About
About us
The Big Data Systems for AI Lab (BDSLAB) at the University of Houston develops interoperable, scalable and parallel algorithms for machine learning and AI, with a current focus on neural networks. We study how to train, optimize, monitor and explain neural networks on large and constantly changing data sets, applying techniques from database systems: relational storage, query processing, sparse matrix representations and I/O-efficient algorithms that balance CPU, memory and accelerators such as GPUs.
Our approach is distinctive: rather than relying on existing libraries, we build our own libraries, tools and systems, aiming to understand how analytic algorithms work end to end — from reading a data set off secondary storage to processing it in main memory on modern CPUs with a small RAM footprint. We work mainly in Python, R and C++, where data sources are large, diverse files that need not come from a relational database or data lake. Earlier work in the group centered on parallel DBMSs, Hadoop "Big Data" systems, graph analytics and machine learning models computed inside a DBMS.
Focus
Research topics
- Monitoring and inspecting neural networks with SQL queries over a relational database.
- I/O-efficient sparse matrix algorithms (multiplication, addition, updates) using database-style coordinate storage.
- Optimizing and evolving neural networks on large, frequently changing data sets.
- Database-inspired algorithms that balance CPU, I/O and accelerators (GPUs) for machine learning.
- Power- and energy-efficient AI and AI-optimized databases.
- LLM-guided data engineering: dataset harmonization, linking, and data quality with privacy constraints.
- Comparing classical models (association rules) with deep neural networks on medical data.
- Scalable machine learning with data summarization in Python, R and C++.
Director