Human-centric AI that aims at Interaction between human and ML&DL models via explainability and visualization
Augmenting BERT-based TOD system with financial ontology (Jisoo Jang) – combining ML model and knowledge base [LINK]
Human-centric model training (Jisoo Jang) – new design of NN structure and converting human input to part of the model
Fusion of ML and knowledge base/human input (WSL’s BP project): Bi-direction exchange between ML and knowledge base with flexible converting operators
XAI aims at Interpretation, reliability, verifiability, traceability, auditability, trustworthiness, fairness, and safety of ML&DL models
XAI for professional classic piano music (Jisoo Park, Bjørn Are Therkelsen)
Device Life Expectation Estimation (Jisoo / Chang) [LINK]
ML model reliability testing and safety assessment (TBD)
XAI for high criminal rate (student group project)
Bias analysis for criminal sentencing (XAI class student project)
XAI for factors with time delay effects or continuous effects (WSL, Jisoo Jang, Naomi) scenarios include
The high crime rate, supply chain bottleneck,
Air pollution, global warming. This is a combination of ambient AI and XAI (Cha, WSL)
Sleeping disorder analysis (Chang Sub Chang)
Causal discovery aims to find causal structures by analyzing observational data.
Data partition and parallel causal discovery (Iju Lee)
The architectural design of GNN Ensemble (Duc) – similar to TPOT/AutoML
Graph database system (Iiju, Alex)
Knowledge discovery in supply chain graphic data (Naomi)
Human-centric AI Research Topic
1. Human-AI Collaboration & Personalization
Research that incorporates human knowledge, feedback, and preferences into learning and reasoning, providing support tailored to individual or group contexts.
Human-Driven Active Verification for Efficient and Trustworthy Graph Classification (Learn More)
Scaffolding Deliberation: A Human-in-the-Loop Framework for Guiding LLM Reasoning (Learn More)
Pathfinder: Context Engineering and Knowledge Management for Domain-Specific Horizontal Reasoning (Learn More)
From Reasoning Traces to Best Practice: Context-aware Path Finding on Strategy Maps (Learn More)
TwinBI: An Agentic Digital Twin for Efficient Augmented Interactions with Business Intelligence Dashboards (Learn More)
Piano Performance Evaluation Dataset with Multilevel Perceptual Features (Learn More)
Identifying Critical Segments Affecting Piano Performance Evaluation (Learn More)
Discovering Piano Performance Evaluation Topology: Hierarchical and Semantic Synergies in Features and Segments (Learn More)
Toward Interpretable Machine Learning: Constructing Polynomial Models Based on Feature Interaction Trees (Learn More)
2. Explainable, Trustworthy & Responsible AI
Research that enables people to review and trust AI decisions by making their rationale, uncertainty, evaluation validity, compliance, and sources transparent.
Toward Interpretable Machine Learning: Constructing Polynomial Models Based on Feature Interaction Trees (Learn More)
Generating Real-Time Explanations for GNNs via Multiple Specialty Learners and Online Knowledge Distillation (Learn More)
Human-Driven Active Verification for Efficient and Trustworthy Graph Classification (Learn More)
UAPRM: Uncertainty-Aware Process Reward Models (Learn More)
Certifying Agent Traces: Factorized Validity, Complete Evaluators, and Testable Interfaces (Learn More)
Witness-Carrying Context Updates for Dependency-Aware LLM Agent Runtimes (Learn More)
Beyond Top-k: Reliability-First Context Compilation for Conversational Information Retrieval (Learn More)
Graph-of-Context: Recoverable Fold/Unfold and Scoped Fork for Instrumented Long-Horizon Multi-Turn LLM Agentic Workflows (Learn More)
Constraint-Based Adaptive AI Certificate Document Generation (Learn More)
Identifying Critical Segments Affecting Piano Performance Evaluation (Learn More)
Discovering Piano Performance Evaluation Topology: Hierarchical and Semantic Synergies in Features and Segments (Learn More)
Toward Interpretable Graph Neural Networks via Concept Matching Model (Learn More)
Toward Interpretable Graph Classification via Concept-Focused Structural Correspondence (Learn More)
INGREX: An Interactive Explanation Framework for Graph Neural Networks (Learn More)
3. Human-Centered Data Access & Decision Intelligence
Research that helps non-experts use complex data and models through natural language and intuitive interfaces to make evidence-based decisions.
A Framework for Enhancing Boardroom TODS with Financial Ontology and Embedded Machine Learning (Learn More)
NL2SQL+: An Extended Framework for SQL Generation with On-Demand Embedded ML Functions (Learn More)
Retrieval-Augmented NL2SQL Generation with Data-Centric Query Capsules for Enterprise Applications (Learn More)
Prompt Tuning for Natural Language to SQL with Embedding Fine-Tuning and RAG (Learn More)
AU-RAG: Agent-based Universal Retrieval Augmented Generation (Learn More)
TwinBI: An Agentic Digital Twin for Efficient Augmented Interactions with Business Intelligence Dashboards (Learn More)
Pathfinder: Context Engineering and Knowledge Management for Domain-Specific Horizontal Reasoning (Learn More)
Piano Performance Evaluation Dataset with Multilevel Perceptual Features (Learn More)
Identifying Critical Segments Affecting Piano Performance Evaluation (Learn More)
Discovering Piano Performance Evaluation Topology: Hierarchical and Semantic Synergies in Features and Segments (Learn More)
Efficient ML Lifecycle Transferring for Large-Scale and High-Dimensional Data via Core Set-Based Dataset Similarity (Learn More)
Constraint-Based Adaptive AI Certificate Document Generation (Learn More)
4. Process-Aware & Reliable Agentic AI
Research that enables agents to act beyond generating answers while respecting procedures, dependencies, context, and permissions in verifiable ways.
Strategy Map: Milestone-Guided Process State for Dependency-Sensitive Tool-Use Agents (Learn More)
Trace2Map: Milestone-Guided Context Management for Tool-Use Agents (Learn More)
Pathfinder: Context Engineering and Knowledge Management for Domain-Specific Horizontal Reasoning (Learn More)
Graph-of-Context: Recoverable Fold/Unfold and Scoped Fork for Instrumented Long-Horizon Multi-Turn LLM Agentic Workflows (Learn More)
Witness-Carrying Context Updates for Dependency-Aware LLM Agent Runtimes (Learn More)
Extracting and Constructing Execution-Ready Skeletons via Mining Multi-Source Agent Traces (Learn More)
Certifying Agent Traces: Factorized Validity, Complete Evaluators, and Testable Interfaces (Learn More)
Beyond Top-k: Reliability-First Context Compilation for Conversational Information Retrieval (Learn More)
UAPRM: Uncertainty-Aware Process Reward Models (Learn More)
Scaffolding Deliberation: A Human-in-the-Loop Framework for Guiding LLM Reasoning (Learn More)
From Reasoning Traces to Best Practice: Context-aware Path Finding on Strategy Maps (Learn More)
Constraint-Based Adaptive AI Certificate Document Generation (Learn More)
AU-RAG: Agent-based Universal Retrieval Augmented Generation (Learn More)