BDAI Lab - Research

Research

Research Statement

Our research is dedicated to advancing the next generation of intelligent data and computing systems by holistically integrating innovations across six key domains: (i) Big Data / Database & Cloud Systems, (ii) Systems for AI, (iii) AI for Systems, (iv) AI Model & Architecture Innovation, (v) Hardware-Software Co-Design, and (vi) Quantum Data & AI.

We envision a future in which data, AI, and systems co-evolve through tightly integrated, cross-disciplinary innovation, enabling scalable and intelligent computing platforms for emerging scientific and industrial applications.

• Big Data / Database & Cloud Systems: We design high-performance, AI-centric data platforms that unify structured, semi-structured, and unstructured data processing across cloud and distributed environments. Our research spans next-generation vector and tensor-native engines, GPU-accelerated query processing, cloud-native data systems, lakehouse architectures, and multimodal analytics. Through innovations in indexing, query optimization, adaptive execution, and elastic resource management, we aim to build scalable and intelligent data infrastructure for modern AI-driven applications.

• Systems for AI: We develop system infrastructures optimized for AI workloads, including data-aware scheduling, distributed runtime support, and ML-accelerated compilers. Our goal is to bridge the gap between algorithm design and real-world deployment by ensuring AI systems are both performant and resource-efficient across diverse environments.

• AI for Systems: We harness machine learning to rethink how systems are built and optimized. This includes learned query optimizers, self-tuning configuration engines, predictive scheduling frameworks, etc. forming a new paradigm of self-managing systems driven by adaptive intelligence.

• AI Model & Architecture Innovation: We develop new architectural paradigms for large-scale AI models — including LLMs, diffusion models, and multimodal AI — with an emphasis on accuracy, efficiency, and deployment scalability. Our research includes sparse modeling, hybrid modularity, and systems-aware model design for high-throughput, cost-effective intelligence.

• Hardware-Software Co-Design: Recognizing the growing complexity of modern hardware, we design vertically integrated solutions that co-optimize algorithms, software stacks, and emerging hardware such as GPUs, NPUs, TPUs, DPUs, QPUs, and near-memory accelerators. Our research spans compilers, memory management, and runtime optimization tailored for data- and AI-intensive workloads.

• Quantum Data & AI: We explore how quantum computing can unlock new capabilities in data management and learning. This includes quantum machine learning algorithms, quantum-enhanced data analytics, and hybrid classical-quantum pipelines for solving intractable problems in optimization and representation learning.

Together, these research thrusts form an integrated strategy for building data-centric, AI-native, and system-intelligent platforms. Through deep interdisciplinary collaboration and fundamental innovation, we aim to shape the foundations of next-generation computing infrastructure that will power scientific discovery, industrial intelligence, and societal-scale digital transformation.

Research Areas

Big Data / Database & Cloud Systems

• AI-Centric Data System Design
• Vector Database System Design
• GPU-Accelerated Database System Design
• Cloud-Native Data System Architectures
• Distributed Data Processing and Cloud Data Platforms
• HTAP (OLTP/OLAP) Systems Design
• Query Optimization / Execution
• Tensor-Relational Hybrid Query Processing
• Data Lakehouse System Architecture
• Multi-Model and Multimodal Data Management
• Learned and Adaptive Indexing Techniques
• Storage-Aware Query Planning
• Privacy-Aware and Secure Data Systems
• Data-Centric AI Pipeline Integration (ML-in-DB / DB-in-ML)
• SQL Extensions for AI Workloads (e.g., SQL + ML inference)

Publications

Systems for AI

• Scalable Infrastructure for AI Model Training
• High-Performance Inference Systems
• Distributed and Heterogeneous Systems for Deep Learning
• System Software for AI Workload Scheduling and Resource Management
• AI Cluster Resource Scheduling and Job Orchestration (e.g., Kubernetes, Ray)
• Model Serving Systems and Low-Latency Inference Serving
• Model Weight Sharding and Storage Optimization
• Security and Privacy-Preserving AI Systems
• Data/Model/Pipeline Parallelism Infrastructure

Publications

AI for Systems

• Machine Learning-Based Query Optimization
• Learned Indexes and Data Structure
• AutoML for Database and System Configuration
• AI-Guided Scheduling and Job Placement
• Predictive Modeling for Resource Management
• Workload Forecasting and Autoscaling
• AI-Powered Storage Tiering and Caching
• Reinforcement Learning for System Parameter Tuning
• AI-Augmented System Profiling and Bottleneck Detection
• Data Placement and Partitioning via ML Techniques

Publications

AI Model & Architecture Innovation

• LLM Architecture Optimization (e.g., MoE, weight sharing, KV cache compression)
• Diffusion Model Acceleration for training and inference (e.g., distillation, scheduler design)
• Multimodal Foundation Models (text-image-video-audio integration)
• Domain-specialized Architecture (e.g., medical, legal, scientific LLMs)
• Continual and Lifelong Learning Model Design

Publications

Hardware-Software Co-Design

• Architectures for tight integration of memory, storage, and compute resources
• Co-optimization of query engines and machine learning pipelines with emerging hardware (e.g., GPUs, TPUs, NPUs, DPUs, CXL)
• In-memory and near-data processing for high-throughput analytics
• CXL-enabled memory disaggregation and composable infrastructure

Publications