HP-JEPA: Hierarchical Partitioning for Multi-Resolution Graph Joint-Embedding Predictive Learning
Multi-resolution latent prediction learns transferable graph representations across local and global structural scales.
Publications
You can also find my articles on my Google Scholar profile.
My work spans trustworthy learning over structured data, efficient and aligned multimodal foundation models, and machine learning for scientific discovery.
Trustworthy and Structured Machine Intelligence
Beyond Soft Masks: Hard-Perturbation Mixup Explainer for Robust GNN Explainability
Hard structural perturbations and mixup produce faithful GNN explanations while reducing distribution shift.
Addressing Structural Distribution Shift in Explanations for Graph Neural Networks
A theoretical and empirical treatment of structural distribution shift in GNN explanations, using proxy graphs to preserve explanatory information while remaining in distribution.
Is Your Explanation Reliable: Confidence-Aware Explanation on Graph Neural Networks
Confidence-calibrated framework for reliable GNN explanations, especially under OOD settings.
GE-Chat: A Graph Enhanced RAG Framework for Evidential Response Generation of LLMs
Graph-enhanced RAG pipeline for evidence-grounded LLM responses.
RISE: Radius of Influence based Subgraph Extraction for 3D Molecular Graph Explanation
A 3D GNN explainer that localizes explanations through node-wise radius of influence.
RegExplainer: Generating Explanations for Graph Neural Networks in Regression Tasks
An explanation framework for graph regression models with improved reliability under distribution shift.
LLMExplainer: Large Language Model based Bayesian Inference for Graph Explanation Generation
Uses LLM-guided Bayesian inference to mitigate learning bias in graph explanation.
Interpreting Graph Neural Networks with In-Distributed Proxies
Generates in-distribution proxy graphs to improve faithfulness of GNN explanations.
MixupExplainer: Generalizing Explanations for Graph Neural Networks with Data Augmentation
A data-augmentation-based framework to improve OOD robustness of GNN explanations.
DeMinify: Neural Variable Name Recovery and Type Inference
Neural recovery of variable names and type inference from minified code.
Commit-level, Neural Vulnerability Detection and Assessment
Commit-level vulnerability detection and CVSS assessment via context-aware graph learning.
Efficient Multimodal and Video Foundation Models
Alignment in Multimodal Large Language Models: A Survey
A review of multimodal alignment spanning evidence use, human intent, training methods, and evaluation.
Shot-Aware Frame Sampling for Video Understanding
InfoShot is a training-free, shot-aware sampler that preserves both representative context and brief decision-critical events in long videos.
RB-FT: Rationale-Bootstrapped Fine-Tuning for Video Classification
A two-stage self-improvement framework that uses model-generated rationales to adapt vision-language models to domain-specific video classification without additional annotations.
Machine Learning for Scientific Discovery
SCOPE: Observation-Conditioned Full-Target Prediction for Sparse PDE Inference
Single-pass recovery of complete PDE fields from sparse measurements through latent prediction and physical reconstruction.
PDE-OBS: Controlled Evaluation Across Observation Patterns
A benchmark for evaluating physical-field reconstruction and forecasting as observation density and spatial layout change.