About
I am a PhD candidate in Computer Science at the University of Auckland. My current research focuses on relational databases and anomaly detection from system logs.
Schema design decides what it costs to keep data correct, and classical normalization leaves that cost largely unquantified: it tells you to reach Third Normal Form, or Boyce-Codd where you can, but not how to choose among the many schemata that qualify. My work makes the choice measurable. I parameterize normal forms by the numbers of minimal keys and functional dependencies they exhibit, which exposes a concrete trade-off between how many ways an entity can be addressed and how much work each update costs, and I build covers that push as much of a constraint set as possible onto the keys that database systems already enforce natively.
Alongside this I work on discovering constraints from data — settling the complexity of cross join discovery and giving the first algorithms for it — and on graph-based models for detecting anomalies in system logs. My work has appeared at SIGMOD and VLDB, and in TKDE and the VLDB Journal.
Research
-
Parameterized Normal Forms
Textbook design stops at "3NF, and BCNF where possible", which admits many schemata and offers no principled way to choose among them. I parameterize BCNF by its number of minimal keys — a count that turns out to trade access variety against update complexity — and 3NF by its numbers of minimal keys and functional dependencies, so the effort of maintaining integrity becomes something you can measure and optimize at design time.
-
Mixed Covers of Keys and Dependencies
Database systems enforce keys natively but leave functional dependencies to triggers or application code. A mixed cover pairs the minimal keys with a cover of the remaining non-key FDs, putting as much of a constraint set as possible on native support: orders of magnitude less integrity maintenance overhead, and faster queries, refreshes and inserts on TPC-H.
-
Discovering Constraints in Data
Which dependencies does a dataset actually satisfy? For cross joins I settled the complexity — NP-complete, and W[3]-complete in the arity — and gave the first algorithms that discover them, including approximate variants that trade discovery time against a target ratio.
-
Anomaly Detection in System Logs
Graph models capture how log events depend on one another but overlook the substructures that betray a fault. My recent work searches for representative substructures and distills them for fast inference, and extends detection to the file level and to anomalies never seen in training.
News
- A Graph-based File-level Anomaly Detection Framework for System Logs has been accepted to EAAI.
- Unseen Anomaly Detection from System Logs has been accepted to SIGMOD 2026.
- Mixed Covers is out in the VLDB Journal, extending our VLDB 2024 paper.
- Our 3NF synthesis paper has been accepted to SIGMOD 2025.
Publications 10
Authors in publication order. Also on DBLP and Google Scholar.
2026
-
A Graph-based File-level Anomaly Detection Framework for System Logs
-
Unseen Anomaly Detection from System Logs
2025
-
Synthesizing Third Normal Form Schemata that Minimize Integrity Maintenance and Update Overheads: Parameterizing 3NF by the Numbers of Minimal Keys and Functional Dependencies
-
Mixed Covers: Optimizing Updates and Queries Using Minimal Keys and Functional Dependencies
-
Substructure-aware Log Anomaly Detection
2024
-
Mixed Covers of Keys and Functional Dependencies for Maintaining the Integrity of Data under Updates
2023
-
Composite Object Normal Forms: Parameterizing Boyce-Codd Normal Form by the Number of Minimal Keys
2022
-
Discovery of Cross Joins
2019
-
Simple SQL Validation of Generalized Entity Integrity
2018
-
Establishing Connections in a Social Network — Radial Versus Medial Centrality Indices
Contact
If you are interested in exploring possible collaborations, please don't hesitate to get in touch.
zhuoxing.zhang [at] auckland.ac.nzSchool of Computer Science, University of Auckland · Auckland, New Zealand