EWHA WOMANS UNIVERSITY
 G-LAMP Project Group

[Research Objectives]

● Addressing Data Challenges: Modern AI systems increasingly encounter non-typical data or data that deviates from ideal assumptions.

● Theoretical Innovation: We aim to develop methodologies and theories using mathematics and statistics to solve these issues and enable effective data analysis.

● Expanding Foundations: Our research supplements traditional theories based on ideal assumptions, establishing a robust foundation for analyzing diverse and complex datasets.

[Research Content]

1. Development of Math-Stat Based AI Theory & AI-Driven Mathematical Theory 

● Based on multidisciplinary knowledge in mathematics and statistics—including differential geometry, probability, number theory, cryptography, and analysis—we develop methodologies for non-parametric regression, neural networks, distribution inference, and cryptographic models that handle non-typical data without requiring ideal assumptions. Furthermore, we aim to develop AI algorithms and rigorous analytical methods to explore and clarify the mathematical properties of special function formulas, arithmetic functions, product structures of Gauss sums, and mathematical physics models.

▷ Sub-Project A1: Development of Neural Network-Based Distribution-Free Statistical Inference and Uncertainty Quantification Methodologies

▷ Sub-Project A3: Analysis of Arithmetic and Special Functions Using AI Models

▷ Sub-Project A4: Cryptographic Foundation Research for Secure Machine Learning

▷ Sub-Project A7: Derivation of Dispersive Equations from Many-Body Schrödinger Dynamics via AI-Based Structure Exploration and Rigorous Analysis

2. Advanced Learning and Inference Methodologies through Math-Stat Behavioral Analysis of Generative & Discriminative AI 

● We systematically analyze the mathematical and statistical behavior of Generative and Discriminative models. Based on these findings, we focus on establishing effective machine learning methodologies under restricted environments and developing fair, reliable generative AI models.

▷ Sub-Project A5: Establishment of Learning and Analysis Methodologies for Discriminative AI under Restricted Conditions

▷ Sub-Project A6: Mathematical and Statistical Analysis of Data Generation Processes in Generative AI

[Research Objectives]

● Securing Trust and Transparency: We define four critical technical challenges (Explainability, Fairness, Versatility, and Interactivity) essential for the coexistence of humans and Artificial General Intelligence (AGI) to develop AI technologies that ensure reliability and transparency.

● Developing Innovative Solutions: Our goal is to provide solutions that enhance the safety and utility of AI. This includes developing Large Language Models (LLMs) that detect and mitigate bias, Multi-modal AI with superior scalability and adaptability, Generative AI for human-agent interaction, and Explainable "White-box" AI technologies.

[Research Content]

1. Fair Next-Generation AI and Privacy Protection Technology

 - Mitigating Bias: Development of benchmarks to measure socio-cultural and linguistic biases and technologies for bias mitigation.

 - Advanced Strategies: Design of bias evaluation metrics and visualization tools, alongside the advancement of bias mitigation strategies based on prompt engineering and optimization.

2. Next-Generation Multi-modal Generative AI for Human and Environmental Interaction

 - Robust Alignment: Development of topology-aware alignment technologies in non-congruent multi-modal environments and reinforcement of privacy and data security. 

 - Affective Computing: Development of technologies for interpreting multi-modal expressions based on user emotions and controlling generated responses accordingly.

3. Explainable Next-Generation Multi-modal AI Technology

 - White-box Modeling: Development of explainable "White-box" multi-modal models that provide transparent internal logic.

 - Global Validation: Verification and advancement of explainable models utilizing diverse global multi-modal datasets.

[Research Objectives]

● Solving Real-World Challenges: We aim to develop AI-based convergence technologies that address various realistic issues in physical, social, and biological environments to realize human-centric environmental responses.

● Developing Core Technologies: To achieve this, we focus on:

    ◆ AI-Driven Sensing & Communication: Integrated sensing and communication technologies powered by AI.

    ◆ Causal Inference for Global Health: AI-based causal analysis connecting climate change with global health outcomes. 

    ◆ Bio-Information Based AI: Advanced AI technologies for precise drug response prediction utilizing biological data. 

    ◆ Autonomous Control for Sustainability: AI-based autonomous control systems for sustainable energy infrastructure.  

[Research Content]

1.Establishment of an Information-Circulating Convergence AI Platform

● We are building an integrated AI platform that organically connects the technological flow of Sensing-Analysis-Prediction-Control, ensuring a seamless cycle of information.

2.Integration of Heterogeneous Data and Interactive System Design

● Our research involves the integrated processing of heterogeneous data collected from physical, social, and biological environments, alongside the design of systems capable of real-time interaction.

3.Design of Interconnected Architecture for Sub-Projects

● To ensure technical synergy between sub-projects, we design:

    ◆ Common Data Interfaces: For standardized data exchange. 

    ◆ Modular Functional Structures: For flexible system scaling and integration. 

    ◆ Mutual Feedback Paths: To enable continuous system optimization and learning across different domains.