Code Quick Start

Multilevel motif-based graph convolution framework

Research Code IEEE TCSS 2024

Motif-based GNN

A multilevel graph convolution framework that combines simple graphs and motifs to represent higher-order topology.

Graph Neural NetworksMotifs

Paper Code

Surveys & Resources

Structured literature collections and practical guides for research and academic writing.

Taxonomy of large-small model collaboration for domain tasks

Survey & Resources

LM-SM Domain Collaboration

A survey and resource collection that organizes large-small model collaboration by transfer direction and inference-time interaction.

Model CollaborationEfficient AI

Paper Repository

Survey & Resources TMLR 2026

Multimodal Token Compression

A modality-centered survey and resource collection for token compression in multimodal large language models.

Token CompressionMultimodal Models

Paper Repository

Practical Guide

Academic Paper Writing

A practical LaTeX guide covering project structure, citations, tables, figures, and conference submission.

LaTeXAcademic Writing

Repository

Additional experiments and reading lists

Browse the full repository collection, including earlier prototypes and maintained paper lists.

View all repositories